基于单目视觉的运动行人检测与跟踪方法

Method of moving pedestrian detection and tracking based on monocular vision technology

为了有效检测与跟踪城市交通环境中的行人,提出了一种在摄像机静止情况下基于单目视觉的运动行人检测与跟踪方法。检测阶段通过自适应背景模型快速提取背景图像,用动态多阈值方法二值化差分图分割运动行人;跟踪阶段引入灰色模型作为行人运动模型,预测行人运动,融合行人多种特征建立目标匹配模板,对行人连续跟踪。通过单个行人通行和多个行人同时出现这两种交通环境下的视频图像对本方法进行了验证,单个行人通行时,跟踪的正确率为95%;多个行人同时通行时,识别每个行人并分别跟踪的正确率为87%。

In order to effectively detect and track moving pedestrian in urban traffic scenes, a method of moving pedestrian detection and tracking by using the data from a fixed CCD camera was presented. Self-adaptive background subtraction method and dynamic multi-threshold method were adopted for background subtraction and image segmentation respectively. During the process of tracking, a new method based on gray model GM （1,1） was proposed to predict pedestrian motion, a template of pedestrian continuous tracking was presented by fusing several characters of targets. Experimental results of two real urban traffic scenes show that 95% of single pedestrians can be detected and tracked, 87% of multi-pedestrians can be detected and tracked. 4 figs, 10 refs.