HandTrackingClient::OpenCVCamera Struct Reference

A class representing the intrinsic/extrinsic parameters of the depth camera. More...

#include <OpenCVCamera.h>

Public Member Functions
	OpenCVCamera (const size_t imageWidth_in, const size_t imageHeight_in, const Transformd &extrinsics_in, const double fx_in, const double fy_in, const double cx_in, const double cy_in, const double k1_in, const double k2_in, const double p1_in, const double p2_in, const double k3_in)
	Initialize the camera model.
	OpenCVCamera ()
	Initializes the camera to be a basic QVGA camera looking down the z axis.
Matrix4d	glModelViewMatrix () const
	Computes the OpenGL modelview matrix corresponding to this camera. More...
Matrix4d	glProjectionMatrix (double near_z, double far_z) const
	Computes the OpenGL projection matrix corresponding to this camera. More...
Vector3d	worldToCamera (const Vector3d &p_world) const
	Takes a world space point and transforms it into camera space. More...
Vector3d	cameraToImage (const Vector3d &p_camera) const
	Takes a point in camera space and transforms it to the image plane. More...
Vector3d	project (const Vector3d &p_world) const
	Takes a world space point and projects it to the image plane. More...

Public Attributes
Transformd	extrinsics
	The transform taking desk to camera space. More...
size_t	imageWidth
	The width of the depth image, in pixels.
size_t	imageHeight
	The height of the depth image, in pixels.
Intrinsics
double	fx
	x focal length (in pixels).
double	fy
	y focal length (in pixels).
Distortion parameters
double	cx
	Camera center x coordinate (in pixels, from left).
double	cy
	Camera center y coordinate (in pixels, from top).
double	k1
	Radial distortion quadratic term.
double	k2
	Radial distortion r^4 term.
double	k3
	Radial distortion r^6 term.
double	p1
	Tangential distortion term.
double	p2
	Tangential distortion term.

Detailed Description

A class representing the intrinsic/extrinsic parameters of the depth camera.

Our camera model is identical to the one used in OpenCV (and in OpenNI). A full description can be found here: http://docs.opencv.org/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html

The extrinsic matrix is set up such that world space corresponds to the desk frame used in hand tracking, with y pointing up and z pointing toward the user.

Everything here will be in double precision, because this is necessary for dealing robustly with distortion (some distortion terms use the 6th power of the radius, which would cause precision loss when working with 32-bit floats).

One possible gotcha: in OpenCV camera space, the coordinate system has the positive z vector pointing into the frame and the positive y vector pointing down from the top of the image. Note that this is flipped from OpenGL (where the negative z vector points into the frame).

This class includes the worldToCamera and cameraToImage functions for projecting a world space point down to the image plane. Because inverting camera distortion is considerably complicated, the inverses of these functions are not provided. If you need undistortion functionality, you can find it within OpenCV; take a look at the undistortPoints function.

Member Function Documentation

Vector3d HandTrackingClient::OpenCVCamera::cameraToImage

(

const Vector3d &

p_camera

)

const

Takes a point in camera space and transforms it to the image plane.

In image space, x and y correspond to actual pixel values, measured from the top left corner of the image.

The z coordinate gets passed straight through from camera space without modification.

Matrix4d HandTrackingClient::OpenCVCamera::glModelViewMatrix

(

)

const

Computes the OpenGL modelview matrix corresponding to this camera.

In most cases it should be used together with the result of glProjectionMatrix. It can be passed to OpenGL using

const Matrix4d modelViewMat = camera.glModelViewMatrix();
glMatrixMode(GL_MODELVIEW);
glLoadIdentity();
glMultMatrixd(projectionMat.data);

Matrix4d HandTrackingClient::OpenCVCamera::glProjectionMatrix	(	double	near_z,
		double	far_z
	)		const

Computes the OpenGL projection matrix corresponding to this camera.

In most cases it should be used together with the result of glModelViewMatrix. It can be passed to OpenGL using

// Distances are in millimeters; 2m is a reasonable far plane:
const Matrix4d projMat = camera.glProjectionMatrix(10.0f, 2000.0f);
glMatrixMode(GL_PROJECTION);
glLoadIdentity();
glMultMatrixd(projectionMat.data);

• near_z The distance to the near z plane (in mm).
• far_z The distance to the near z plane (in mm).

Vector3d HandTrackingClient::OpenCVCamera::project ( const Vector3d &  p_world ) const

inline

Takes a world space point and projects it to the image plane.

Combines worldToCamera with cameraToImage.

Vector3d HandTrackingClient::OpenCVCamera::worldToCamera

(

const Vector3d &

p_world

)

const

Takes a world space point and transforms it into camera space.

Camera space is the space aligned with the camera center where the positive z vector points into the frame.

Member Data Documentation

Transformd HandTrackingClient::OpenCVCamera::extrinsics

The transform taking desk to camera space.

This is the camera extrinsics matrix which takes a point from desk space (where the origin is roughly at the center of the checkerboard you used to calibrate) to camera space (looking down the z axis).

---

The documentation for this struct was generated from the following files:

• OpenCVCamera.h
• OpenCVCamera.cpp