2011-11-07 12:55:18 +00:00
|
|
|
/*
|
|
|
|
* ***** BEGIN GPL LICENSE BLOCK *****
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation; either version 2
|
|
|
|
* of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program; if not, write to the Free Software Foundation,
|
|
|
|
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
|
|
*
|
|
|
|
* The Original Code is Copyright (C) 2011 Blender Foundation.
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Contributor(s): Blender Foundation,
|
|
|
|
* Sergey Sharybin
|
|
|
|
*
|
|
|
|
* ***** END GPL LICENSE BLOCK *****
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* define this to generate PNG images with content of search areas
|
|
|
|
tracking between which failed */
|
|
|
|
#undef DUMP_FAILURE
|
|
|
|
|
|
|
|
#include "libmv-capi.h"
|
|
|
|
|
|
|
|
#include "glog/logging.h"
|
2011-11-28 13:49:42 +00:00
|
|
|
#include "libmv/logging/logging.h"
|
|
|
|
|
2011-11-07 12:55:18 +00:00
|
|
|
#include "Math/v3d_optimization.h"
|
|
|
|
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
#include "libmv/tracking/esm_region_tracker.h"
|
2011-11-07 12:55:18 +00:00
|
|
|
#include "libmv/tracking/klt_region_tracker.h"
|
|
|
|
#include "libmv/tracking/trklt_region_tracker.h"
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
#include "libmv/tracking/lmicklt_region_tracker.h"
|
2011-11-07 12:55:18 +00:00
|
|
|
#include "libmv/tracking/pyramid_region_tracker.h"
|
|
|
|
|
|
|
|
#include "libmv/tracking/sad.h"
|
|
|
|
|
2011-11-28 13:49:42 +00:00
|
|
|
#include "libmv/simple_pipeline/callbacks.h"
|
2011-11-07 12:55:18 +00:00
|
|
|
#include "libmv/simple_pipeline/tracks.h"
|
|
|
|
#include "libmv/simple_pipeline/initialize_reconstruction.h"
|
|
|
|
#include "libmv/simple_pipeline/bundle.h"
|
|
|
|
#include "libmv/simple_pipeline/detect.h"
|
|
|
|
#include "libmv/simple_pipeline/pipeline.h"
|
|
|
|
#include "libmv/simple_pipeline/camera_intrinsics.h"
|
|
|
|
|
|
|
|
#include <stdlib.h>
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
#include <assert.h>
|
2011-11-07 12:55:18 +00:00
|
|
|
|
|
|
|
#ifdef DUMP_FAILURE
|
|
|
|
# include <png.h>
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef _MSC_VER
|
|
|
|
# define snprintf _snprintf
|
|
|
|
#endif
|
|
|
|
|
|
|
|
typedef struct libmv_Reconstruction {
|
|
|
|
libmv::EuclideanReconstruction reconstruction;
|
|
|
|
|
|
|
|
/* used for per-track average error calculation after reconstruction */
|
|
|
|
libmv::Tracks tracks;
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
double error;
|
|
|
|
} libmv_Reconstruction;
|
|
|
|
|
|
|
|
typedef struct libmv_Features {
|
|
|
|
int count, margin;
|
|
|
|
libmv::Feature *features;
|
|
|
|
} libmv_Features;
|
|
|
|
|
|
|
|
/* ************ Logging ************ */
|
|
|
|
|
|
|
|
void libmv_initLogging(const char *argv0)
|
|
|
|
{
|
|
|
|
google::InitGoogleLogging(argv0);
|
|
|
|
google::SetCommandLineOption("logtostderr", "1");
|
|
|
|
google::SetCommandLineOption("v", "0");
|
|
|
|
google::SetCommandLineOption("stderrthreshold", "7");
|
|
|
|
google::SetCommandLineOption("minloglevel", "7");
|
|
|
|
V3D::optimizerVerbosenessLevel = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_startDebugLogging(void)
|
|
|
|
{
|
|
|
|
google::SetCommandLineOption("logtostderr", "1");
|
|
|
|
google::SetCommandLineOption("v", "0");
|
|
|
|
google::SetCommandLineOption("stderrthreshold", "1");
|
|
|
|
google::SetCommandLineOption("minloglevel", "0");
|
|
|
|
V3D::optimizerVerbosenessLevel = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_setLoggingVerbosity(int verbosity)
|
|
|
|
{
|
|
|
|
char val[10];
|
|
|
|
snprintf(val, sizeof(val), "%d", verbosity);
|
|
|
|
|
|
|
|
google::SetCommandLineOption("v", val);
|
|
|
|
V3D::optimizerVerbosenessLevel = verbosity;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ RegionTracker ************ */
|
|
|
|
|
2011-11-14 06:41:32 +00:00
|
|
|
libmv_RegionTracker *libmv_regionTrackerNew(int max_iterations, int pyramid_level, int half_window_size)
|
2011-11-07 12:55:18 +00:00
|
|
|
{
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
libmv::EsmRegionTracker *klt_region_tracker = new libmv::EsmRegionTracker;
|
2011-11-07 12:55:18 +00:00
|
|
|
|
2011-11-14 06:41:32 +00:00
|
|
|
klt_region_tracker->half_window_size = half_window_size;
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
klt_region_tracker->max_iterations = max_iterations;
|
|
|
|
klt_region_tracker->min_determinant = 1e-4;
|
2011-11-07 12:55:18 +00:00
|
|
|
|
|
|
|
libmv::PyramidRegionTracker *region_tracker =
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
new libmv::PyramidRegionTracker(klt_region_tracker, pyramid_level);
|
2011-11-07 12:55:18 +00:00
|
|
|
|
2011-11-14 06:41:32 +00:00
|
|
|
return (libmv_RegionTracker *)region_tracker;
|
2011-11-07 12:55:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void floatBufToImage(const float *buf, int width, int height, libmv::FloatImage *image)
|
|
|
|
{
|
|
|
|
int x, y, a = 0;
|
|
|
|
|
|
|
|
image->resize(height, width);
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
(*image)(y, x, 0) = buf[a++];
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef DUMP_FAILURE
|
|
|
|
void savePNGImage(png_bytep *row_pointers, int width, int height, int depth, int color_type, char *file_name)
|
|
|
|
{
|
|
|
|
png_infop info_ptr;
|
|
|
|
png_structp png_ptr;
|
|
|
|
FILE *fp = fopen(file_name, "wb");
|
|
|
|
|
|
|
|
if (!fp)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Initialize stuff */
|
|
|
|
png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, NULL, NULL, NULL);
|
|
|
|
info_ptr = png_create_info_struct(png_ptr);
|
|
|
|
|
|
|
|
if (setjmp(png_jmpbuf(png_ptr))) {
|
|
|
|
fclose(fp);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
png_init_io(png_ptr, fp);
|
|
|
|
|
|
|
|
/* write header */
|
|
|
|
if (setjmp(png_jmpbuf(png_ptr))) {
|
|
|
|
fclose(fp);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
png_set_IHDR(png_ptr, info_ptr, width, height,
|
|
|
|
depth, color_type, PNG_INTERLACE_NONE,
|
|
|
|
PNG_COMPRESSION_TYPE_BASE, PNG_FILTER_TYPE_BASE);
|
|
|
|
|
|
|
|
png_write_info(png_ptr, info_ptr);
|
|
|
|
|
|
|
|
/* write bytes */
|
|
|
|
if (setjmp(png_jmpbuf(png_ptr))) {
|
|
|
|
fclose(fp);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
png_write_image(png_ptr, row_pointers);
|
|
|
|
|
|
|
|
/* end write */
|
|
|
|
if (setjmp(png_jmpbuf(png_ptr))) {
|
|
|
|
fclose(fp);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
png_write_end(png_ptr, NULL);
|
|
|
|
|
|
|
|
fclose(fp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void saveImage(char *prefix, libmv::FloatImage image, int x0, int y0)
|
|
|
|
{
|
|
|
|
int x, y;
|
|
|
|
png_bytep *row_pointers;
|
|
|
|
|
|
|
|
row_pointers= (png_bytep*)malloc(sizeof(png_bytep)*image.Height());
|
|
|
|
|
|
|
|
for (y = 0; y < image.Height(); y++) {
|
|
|
|
row_pointers[y]= (png_bytep)malloc(sizeof(png_byte)*4*image.Width());
|
|
|
|
|
|
|
|
for (x = 0; x < image.Width(); x++) {
|
|
|
|
if (x0 == x && y0 == y) {
|
|
|
|
row_pointers[y][x*4+0]= 255;
|
|
|
|
row_pointers[y][x*4+1]= 0;
|
|
|
|
row_pointers[y][x*4+2]= 0;
|
|
|
|
row_pointers[y][x*4+3]= 255;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
float pixel = image(y, x, 0);
|
|
|
|
row_pointers[y][x*4+0]= pixel*255;
|
|
|
|
row_pointers[y][x*4+1]= pixel*255;
|
|
|
|
row_pointers[y][x*4+2]= pixel*255;
|
|
|
|
row_pointers[y][x*4+3]= 255;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
{
|
|
|
|
static int a= 0;
|
|
|
|
char buf[128];
|
|
|
|
snprintf(buf, sizeof(buf), "%s_%02d.png", prefix, ++a);
|
|
|
|
savePNGImage(row_pointers, image.Width(), image.Height(), 8, PNG_COLOR_TYPE_RGBA, buf);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (y = 0; y < image.Height(); y++) {
|
|
|
|
free(row_pointers[y]);
|
|
|
|
}
|
|
|
|
free(row_pointers);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void saveBytesImage(char *prefix, unsigned char *data, int width, int height)
|
|
|
|
{
|
|
|
|
int x, y;
|
|
|
|
png_bytep *row_pointers;
|
|
|
|
|
|
|
|
row_pointers= (png_bytep*)malloc(sizeof(png_bytep)*height);
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
row_pointers[y]= (png_bytep)malloc(sizeof(png_byte)*4*width);
|
|
|
|
|
|
|
|
for (x = 0; x < width; x++) {
|
|
|
|
char pixel = data[width*y+x];
|
|
|
|
row_pointers[y][x*4+0]= pixel;
|
|
|
|
row_pointers[y][x*4+1]= pixel;
|
|
|
|
row_pointers[y][x*4+2]= pixel;
|
|
|
|
row_pointers[y][x*4+3]= 255;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
{
|
|
|
|
static int a= 0;
|
|
|
|
char buf[128];
|
|
|
|
snprintf(buf, sizeof(buf), "%s_%02d.png", prefix, ++a);
|
|
|
|
savePNGImage(row_pointers, width, height, 8, PNG_COLOR_TYPE_RGBA, buf);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (y = 0; y < height; y++) {
|
|
|
|
free(row_pointers[y]);
|
|
|
|
}
|
|
|
|
free(row_pointers);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
int libmv_regionTrackerTrack(libmv_RegionTracker *libmv_tracker, const float *ima1, const float *ima2,
|
2011-11-14 06:41:32 +00:00
|
|
|
int width, int height, double x1, double y1, double *x2, double *y2)
|
2011-11-07 12:55:18 +00:00
|
|
|
{
|
2011-11-14 06:41:32 +00:00
|
|
|
libmv::RegionTracker *region_tracker = (libmv::RegionTracker *)libmv_tracker;
|
2011-11-07 12:55:18 +00:00
|
|
|
libmv::FloatImage old_patch, new_patch;
|
|
|
|
|
|
|
|
floatBufToImage(ima1, width, height, &old_patch);
|
|
|
|
floatBufToImage(ima2, width, height, &new_patch);
|
|
|
|
|
|
|
|
#ifndef DUMP_FAILURE
|
|
|
|
return region_tracker->Track(old_patch, new_patch, x1, y1, x2, y2);
|
|
|
|
#else
|
|
|
|
{
|
|
|
|
double sx2 = *x2, sy2 = *y2;
|
|
|
|
int result = region_tracker->Track(old_patch, new_patch, x1, y1, x2, y2);
|
|
|
|
|
|
|
|
if (!result) {
|
|
|
|
saveImage("old_patch", old_patch, x1, y1);
|
|
|
|
saveImage("new_patch", new_patch, sx2, sy2);
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_regionTrackerDestroy(libmv_RegionTracker *libmv_tracker)
|
|
|
|
{
|
2011-11-14 06:41:32 +00:00
|
|
|
libmv::RegionTracker *region_tracker= (libmv::RegionTracker *)libmv_tracker;
|
|
|
|
|
|
|
|
delete region_tracker;
|
2011-11-07 12:55:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ Tracks ************ */
|
|
|
|
|
|
|
|
void libmv_SADSamplePattern(unsigned char *image, int stride,
|
|
|
|
float warp[3][2], unsigned char *pattern)
|
|
|
|
{
|
|
|
|
libmv::mat32 mat32;
|
|
|
|
|
|
|
|
memcpy(mat32.data, warp, sizeof(float)*3*2);
|
|
|
|
|
|
|
|
libmv::SamplePattern(image, stride, mat32, pattern, 16);
|
|
|
|
}
|
|
|
|
|
|
|
|
float libmv_SADTrackerTrack(unsigned char *pattern, unsigned char *warped, unsigned char *image, int stride,
|
|
|
|
int width, int height, float warp[3][2])
|
|
|
|
{
|
|
|
|
float result;
|
|
|
|
libmv::mat32 mat32;
|
|
|
|
|
|
|
|
memcpy(mat32.data, warp, sizeof(float)*3*2);
|
|
|
|
|
|
|
|
result = libmv::Track(pattern, warped, 16, image, stride, width, height, &mat32, 16, 16);
|
|
|
|
|
|
|
|
memcpy(warp, mat32.data, sizeof(float)*3*2);
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ Tracks ************ */
|
|
|
|
|
|
|
|
libmv_Tracks *libmv_tracksNew(void)
|
|
|
|
{
|
|
|
|
libmv::Tracks *libmv_tracks = new libmv::Tracks();
|
|
|
|
|
|
|
|
return (libmv_Tracks *)libmv_tracks;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_tracksInsert(struct libmv_Tracks *libmv_tracks, int image, int track, double x, double y)
|
|
|
|
{
|
|
|
|
((libmv::Tracks*)libmv_tracks)->Insert(image, track, x, y);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_tracksDestroy(libmv_Tracks *libmv_tracks)
|
|
|
|
{
|
|
|
|
delete (libmv::Tracks*)libmv_tracks;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ Reconstruction solver ************ */
|
|
|
|
|
2011-11-28 13:49:42 +00:00
|
|
|
class ReconstructUpdateCallback : public libmv::ProgressUpdateCallback {
|
|
|
|
public:
|
|
|
|
ReconstructUpdateCallback(reconstruct_progress_update_cb progress_update_callback,
|
|
|
|
void *callback_customdata)
|
|
|
|
{
|
|
|
|
progress_update_callback_ = progress_update_callback;
|
|
|
|
callback_customdata_ = callback_customdata;
|
|
|
|
}
|
|
|
|
|
|
|
|
void invoke(double progress, const char *message)
|
|
|
|
{
|
|
|
|
if(progress_update_callback_) {
|
|
|
|
progress_update_callback_(callback_customdata_, progress, message);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
protected:
|
|
|
|
reconstruct_progress_update_cb progress_update_callback_;
|
|
|
|
void *callback_customdata_;
|
|
|
|
};
|
|
|
|
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
int libmv_refineParametersAreValid(int parameters) {
|
|
|
|
return (parameters == (LIBMV_REFINE_FOCAL_LENGTH)) ||
|
|
|
|
(parameters == (LIBMV_REFINE_FOCAL_LENGTH |
|
|
|
|
LIBMV_REFINE_PRINCIPAL_POINT)) ||
|
|
|
|
(parameters == (LIBMV_REFINE_FOCAL_LENGTH |
|
|
|
|
LIBMV_REFINE_PRINCIPAL_POINT |
|
|
|
|
LIBMV_REFINE_RADIAL_DISTORTION_K1 |
|
|
|
|
LIBMV_REFINE_RADIAL_DISTORTION_K2)) ||
|
|
|
|
(parameters == (LIBMV_REFINE_FOCAL_LENGTH |
|
|
|
|
LIBMV_REFINE_RADIAL_DISTORTION_K1 |
|
|
|
|
LIBMV_REFINE_RADIAL_DISTORTION_K2)) ||
|
|
|
|
(parameters == (LIBMV_REFINE_FOCAL_LENGTH |
|
|
|
|
LIBMV_REFINE_RADIAL_DISTORTION_K1));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2011-11-07 12:55:18 +00:00
|
|
|
libmv_Reconstruction *libmv_solveReconstruction(libmv_Tracks *tracks, int keyframe1, int keyframe2,
|
2011-11-28 13:49:42 +00:00
|
|
|
int refine_intrinsics, double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
reconstruct_progress_update_cb progress_update_callback, void *callback_customdata)
|
2011-11-07 12:55:18 +00:00
|
|
|
{
|
|
|
|
/* Invert the camera intrinsics. */
|
|
|
|
libmv::vector<libmv::Marker> markers = ((libmv::Tracks*)tracks)->AllMarkers();
|
|
|
|
libmv_Reconstruction *libmv_reconstruction = new libmv_Reconstruction();
|
|
|
|
libmv::EuclideanReconstruction *reconstruction = &libmv_reconstruction->reconstruction;
|
|
|
|
libmv::CameraIntrinsics *intrinsics = &libmv_reconstruction->intrinsics;
|
|
|
|
|
2011-11-28 13:49:42 +00:00
|
|
|
ReconstructUpdateCallback update_callback =
|
|
|
|
ReconstructUpdateCallback(progress_update_callback, callback_customdata);
|
|
|
|
|
2011-11-07 12:55:18 +00:00
|
|
|
intrinsics->SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics->SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics->SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
2011-11-16 10:00:02 +00:00
|
|
|
for (int i = 0; i < markers.size(); ++i) {
|
|
|
|
intrinsics->InvertIntrinsics(markers[i].x,
|
|
|
|
markers[i].y,
|
|
|
|
&(markers[i].x),
|
|
|
|
&(markers[i].y));
|
2011-11-07 12:55:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
libmv::Tracks normalized_tracks(markers);
|
|
|
|
|
2011-11-28 13:49:42 +00:00
|
|
|
LG << "frames to init from: " << keyframe1 << " " << keyframe2;
|
2011-11-07 12:55:18 +00:00
|
|
|
libmv::vector<libmv::Marker> keyframe_markers =
|
|
|
|
normalized_tracks.MarkersForTracksInBothImages(keyframe1, keyframe2);
|
2011-11-28 13:49:42 +00:00
|
|
|
LG << "number of markers for init: " << keyframe_markers.size();
|
|
|
|
|
|
|
|
update_callback.invoke(0, "Initial reconstruction");
|
2011-11-07 12:55:18 +00:00
|
|
|
|
|
|
|
libmv::EuclideanReconstructTwoFrames(keyframe_markers, reconstruction);
|
|
|
|
libmv::EuclideanBundle(normalized_tracks, reconstruction);
|
2011-11-28 13:49:42 +00:00
|
|
|
libmv::EuclideanCompleteReconstruction(normalized_tracks, reconstruction, &update_callback);
|
2011-11-07 12:55:18 +00:00
|
|
|
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
if (refine_intrinsics) {
|
|
|
|
/* only a few combinations are supported but trust the caller */
|
|
|
|
int libmv_refine_flags = 0;
|
|
|
|
if (refine_intrinsics & LIBMV_REFINE_FOCAL_LENGTH) {
|
|
|
|
libmv_refine_flags |= libmv::BUNDLE_FOCAL_LENGTH;
|
|
|
|
}
|
|
|
|
if (refine_intrinsics & LIBMV_REFINE_PRINCIPAL_POINT) {
|
|
|
|
libmv_refine_flags |= libmv::BUNDLE_PRINCIPAL_POINT;
|
|
|
|
}
|
|
|
|
if (refine_intrinsics & LIBMV_REFINE_RADIAL_DISTORTION_K1) {
|
|
|
|
libmv_refine_flags |= libmv::BUNDLE_RADIAL_K1;
|
|
|
|
}
|
|
|
|
if (refine_intrinsics & LIBMV_REFINE_RADIAL_DISTORTION_K2) {
|
|
|
|
libmv_refine_flags |= libmv::BUNDLE_RADIAL_K2;
|
|
|
|
}
|
2011-11-28 13:49:42 +00:00
|
|
|
|
|
|
|
progress_update_callback(callback_customdata, 1.0, "Refining solution");
|
|
|
|
libmv::EuclideanBundleCommonIntrinsics(*(libmv::Tracks *)tracks, libmv_refine_flags,
|
|
|
|
reconstruction, intrinsics);
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
}
|
|
|
|
|
2011-11-28 13:49:42 +00:00
|
|
|
progress_update_callback(callback_customdata, 1.0, "Finishing solution");
|
2011-11-07 12:55:18 +00:00
|
|
|
libmv_reconstruction->tracks = *(libmv::Tracks *)tracks;
|
|
|
|
libmv_reconstruction->error = libmv::EuclideanReprojectionError(*(libmv::Tracks *)tracks, *reconstruction, *intrinsics);
|
|
|
|
|
|
|
|
return (libmv_Reconstruction *)libmv_reconstruction;
|
|
|
|
}
|
|
|
|
|
|
|
|
int libmv_reporojectionPointForTrack(libmv_Reconstruction *libmv_reconstruction, int track, double pos[3])
|
|
|
|
{
|
|
|
|
libmv::EuclideanReconstruction *reconstruction = &libmv_reconstruction->reconstruction;
|
|
|
|
libmv::EuclideanPoint *point = reconstruction->PointForTrack(track);
|
|
|
|
|
|
|
|
if(point) {
|
|
|
|
pos[0] = point->X[0];
|
|
|
|
pos[1] = point->X[2];
|
|
|
|
pos[2] = point->X[1];
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static libmv::Marker ProjectMarker(const libmv::EuclideanPoint &point, const libmv::EuclideanCamera &camera,
|
|
|
|
const libmv::CameraIntrinsics &intrinsics) {
|
|
|
|
libmv::Vec3 projected = camera.R * point.X + camera.t;
|
|
|
|
projected /= projected(2);
|
|
|
|
|
|
|
|
libmv::Marker reprojected_marker;
|
|
|
|
intrinsics.ApplyIntrinsics(projected(0), projected(1), &reprojected_marker.x, &reprojected_marker.y);
|
|
|
|
|
|
|
|
reprojected_marker.image = camera.image;
|
|
|
|
reprojected_marker.track = point.track;
|
|
|
|
|
|
|
|
return reprojected_marker;
|
|
|
|
}
|
|
|
|
|
|
|
|
double libmv_reporojectionErrorForTrack(libmv_Reconstruction *libmv_reconstruction, int track)
|
|
|
|
{
|
|
|
|
libmv::EuclideanReconstruction *reconstruction = &libmv_reconstruction->reconstruction;
|
|
|
|
libmv::CameraIntrinsics *intrinsics = &libmv_reconstruction->intrinsics;
|
|
|
|
libmv::vector<libmv::Marker> markers = libmv_reconstruction->tracks.MarkersForTrack(track);
|
|
|
|
|
|
|
|
int num_reprojected = 0;
|
|
|
|
double total_error = 0.0;
|
|
|
|
|
|
|
|
for (int i = 0; i < markers.size(); ++i) {
|
|
|
|
const libmv::EuclideanCamera *camera = reconstruction->CameraForImage(markers[i].image);
|
|
|
|
const libmv::EuclideanPoint *point = reconstruction->PointForTrack(markers[i].track);
|
|
|
|
|
|
|
|
if (!camera || !point) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
num_reprojected++;
|
|
|
|
|
|
|
|
libmv::Marker reprojected_marker = ProjectMarker(*point, *camera, *intrinsics);
|
|
|
|
double ex = reprojected_marker.x - markers[i].x;
|
|
|
|
double ey = reprojected_marker.y - markers[i].y;
|
|
|
|
|
|
|
|
total_error += sqrt(ex*ex + ey*ey);
|
|
|
|
}
|
|
|
|
|
|
|
|
return total_error / num_reprojected;
|
|
|
|
}
|
|
|
|
|
|
|
|
double libmv_reporojectionErrorForImage(libmv_Reconstruction *libmv_reconstruction, int image)
|
|
|
|
{
|
|
|
|
libmv::EuclideanReconstruction *reconstruction = &libmv_reconstruction->reconstruction;
|
|
|
|
libmv::CameraIntrinsics *intrinsics = &libmv_reconstruction->intrinsics;
|
|
|
|
libmv::vector<libmv::Marker> markers = libmv_reconstruction->tracks.MarkersInImage(image);
|
|
|
|
const libmv::EuclideanCamera *camera = reconstruction->CameraForImage(image);
|
|
|
|
int num_reprojected = 0;
|
|
|
|
double total_error = 0.0;
|
|
|
|
|
|
|
|
if (!camera)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
for (int i = 0; i < markers.size(); ++i) {
|
|
|
|
const libmv::EuclideanPoint *point = reconstruction->PointForTrack(markers[i].track);
|
|
|
|
|
|
|
|
if (!point) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
num_reprojected++;
|
|
|
|
|
|
|
|
libmv::Marker reprojected_marker = ProjectMarker(*point, *camera, *intrinsics);
|
|
|
|
double ex = reprojected_marker.x - markers[i].x;
|
|
|
|
double ey = reprojected_marker.y - markers[i].y;
|
|
|
|
|
|
|
|
total_error += sqrt(ex*ex + ey*ey);
|
|
|
|
}
|
|
|
|
|
|
|
|
return total_error / num_reprojected;
|
|
|
|
}
|
|
|
|
|
|
|
|
int libmv_reporojectionCameraForImage(libmv_Reconstruction *libmv_reconstruction, int image, double mat[4][4])
|
|
|
|
{
|
|
|
|
libmv::EuclideanReconstruction *reconstruction = &libmv_reconstruction->reconstruction;
|
|
|
|
libmv::EuclideanCamera *camera = reconstruction->CameraForImage(image);
|
|
|
|
|
|
|
|
if(camera) {
|
|
|
|
for (int j = 0; j < 3; ++j) {
|
|
|
|
for (int k = 0; k < 3; ++k) {
|
|
|
|
int l = k;
|
|
|
|
|
|
|
|
if (k == 1) l = 2;
|
|
|
|
else if (k == 2) l = 1;
|
|
|
|
|
|
|
|
if (j == 2) mat[j][l] = -camera->R(j,k);
|
|
|
|
else mat[j][l] = camera->R(j,k);
|
|
|
|
}
|
|
|
|
mat[j][3]= 0.0;
|
|
|
|
}
|
|
|
|
|
|
|
|
libmv::Vec3 optical_center = -camera->R.transpose() * camera->t;
|
|
|
|
|
|
|
|
mat[3][0] = optical_center(0);
|
|
|
|
mat[3][1] = optical_center(2);
|
|
|
|
mat[3][2] = optical_center(1);
|
|
|
|
|
|
|
|
mat[3][3]= 1.0;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
double libmv_reprojectionError(libmv_Reconstruction *libmv_reconstruction)
|
|
|
|
{
|
|
|
|
return libmv_reconstruction->error;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_destroyReconstruction(libmv_Reconstruction *libmv_reconstruction)
|
|
|
|
{
|
|
|
|
delete libmv_reconstruction;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ feature detector ************ */
|
|
|
|
|
|
|
|
struct libmv_Features *libmv_detectFeaturesFAST(unsigned char *data, int width, int height, int stride,
|
|
|
|
int margin, int min_trackness, int min_distance)
|
|
|
|
{
|
|
|
|
libmv::Feature *features = NULL;
|
|
|
|
std::vector<libmv::Feature> v;
|
|
|
|
libmv_Features *libmv_features = new libmv_Features();
|
|
|
|
int i= 0, count;
|
|
|
|
|
|
|
|
if(margin) {
|
|
|
|
data += margin*stride+margin;
|
|
|
|
width -= 2*margin;
|
|
|
|
height -= 2*margin;
|
|
|
|
}
|
|
|
|
|
|
|
|
v = libmv::DetectFAST(data, width, height, stride, min_trackness, min_distance);
|
|
|
|
|
|
|
|
count = v.size();
|
|
|
|
|
|
|
|
if(count) {
|
|
|
|
features= new libmv::Feature[count];
|
|
|
|
|
|
|
|
for(std::vector<libmv::Feature>::iterator it = v.begin(); it != v.end(); it++) {
|
|
|
|
features[i++]= *it;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
libmv_features->features = features;
|
|
|
|
libmv_features->count = count;
|
|
|
|
libmv_features->margin = margin;
|
|
|
|
|
|
|
|
return (libmv_Features *)libmv_features;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct libmv_Features *libmv_detectFeaturesMORAVEC(unsigned char *data, int width, int height, int stride,
|
|
|
|
int margin, int count, int min_distance)
|
|
|
|
{
|
|
|
|
libmv::Feature *features = NULL;
|
|
|
|
libmv_Features *libmv_features = new libmv_Features;
|
|
|
|
|
|
|
|
if(count) {
|
|
|
|
if(margin) {
|
|
|
|
data += margin*stride+margin;
|
|
|
|
width -= 2*margin;
|
|
|
|
height -= 2*margin;
|
|
|
|
}
|
|
|
|
|
|
|
|
features = new libmv::Feature[count];
|
|
|
|
libmv::DetectMORAVEC(data, stride, width, height, features, &count, min_distance, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
libmv_features->count = count;
|
|
|
|
libmv_features->margin = margin;
|
|
|
|
libmv_features->features = features;
|
|
|
|
|
|
|
|
return libmv_features;
|
|
|
|
}
|
|
|
|
|
|
|
|
int libmv_countFeatures(struct libmv_Features *libmv_features)
|
|
|
|
{
|
|
|
|
return libmv_features->count;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_getFeature(struct libmv_Features *libmv_features, int number, double *x, double *y, double *score, double *size)
|
|
|
|
{
|
|
|
|
libmv::Feature feature= libmv_features->features[number];
|
|
|
|
|
|
|
|
*x = feature.x + libmv_features->margin;
|
|
|
|
*y = feature.y + libmv_features->margin;
|
|
|
|
*score = feature.score;
|
|
|
|
*size = feature.size;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_destroyFeatures(struct libmv_Features *libmv_features)
|
|
|
|
{
|
|
|
|
if(libmv_features->features)
|
|
|
|
delete [] libmv_features->features;
|
|
|
|
|
|
|
|
delete libmv_features;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ camera intrinsics ************ */
|
|
|
|
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
struct libmv_CameraIntrinsics *libmv_ReconstructionExtractIntrinsics(struct libmv_Reconstruction *libmv_Reconstruction) {
|
|
|
|
return (struct libmv_CameraIntrinsics *)&libmv_Reconstruction->intrinsics;
|
|
|
|
}
|
|
|
|
|
2011-11-07 12:55:18 +00:00
|
|
|
struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsNew(double focal_length, double principal_x, double principal_y,
|
|
|
|
double k1, double k2, double k3, int width, int height)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics= new libmv::CameraIntrinsics();
|
|
|
|
|
|
|
|
intrinsics->SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics->SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics->SetRadialDistortion(k1, k2, k3);
|
|
|
|
intrinsics->SetImageSize(width, height);
|
|
|
|
|
|
|
|
return (struct libmv_CameraIntrinsics *) intrinsics;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct libmv_CameraIntrinsics *libmv_CameraIntrinsicsCopy(struct libmv_CameraIntrinsics *libmvIntrinsics)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *orig_intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
libmv::CameraIntrinsics *new_intrinsics= new libmv::CameraIntrinsics(*orig_intrinsics);
|
|
|
|
|
|
|
|
return (struct libmv_CameraIntrinsics *) new_intrinsics;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_CameraIntrinsicsDestroy(struct libmv_CameraIntrinsics *libmvIntrinsics)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
|
|
|
|
delete intrinsics;
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_CameraIntrinsicsUpdate(struct libmv_CameraIntrinsics *libmvIntrinsics, double focal_length,
|
|
|
|
double principal_x, double principal_y, double k1, double k2, double k3, int width, int height)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
|
|
|
|
if (intrinsics->focal_length() != focal_length)
|
|
|
|
intrinsics->SetFocalLength(focal_length, focal_length);
|
|
|
|
|
|
|
|
if (intrinsics->principal_point_x() != principal_x || intrinsics->principal_point_y() != principal_y)
|
|
|
|
intrinsics->SetFocalLength(focal_length, focal_length);
|
|
|
|
|
|
|
|
if (intrinsics->k1() != k1 || intrinsics->k2() != k2 || intrinsics->k3() != k3)
|
|
|
|
intrinsics->SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
if (intrinsics->image_width() != width || intrinsics->image_height() != height)
|
|
|
|
intrinsics->SetImageSize(width, height);
|
|
|
|
}
|
|
|
|
|
Assorted camera tracker improvements
- Add support for refining the camera's intrinsic parameters
during a solve. Currently, refining supports only the following
combinations of intrinsic parameters:
f
f, cx, cy
f, cx, cy, k1, k2
f, k1
f, k1, k2
This is not the same as autocalibration, since the user must
still make a reasonable initial guess about the focal length and
other parameters, whereas true autocalibration would eliminate
the need for the user specify intrinsic parameters at all.
However, the solver works well with only rough guesses for the
focal length, so perhaps full autocalibation is not that
important.
Adding support for the last two combinations, (f, k1) and (f,
k1, k2) required changes to the library libmv depends on for
bundle adjustment, SSBA. These changes should get ported
upstream not just to libmv but to SSBA as well.
- Improved the region of convergence for bundle adjustment by
increasing the number of Levenberg-Marquardt iterations from 50
to 500. This way, the solver is able to crawl out of the bad
local minima it gets stuck in when changing from, for example,
bundling k1 and k2 to just k1 and resetting k2 to 0.
- Add several new region tracker implementations. A region tracker
is a libmv concept, which refers to tracking a template image
pattern through frames. The impact to end users is that tracking
should "just work better". I am reserving a more detailed
writeup, and maybe a paper, for later.
- Other libmv tweaks, such as detecting that a tracker is headed
outside of the image bounds.
This includes several changes made directly to the libmv extern
code rather expecting to get those changes through normal libmv
channels, because I, the libmv BDFL, decided it was faster to work
on libmv directly in Blender, then later reverse-port the libmv
changes from Blender back into libmv trunk. The interesting part
is that I added a full Levenberg-Marquardt loop to the region
tracking code, which should lead to a more stable solutions. I
also added a hacky implementation of "Efficient Second-Order
Minimization" for tracking, which works nicely. A more detailed
quantitative evaluation will follow.
Original patch by Keir, cleaned a bit by myself.
2011-11-14 06:41:23 +00:00
|
|
|
void libmv_CameraIntrinsicsExtract(struct libmv_CameraIntrinsics *libmvIntrinsics, double *focal_length,
|
|
|
|
double *principal_x, double *principal_y, double *k1, double *k2, double *k3, int *width, int *height) {
|
|
|
|
libmv::CameraIntrinsics *intrinsics= (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
*focal_length = intrinsics->focal_length();
|
|
|
|
*principal_x = intrinsics->principal_point_x();
|
|
|
|
*principal_y = intrinsics->principal_point_y();
|
|
|
|
*k1 = intrinsics->k1();
|
|
|
|
*k2 = intrinsics->k2();
|
|
|
|
}
|
|
|
|
|
2011-11-07 12:55:18 +00:00
|
|
|
void libmv_CameraIntrinsicsUndistortByte(struct libmv_CameraIntrinsics *libmvIntrinsics,
|
|
|
|
unsigned char *src, unsigned char *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
|
|
|
|
intrinsics->Undistort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_CameraIntrinsicsUndistortFloat(struct libmv_CameraIntrinsics *libmvIntrinsics,
|
|
|
|
float *src, float *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
|
|
|
|
intrinsics->Undistort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_CameraIntrinsicsDistortByte(struct libmv_CameraIntrinsics *libmvIntrinsics,
|
|
|
|
unsigned char *src, unsigned char *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
intrinsics->Distort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_CameraIntrinsicsDistortFloat(struct libmv_CameraIntrinsics *libmvIntrinsics,
|
|
|
|
float *src, float *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics *intrinsics = (libmv::CameraIntrinsics *) libmvIntrinsics;
|
|
|
|
|
|
|
|
intrinsics->Distort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ distortion ************ */
|
|
|
|
|
|
|
|
void libmv_undistortByte(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
unsigned char *src, unsigned char *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
intrinsics.Undistort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_undistortFloat(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
float *src, float *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
intrinsics.Undistort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_distortByte(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
unsigned char *src, unsigned char *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
intrinsics.Distort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_distortFloat(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
float *src, float *dst, int width, int height, float overscan, int channels)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
intrinsics.Distort(src, dst, width, height, overscan, channels);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ************ utils ************ */
|
|
|
|
|
|
|
|
void libmv_applyCameraIntrinsics(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
double x, double y, double *x1, double *y1)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
if(focal_length) {
|
|
|
|
/* do a lens undistortion if focal length is non-zero only */
|
|
|
|
|
|
|
|
intrinsics.ApplyIntrinsics(x, y, x1, y1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void libmv_InvertIntrinsics(double focal_length, double principal_x, double principal_y, double k1, double k2, double k3,
|
|
|
|
double x, double y, double *x1, double *y1)
|
|
|
|
{
|
|
|
|
libmv::CameraIntrinsics intrinsics;
|
|
|
|
|
|
|
|
intrinsics.SetFocalLength(focal_length, focal_length);
|
|
|
|
intrinsics.SetPrincipalPoint(principal_x, principal_y);
|
|
|
|
intrinsics.SetRadialDistortion(k1, k2, k3);
|
|
|
|
|
|
|
|
if(focal_length) {
|
|
|
|
/* do a lens distortion if focal length is non-zero only */
|
|
|
|
|
|
|
|
intrinsics.InvertIntrinsics(x, y, x1, y1);
|
|
|
|
}
|
|
|
|
}
|