关键词: 车辆环境感知, 车身姿态, 角点检测, 立体视觉, 弱极线约束

Abstract: Results of range measurement test imply that calibration errors will degrade the accuracy of the stereo vision system as the sensing distance increases, and make the epipolar constraint invalid. Therefore, conventional stereo vision algorithms are not applicable when the sensing range is large. A corner-based stereo vision system is designed for vehicle surrounding sensing applications where the sensing range is up to 80 m. Errors in the range measurement are corrected by a secondary calibration. Based on loose epipolar constraint, stereo vision processing of the corners detected in the image is performed to improve the success rate of matching. Lane detection is integrated with the obstacle-detection system to estimate the vision system’s pitch angle, which is indispensable in the transformation from the camera coordinates to the ground coordinate. For the purpose of real time processing, the sensing range is dynamically adjusted on the basis of the vehicle speed. Experimental results show that the system is able to reliably detect obstacles such as vehicles and pedestrians from 3 m to 80 m in front in real time.

Key words: Corner detection, Loose epipolar constraint, Stereovision, Vehicle pose calibration, Vehicular surroundings sensing