基于阴影和类Haar特征的动态车辆检测

Dynamic vehicle detection based on shadow and Haar-like feature

提出了一种基于车载相机的前向车辆检测方法。首先联合车底阴影和车辆垂直边缘产生车辆假设,针对阴影分割易受光照和路面灰度突变的影响,提出了直方图峰谷分析法(HVAM)来获得自适应阴影分割阈值;采用类Haar特征来检测车辆垂直边缘,以局部区域统计学方法克服了传统边缘检测算子易受外界因素影响的缺点;然后使用基于V-HOG特征结合SVM的车辆分类器对假设区域进行验证;最后推算检测到的车辆目标航迹,建立加速度模型,对多个车辆目标进行跟踪以提高整体性能。多种交通场景试验表明:该方法能够稳定准确地检测到前方车辆,每帧平均检测时间仅需35 ms,跟踪7 ms,远远满足实时性要求。

A vision based vehicle detection system is developed to detect front vehicles in a driver assistance system to improve vehicle safety. Vehicle hypotheses are generated by combining shadow underneath vehicle and vehicle vertical edge. Histogram valley analysis method （HVAM） is proposed to get adaptive shadow threshold. This meth- od can significantly solve the influence of illumination and the gray value mutation of road surface to shadow seg- mentation. Haar-like feature is used to detect the vertical edges of vehicle. It overcomes the weakness that tradi- tional edge deteetion operators are easily affected by surroundings. These hypotheses are further verified by using a classifier based on V-HOG features and SVM. Object trace prediction algorithm is used to track the detected vehi- cles to improve system performance. The system is tested under different scene and illumination circumstance. The results show that the system can accurately detect the front vehiele stability with only takes 35 ms on detection per frame and 7ms on tracking.