利用D-S证据理论的夜间车辆检测

为了有效检测夜间车辆,提出了一种利用D-S证据理论进行夜间车辆检测的方法。首先在YCrCb颜色空间中采用阈值法对道路场景图像进行分割得到明亮块,提取各个明亮块的轮廓,利用轮廓四邻域偏红度水平消除非尾灯等虚假目标。其次,使用尾灯聚类算法组合车灯对,得到车辆假设。最后,利用车辆车尾中车灯对的面积比、互相关值以及车灯对组合框长宽比等结构化特征信息来构建基本信任分配函数,运用D-S证据理论,融合这些特征信息得到总的信任度值,最终设立信任度阈值验证车辆假设。该方法减少了主观阈值的数量,可有效降低经验不足导致阈值确定不当的风险,提高辨别率。实验结果表明,该方法效果明显,提高了检测精度,减少了误判,同时提高了系统的鲁棒性。

Night-time vehicle detection using D-S evidence theory

In order to effectively detect vehicles moving at night, this paper presented a new vehicle detection method for night-time vehicle using D-S evidence theory. Firstly, to collect all kind of contours of bright region in image, it segmented the road scene image by using threshold value in YCrCb color space and extracted the contour of each bright region, in which false targets such as non-rear light were excluded by using contour four-neighborhood red level method. Secondly, it paired the rear lamps using taillights clustering algorithm to get vehicles candidate hypothesis. At last, it used the structured feature information of taillights such as the area ratio, cross-image correlation and length-width ratio of the rear light combo box to construct the basic belief distribution function, and then constructed the general trust value by fusing these feature information and using D-S evidence theory, ultimately a belief threshold value was set to verify the vehicle hypothesis. The proposed method could decrease the number of subjective threshold, and reduced efficiently the risk of inappropriate threshold definition because of lacking experience, and thus could improve the recognition rate. The experimental results show that the method can improve the detection accuracy and reduce the misjudgment. It also improves the robustness of the system.