一种基于视频的道路行人检测方法

摘要

An approach for detecting road pedestrian based on traffic video is proposed. Firstly, self-adaptive background updating algorithm is used to detect moving target. Then, the characters of pedestrian's outline and linear relationship between the height of pedestrian in pixels and actual distance are adopted to extract pedestrian. In addition,the center of pedestrian gravity is taken to track moving pedestrian, and the speed is obtained. Finally, the speed of the pedestrian features is combined to achieve the pedestrian target detection. Experimental results show that this method, with reasonably realtime performance and robustness, can detect the pedestrian appearing in traffic scene, and effectively solve the influence of shelter by moving vehicles and illumination change.%提出了一种基于交通视频的道路行人检测方法.首先,采用具有自适应背景更新的背景差法提取运动目标.其次,利用行人的轮廓特征和行人在图像上的像素高度与实际距离的线性变化关系特征进一步分割行人目标,并依据行人位置特征对其做多帧匹配跟踪.最后,结合行人速度特征,实现行人目标检测.实验结果表明,该方法能准确地检测交通场景中出现的行人目标,有效地解决运动车辆的遮挡和光照等对行人目标检测的影响,具有良好的实时性和稳健性.