Fundamentals of stereovision

Stereo vision is a possibility to add additional information to images and videos. Disparity information representing the distance of image points to the camera is added to color and brightness information of individual pixels. To achieve this, two cameras are mounted at specific positions to produce two images. The cameras have to be calibrated by means of a chessboard, which is presented at several different positions. When the corners of each individual square are detected and brought into correlation between left and right image the intrinsic and extrinsic parameters of the camera set can be calculated and stored. These parameters are then used to warp and remap the original grabbed video images to undistorted and aligned images (rectified images). To calculate the distance information out of these two images, corresponding points in both images have to be found. The rectification makes the finding of corresponding points much easier and faster, because we know that the corresponding point can be found on the same horizontal line (epipolar line). The disparity image should contain dense information about corresponding points for as much points as possible. Due to occlusions and noise this is not always possible. Different methods are investigated to solve this problem with as less processing effort as possible. This technology can be added to a system, which differentiates between background and new objects in images, as used in video surveillance, to detect new objects. In the tutorial an overview of the technology used will be given, samples of different image and videos will be presented and different methods to calculate disparity maps will be presented. A live demonstration of an application used in object detection will be given and the method “codebook” to differentiate between background and foreground will be explained.