实时行人检测预警系统

针对重特大交通事故中的行人保护问题, 提出了基于侧面行人特征的实时行人检测预警系统(PDWS)。系统由检测模块和预警模块两部分组成, 其中检测模块使用Haar与HOG特征和AdaBoost与SVM分类器, 通过侧面行人样本库完成行人特征的提取与检测, 同时应用窗口拆分法与快速窗口扫描算法提高检测效率, 得到一个具有高检测率与低误检率的结果。在预警模块融合了行人距离、汽车速度及角速度信息, 判断前方行人存在碰撞的危险。使用系统及算法对城市环境复杂背景下横过街道的行人进行了实车验证。测试结果表明: 对704Pixel×576Pixel图像的检测帧率为13~18帧.s^-1, 检测率大于85%, 误检率小于1%, 预警时间小于1s, 实车验证结果达到了车载主动安全系统实时性与准确性的要求。

Real-time pedestrian detecting and warning system

A real-time pedestrian detecting and warning system(PDWS) based on the features of pedestrian side was proposed to solve the problem of pedestrian protection in serious traffic accidents. The system consisted of two parts, detecting module and warning module. The feature extraction and detection pedestrian were completed by using side pedestrian sample dataset in detecting module, Haar and HOG features together with AdaBoost and SVM classifiers were applied to complete the feature extraction and detection. Window-dividing method and operator context scanning(OCS) method were used to improve the detecting efficiency, and a result with both high detection rate and low false alarm rate was obtained. The velocity and angular velocity of automobile together with the distance of detected pedestrian were merged in warning module, so the system could judge the collision risk for the pedestrian in the front. The system and the algorithms were tested toward the pedestrian crossing street with a complex urban envirnment in real vehicle. Test result shows that for the images with 704 Pixel×576 Pixel, the frame rate is about 13-18 frame·s-1, the detecting rate is above 85%, the false detecting rate is below 1%, and the warning response time is less than 1 s. The result meets the requirements of on-board active safety system both in accuracy and real-time use.