基于毫米波雷达和机器视觉的夜间前方车辆检测

摘要

A leading-vehicle night-detection method was proposed based on milimeter-wave radar-vision, using the fusion of data from multi-sensors to investigate the inteligent warning system for avoidance vehicles colision at night with preceding vehicles. A world coordinate of the preceding vehicles was built for milimeter wave radar target to form the region of interesting image after relationship transformation from the world coordinate to image pixels coordinate. Image processing method reduced interference from the outside environment. A general value of reliability was achieved to test vehicles in interest depending on Dempster-Shafer Evidence Theory (D-S) which fused feature information. Several sections of video for vehicles driving on road were colected during experiment. The statistical data of frames of tailight identiifed were achieved and compared by subjective judgment. The results show that the method effectively eliminates the inlfuence of ilumination condition at night, accurately detect leading vehicles and determine their location.%为研究夜间追尾事故中本车智能防撞预警方法，提出了一种基于毫米波雷达和机器视觉的前方车辆检测方法。利用多传感器融合数据，检测前方车辆的距离、速度等。建立传感器之间转换关系，转换雷达目标的世界坐标到图像坐标。在图像上形成感兴趣区域，利用图像处理方法减少干扰点，运用Dempster-Shafer（D-S）证据理论，融合特征信息，得到总的信任度值检验感兴趣区域内的车辆。实验采集多段夜间道路行车视频数据，统计实现尾灯识别的帧数，与主观判断进行比较。结果表明：该方法能够实现对夜间前方车辆的检测和定位。