Stereo Vision based Full-Range Object Detection and Tracking including Pedestrians 2012-01-0286

Active safety applications can be divided into two categories according to the sensor technology the system uses: short range applications and long range applications. Short range applications require reliable sensing of a wider field-of-view (FOV) but limited range, while long range applications require accurate sensing of a long range but within a limited FOV. In this paper, we present a stereo vision based system that is capable of detecting, classifying different type of objects including vehicles and pedestrians, and tracking objects in full range and wide FOV, which effectively enables both short range and long range applications to happen simultaneously in a single system. Further, the system is more strengthened by its ability to maintain multiple sub-classes of detected vehicles and pedestrians. The developed system runs at 20 Hz on embedded hardware, computing 1280×456 range maps, road surface estimation, object detection, segmentation, classification, and tracking. The system can detect pedestrians very reliably up to 60m and has a general long range detection capability of up to. Test results are presented to demonstrate the performance of the developed system.