RGB-D and Volumetric Reconstruction

Detailed Description

Classes
class	cv::Odometry
class	cv::OdometryFrame
	An object that keeps per-frame data for Odometry algorithms from user-provided images to algorithm-specific precalculated data. When not empty, it contains a depth image, a mask of valid pixels and a set of pyramids generated from that data. A BGR/Gray image and normals are optional. OdometryFrame is made to be used together with Odometry class to reuse precalculated data between Rt data calculations. A correct way to do that is to call Odometry::prepareFrames() on prev and next frames and then pass them to Odometry::compute() method. More...
class	cv::OdometrySettings
class	cv::RgbdNormals
class	cv::Volume
class	cv::VolumeSettings

Enumerations
enum class	cv::OdometryAlgoType {   cv::OdometryAlgoType::COMMON = 0 ,   cv::OdometryAlgoType::FAST = 1 }
enum	cv::OdometryFramePyramidType {   cv::PYR_IMAGE = 0 ,   cv::PYR_DEPTH = 1 ,   cv::PYR_MASK = 2 ,   cv::PYR_CLOUD = 3 ,   cv::PYR_DIX = 4 ,   cv::PYR_DIY = 5 ,   cv::PYR_TEXMASK = 6 ,   cv::PYR_NORM = 7 ,   cv::PYR_NORMMASK = 8 ,   cv::N_PYRAMIDS }
enum class	cv::OdometryType {   cv::OdometryType::DEPTH = 0 ,   cv::OdometryType::RGB = 1 ,   cv::OdometryType::RGB_DEPTH = 2 }
enum	cv::RgbdPlaneMethod { cv::RGBD_PLANE_METHOD_DEFAULT }
enum class	cv::VolumeType {   cv::VolumeType::TSDF = 0 ,   cv::VolumeType::HashTSDF = 1 ,   cv::VolumeType::ColorTSDF = 2 }

Functions
void	cv::depthTo3d (InputArray depth, InputArray K, OutputArray points3d, InputArray mask=noArray())
void	cv::depthTo3dSparse (InputArray depth, InputArray in_K, InputArray in_points, OutputArray points3d)
void	cv::findPlanes (InputArray points3d, InputArray normals, OutputArray mask, OutputArray plane_coefficients, int block_size=40, int min_size=40 *40, double threshold=0.01, double sensor_error_a=0, double sensor_error_b=0, double sensor_error_c=0, RgbdPlaneMethod method=RGBD_PLANE_METHOD_DEFAULT)
void	cv::registerDepth (InputArray unregisteredCameraMatrix, InputArray registeredCameraMatrix, InputArray registeredDistCoeffs, InputArray Rt, InputArray unregisteredDepth, const Size &outputImagePlaneSize, OutputArray registeredDepth, bool depthDilation=false)
void	cv::rescaleDepth (InputArray in, int type, OutputArray out, double depth_factor=1000.0)
void	cv::warpFrame (InputArray depth, InputArray image, InputArray mask, InputArray Rt, InputArray cameraMatrix, OutputArray warpedDepth=noArray(), OutputArray warpedImage=noArray(), OutputArray warpedMask=noArray())

Enumeration Type Documentation

OdometryAlgoType

enum class cv::OdometryAlgoType

strong

#include <opencv2/ptcloud/odometry.hpp>

These constants are used to set the speed and accuracy of odometry

Parameters

COMMON	accurate but not so fast
FAST	less accurate but faster

Enumerator
COMMON
FAST

OdometryFramePyramidType

enum cv::OdometryFramePyramidType

#include <opencv2/ptcloud/odometry_frame.hpp>

Indicates what pyramid is to access using getPyramidAt() method:

Enumerator
PYR_IMAGE  Python: cv.PYR_IMAGE	The pyramid of grayscale images.
PYR_DEPTH  Python: cv.PYR_DEPTH	The pyramid of depth images.
PYR_MASK  Python: cv.PYR_MASK	The pyramid of masks.
PYR_CLOUD  Python: cv.PYR_CLOUD	The pyramid of point clouds, produced from the pyramid of depths.
PYR_DIX  Python: cv.PYR_DIX	The pyramid of dI/dx derivative images.
PYR_DIY  Python: cv.PYR_DIY	The pyramid of dI/dy derivative images.
PYR_TEXMASK  Python: cv.PYR_TEXMASK	The pyramid of "textured" masks (i.e. additional masks for normals or grayscale images)
PYR_NORM  Python: cv.PYR_NORM	The pyramid of normals.
PYR_NORMMASK  Python: cv.PYR_NORMMASK	The pyramid of normals masks.
N_PYRAMIDS  Python: cv.N_PYRAMIDS

OdometryType

enum class cv::OdometryType

strong

#include <opencv2/ptcloud/odometry.hpp>

These constants are used to set a type of data which odometry will use

Parameters

DEPTH	only depth data
RGB	only rgb image
RGB_DEPTH	depth and rgb data simultaneously

Enumerator
DEPTH
RGB
RGB_DEPTH

RgbdPlaneMethod

enum cv::RgbdPlaneMethod

#include <opencv2/ptcloud/depth.hpp>

Enumerator
RGBD_PLANE_METHOD_DEFAULT  Python: cv.RGBD_PLANE_METHOD_DEFAULT

VolumeType

enum class cv::VolumeType

strong

#include <opencv2/ptcloud/volume_settings.hpp>

Enumerator
TSDF
HashTSDF
ColorTSDF

Function Documentation

depthTo3d()

void cv::depthTo3d	(	InputArray	depth,
		InputArray	K,
		OutputArray	points3d,
		InputArray	mask = noArray() )

Python:
	cv.depthTo3d(	depth, K[, points3d[, mask]]	) ->	points3d

#include <opencv2/ptcloud/depth.hpp>

Converts a depth image to 3d points. If the mask is empty then the resulting array has the same dimensions as depth, otherwise it is 1d vector containing mask-enabled values only. The coordinate system is x pointing left, y down and z away from the camera

Parameters

depth	the depth image (if given as short int CV_U, it is assumed to be the depth in millimeters (as done with the Microsoft Kinect), otherwise, if given as CV_32F or CV_64F, it is assumed in meters)
K	The calibration matrix
points3d	the resulting 3d points (point is represented by 4 channels value [x, y, z, 0]). They are of the same depth as depth if it is CV_32F or CV_64F, and the depth of K if depth is of depth CV_16U or CV_16S
mask	the mask of the points to consider (can be empty)

depthTo3dSparse()

void cv::depthTo3dSparse	(	InputArray	depth,
		InputArray	in_K,
		InputArray	in_points,
		OutputArray	points3d )

Python:
	cv.depthTo3dSparse(	depth, in_K, in_points[, points3d]	) ->	points3d

#include <opencv2/ptcloud/depth.hpp>

Parameters

depth	the depth image
in_K
in_points	the list of xy coordinates
points3d	the resulting 3d points (point is represented by 4 chanels value [x, y, z, 0])

findPlanes()

void cv::findPlanes	(	InputArray	points3d,
		InputArray	normals,
		OutputArray	mask,
		OutputArray	plane_coefficients,
		int	block_size = 40,
		int	min_size = 40 *40,
		double	threshold = 0.01,
		double	sensor_error_a = 0,
		double	sensor_error_b = 0,
		double	sensor_error_c = 0,
		RgbdPlaneMethod	method = RGBD_PLANE_METHOD_DEFAULT )

Python:
	cv.findPlanes(	points3d, normals[, mask[, plane_coefficients[, block_size[, min_size[, threshold[, sensor_error_a[, sensor_error_b[, sensor_error_c[, method]]]]]]]]]	) ->	mask, plane_coefficients

#include <opencv2/ptcloud/depth.hpp>

Find the planes in a depth image

Parameters

points3d	the 3d points organized like the depth image: rows x cols with 3 channels
normals	the normals for every point in the depth image; optional, can be empty
mask	An image where each pixel is labeled with the plane it belongs to and 255 if it does not belong to any plane
plane_coefficients	the coefficients of the corresponding planes (a,b,c,d) such that ax+by+cz+d=0, norm(a,b,c)=1 and c < 0 (so that the normal points towards the camera)
block_size	The size of the blocks to look at for a stable MSE
min_size	The minimum size of a cluster to be considered a plane
threshold	The maximum distance of a point from a plane to belong to it (in meters)
sensor_error_a	coefficient of the sensor error. 0 by default, use 0.0075 for a Kinect
sensor_error_b	coefficient of the sensor error. 0 by default
sensor_error_c	coefficient of the sensor error. 0 by default
method	The method to use to compute the planes.

registerDepth()

void cv::registerDepth	(	InputArray	unregisteredCameraMatrix,
		InputArray	registeredCameraMatrix,
		InputArray	registeredDistCoeffs,
		InputArray	Rt,
		InputArray	unregisteredDepth,
		const Size &	outputImagePlaneSize,
		OutputArray	registeredDepth,
		bool	depthDilation = false )

Python:
	cv.registerDepth(	unregisteredCameraMatrix, registeredCameraMatrix, registeredDistCoeffs, Rt, unregisteredDepth, outputImagePlaneSize[, registeredDepth[, depthDilation]]	) ->	registeredDepth

#include <opencv2/ptcloud/depth.hpp>

Registers depth data to an external camera Registration is performed by creating a depth cloud, transforming the cloud by the rigid body transformation between the cameras, and then projecting the transformed points into the RGB camera.

uv_rgb = K_rgb * [R | t] * z * inv(K_ir) * uv_ir

Currently does not check for negative depth values.

Parameters

unregisteredCameraMatrix	the camera matrix of the depth camera
registeredCameraMatrix	the camera matrix of the external camera
registeredDistCoeffs	the distortion coefficients of the external camera
Rt	the rigid body transform between the cameras. Transforms points from depth camera frame to external camera frame.
unregisteredDepth	the input depth data
outputImagePlaneSize	the image plane dimensions of the external camera (width, height)
registeredDepth	the result of transforming the depth into the external camera
depthDilation	whether or not the depth is dilated to avoid holes and occlusion errors (optional)

rescaleDepth()

void cv::rescaleDepth	(	InputArray	in,
		int	type,
		OutputArray	out,
		double	depth_factor = 1000.0 )

Python:
	cv.rescaleDepth(	in_, type[, out[, depth_factor]]	) ->	out

#include <opencv2/ptcloud/depth.hpp>

If the input image is of type CV_16UC1 (like the Kinect one), the image is converted to floats, divided by depth_factor to get a depth in meters, and the values 0 are converted to std::numeric_limits<float>::quiet_NaN() Otherwise, the image is simply converted to floats

Parameters

in	the depth image (if given as short int CV_U, it is assumed to be the depth in millimeters (as done with the Microsoft Kinect), it is assumed in meters)
type	the desired output depth (CV_32F or CV_64F)
out	The rescaled float depth image
depth_factor	(optional) factor by which depth is converted to distance (by default = 1000.0 for Kinect sensor)

warpFrame()

void cv::warpFrame	(	InputArray	depth,
		InputArray	image,
		InputArray	mask,
		InputArray	Rt,
		InputArray	cameraMatrix,
		OutputArray	warpedDepth = noArray(),
		OutputArray	warpedImage = noArray(),
		OutputArray	warpedMask = noArray() )

Python:
	cv.warpFrame(	depth, image, mask, Rt, cameraMatrix[, warpedDepth[, warpedImage[, warpedMask]]]	) ->	warpedDepth, warpedImage, warpedMask

#include <opencv2/ptcloud/depth.hpp>

Warps depth or RGB-D image by reprojecting it in 3d, applying Rt transformation and then projecting it back onto the image plane. This function can be used to visualize the results of the Odometry algorithm.

Parameters

depth	Depth data, should be 1-channel CV_16U, CV_16S, CV_32F or CV_64F
image	RGB image (optional), should be 1-, 3- or 4-channel CV_8U
mask	Mask of used pixels (optional), should be CV_8UC1, CV_8SC1 or CV_BoolC1
Rt	Rotation+translation matrix (3x4 or 4x4) to be applied to depth points
cameraMatrix	Camera intrinsics matrix (3x3)
warpedDepth	The warped depth data (optional)
warpedImage	The warped RGB image (optional)
warpedMask	The mask of valid pixels in warped image (optional)