Computer Vision for the Car Industry

This report provides a brief and informal introduction into stereo and motion analysis for driver assistance. Stereo and motion analysis play a central role in computer vision [10]. Many algorithms in this field have been proposed and carefully studied; see, for example, [2, 14] and the website vision.middlebury.edu for stereo and optic flow algorithms. In short, a stereo pair of images allows to identify pairs of corresponding points, and those allow to calculate the distance between the projected point in the three-dimensional world and the recording cameras. Fig. 1. A stereo pair captured by two cameras, installed on the windscreen of an ego-vehicle. The vehicle which is used to host the stereo camera platform for capturing stereo image sequences is called the ego-vehicle. Figure 1 illustrates an example of a stereo video sequence, recorded and processed at Daimler AG, Germany. After years of research on ego-motion estimation [1], the automobile industry has all the tools for producing rectified (i.e., geometrically corrected) stereo image sequences. Calculated distance values can then be visualized in a distance map, where gray values represent various distance levels. Figure 2 illustrates two depth maps