基于立体匹配的改进ViBe算法

An Improved Vi Be Algorithm Based on Stereo Matching

由于深度相机技术的不成熟,限制了融合深度信息的运动目标检测算法的应用范围。对此,提出一种高效的基于随机选择策略的立体匹配算法,用于代替深度相机为运动目标检测算法提供深度信息。首先,建立匹配点样本集合,并在每帧中随机选择匹配点计算其匹配代价和置信度;其次,更新集合并将置信度高的匹配点向邻域传播,随着时间的持续会由粗到精得到深度图;最后,用深度信息改进Vi Be算法(visual background extractor,视觉背景提取)得到最终的检测效果。实验结果表明,改进算法能够在400×300的分辨率下以8~10帧/秒的速度运行,消除鬼影和抑制光照突变的速度要比经典Vi Be算法快3倍以上,同时在普通场景中的运动目标的检测效果和消除阴影的能力也明显优于原算法。

Due to the immaturity of the depth camera technique,the application range of moving object detection algorithm with the fusion of depth information is restricted.For this,we propose an efficient stereo matching algorithm based on random selection strategy,which provides the depth information for moving object detection algorithm instead of the depth camera.First,the matching point sample sets are established,and their matching cost and confidence are computed by random selection of matching points in each frame. Then,the sets are updated and the matching points with high confidence are propagated to the neighborhood. As time goes on,the depth map will be obtained through a coarse-to-fine pattern.Lastly,the depth information is added to the conventional Vi Be algorithm,which leads to the final test results.The experiments showthat in the resolution of 400×300 the improved algorithm can achieve a nearly real-time frame rate of 8-10 frames per second and get the tripe speed of conventional Vi Be on removing ghost and suppressing pseudo-target caused by light changing.Meanwhile,it obtains better detection and shadowelimination performances in common scene.