基于改进SUSAN原则的车辆检测方法

Vehicle Detection Based on Improved SUSAN Algorithm

为解决汽车辅助驾驶系统中目标车辆检测的实时性和鲁棒性问题,提出一种基于单目视觉的车辆检测系统,将改进的SU—SAN(Smallest Univalue Segment Assimilating Nucleus,即最小核值相似区)算法应用到车辆边缘检测中;采用自适应双阈值法检测车底阴影。结合车道线参数动态规划车辆初始检测区域;在检测区域中,采用改进的SUSAN算法定位车辆边缘,生成车辆假设;最后根据车辆的纹理、形状和位置特征来验证车辆假设;为改善系统性能,采用Kalman滤波算法对检测到的目标进行跟踪;使用实际采集的道路图像序列对系统进行测试。实验表明,该系统能够及时准确地检测前方目标车辆。

In order to achieve the real--time and robustness of vehicle detection in a driver assistance system, a method of vehicle detec tion based on a single camera is developed using improved SUSAN algorithm. First, the adaptive double threshold is used to detect shadow under vehicle. The initial detection region is dynamically estimated according to shadow positions and lane parameters. Then improved SU SAN algorithm is used to locate vehicle edge and generate hypotheses of vehicle in this region. The texture, shape and position features of vehicle are used to verify the hypotheses. Finally, The Kalman filter is used to track objects to improve the system performance. The system is tested with image sequences taken on roads. Experiment results show that the system can detect proceeding vehicles effectively.