silhouette based 3D object tracking

クラス
class	cv::rapid::Tracker
	Abstract base class for stateful silhouette trackers [詳解]
class	cv::rapid::Rapid
	wrapper around silhouette based 3D object tracking function for uniform access [詳解]
class	cv::rapid::OLSTracker
class	cv::rapid::GOSTracker

関数
CV_EXPORTS_W void	cv::rapid::drawCorrespondencies (InputOutputArray bundle, InputArray cols, InputArray colors=noArray())
CV_EXPORTS_W void	cv::rapid::drawSearchLines (InputOutputArray img, InputArray locations, const Scalar &color)
CV_EXPORTS_W void	cv::rapid::drawWireframe (InputOutputArray img, InputArray pts2d, InputArray tris, const Scalar &color, int type=LINE_8, bool cullBackface=false)
CV_EXPORTS_W void	cv::rapid::extractControlPoints (int num, int len, InputArray pts3d, InputArray rvec, InputArray tvec, InputArray K, const Size &imsize, InputArray tris, OutputArray ctl2d, OutputArray ctl3d)
CV_EXPORTS_W void	cv::rapid::extractLineBundle (int len, InputArray ctl2d, InputArray img, OutputArray bundle, OutputArray srcLocations)
CV_EXPORTS_W void	cv::rapid::findCorrespondencies (InputArray bundle, OutputArray cols, OutputArray response=noArray())
CV_EXPORTS_W void	cv::rapid::convertCorrespondencies (InputArray cols, InputArray srcLocations, OutputArray pts2d, InputOutputArray pts3d=noArray(), InputArray mask=noArray())
CV_EXPORTS_W float	cv::rapid::rapid (InputArray img, int num, int len, InputArray pts3d, InputArray tris, InputArray K, InputOutputArray rvec, InputOutputArray tvec, CV_OUT double *rmsd=0)

詳解

implements "RAPID-a video rate object tracker" [harris1990rapid] with the dynamic control point extraction of [drummond2002real]

関数詳解

convertCorrespondencies()

CV_EXPORTS_W void cv::rapid::convertCorrespondencies	(	InputArray	cols,
		InputArray	srcLocations,
		OutputArray	pts2d,
		InputOutputArray	pts3d = noArray(),
		InputArray	mask = noArray()
	)

Collect corresponding 2d and 3d points based on correspondencies and mask

引数

cols	correspondence-position per line in line-bundle-space
srcLocations	the source image location
pts2d	2d points
pts3d	3d points
mask	mask containing non-zero values for the elements to be retained

drawCorrespondencies()

CV_EXPORTS_W void cv::rapid::drawCorrespondencies	(	InputOutputArray	bundle,
		InputArray	cols,
		InputArray	colors = noArray()
	)

Debug draw markers of matched correspondences onto a lineBundle

引数

bundle	the lineBundle
cols	column coordinates in the line bundle
colors	colors for the markers. Defaults to white.

drawSearchLines()

CV_EXPORTS_W void cv::rapid::drawSearchLines	(	InputOutputArray	img,
		InputArray	locations,
		const Scalar &	color
	)

Debug draw search lines onto an image

引数

img	the output image
locations	the source locations of a line bundle
color	the line color

drawWireframe()

CV_EXPORTS_W void cv::rapid::drawWireframe	(	InputOutputArray	img,
		InputArray	pts2d,
		InputArray	tris,
		const Scalar &	color,
		int	type = LINE_8,
		bool	cullBackface = false
	)

Draw a wireframe of a triangle mesh

引数

img	the output image
pts2d	the 2d points obtained by projectPoints
tris	triangle face connectivity
color	line color
type	line type. See LineTypes.
cullBackface	enable back-face culling based on CCW order

extractControlPoints()

CV_EXPORTS_W void cv::rapid::extractControlPoints	(	int	num,
		int	len,
		InputArray	pts3d,
		InputArray	rvec,
		InputArray	tvec,
		InputArray	K,
		const Size &	imsize,
		InputArray	tris,
		OutputArray	ctl2d,
		OutputArray	ctl3d
	)

Extract control points from the projected silhouette of a mesh

see [drummond2002real] Sec 2.1, Step b

引数

num	number of control points
len	search radius (used to restrict the ROI)
pts3d	the 3D points of the mesh
rvec	rotation between mesh and camera
tvec	translation between mesh and camera
K	camera intrinsic
imsize	size of the video frame
tris	triangle face connectivity
ctl2d	the 2D locations of the control points
ctl3d	matching 3D points of the mesh

extractLineBundle()

CV_EXPORTS_W void cv::rapid::extractLineBundle	(	int	len,
		InputArray	ctl2d,
		InputArray	img,
		OutputArray	bundle,
		OutputArray	srcLocations
	)

Extract the line bundle from an image

引数

len	the search radius. The bundle will have 2*len + 1 columns.
ctl2d	the search lines will be centered at this points and orthogonal to the contour defined by them. The bundle will have as many rows.
img	the image to read the pixel intensities values from
bundle	line bundle image with size ctl2d.rows() x (2 * len + 1) and the same type as img
srcLocations	the source pixel locations of bundle in img as CV_16SC2

findCorrespondencies()

CV_EXPORTS_W void cv::rapid::findCorrespondencies	(	InputArray	bundle,
		OutputArray	cols,
		OutputArray	response = noArray()
	)

Find corresponding image locations by searching for a maximal sobel edge along the search line (a single row in the bundle)

引数

bundle	the line bundle
cols	correspondence-position per line in line-bundle-space
response	the sobel response for the selected point

rapid()

CV_EXPORTS_W float cv::rapid::rapid	(	InputArray	img,
		int	num,
		int	len,
		InputArray	pts3d,
		InputArray	tris,
		InputArray	K,
		InputOutputArray	rvec,
		InputOutputArray	tvec,
		CV_OUT double *	rmsd = 0
	)

High level function to execute a single rapid [harris1990rapid] iteration

1. extractControlPoints
2. extractLineBundle
3. findCorrespondencies
4. convertCorrespondencies
5. solvePnPRefineLM

引数

img	the video frame
num	number of search lines
len	search line radius
pts3d	the 3D points of the mesh
tris	triangle face connectivity
K	camera matrix
rvec	rotation between mesh and camera. Input values are used as an initial solution.
tvec	translation between mesh and camera. Input values are used as an initial solution.
rmsd	the 2d reprojection difference

戻り値

ratio of search lines that could be extracted and matched