基于改进的PBAS算法的前景目标检测

Foreground Object Detection Based on Improved PBAS

针对像素层自适应分割算法(Pixel Based Adaptive Segmenter,PBAS)在动态背景下检测准确率低、静止或运动缓慢的前景目标被更新为背景以及出现鬼影干扰的问题,提出了一种结合像素级信息和区域级信息的改进的前景检测算法。首先,提出一种融合区域结构信息和区域颜色信息的背景复杂度衡量方式;然后,采用改进的背景复杂度来控制判定阈值和学习率,并检测前景;其次,对像素层的检测结果使用区域窗口进行空间邻域对比,以消除鬼影;最后,引入前景计数机制来保证静止前景不被更新为背景。实验结果表明,该算法对光照条件和前景运动速度不敏感,能有效地从背景中检测出完整的前景目标,并迅速地消除鬼影干扰,准确率达到了92.7%。

To avoid the problems of PABS （Pixel Based Adaptive Segmenter） such as low accuracy rate of detection, re- placement of immobile of slow-moving foreground objects by background ones, and the interference of ghosting, this pa- per presented a new PBAS which is improved by merging pixel-level information and region-level information. A mea- sure of background dynamics fused with structural information and color information at region level is computed firstly. And then this measure will be used to estimate and control the threshold and learning rate, as well as to detect the fore- ground object. A spatial neighborhood contrast on pixel-level result is computed in order to solve the interfe-rence of ghosting, and a foreground count machine is introduced to avoid the missing of static object in foreground. The experi- ment results indicate that the algorithm is insensitive to brightness and velocity of objects, thus the foreground object can be detected effectively, and interference of ghosting can be removed quickly with a high accuracy detection rate 92. 7%.