摘要
由于场景中存在目标遮挡与重叠的影响,以及检测时产生的分割错误,同一个目标往往被识别为若干运动区域,或者多个目标被识别为同一个运动区域,从而导致目标数量识别错误。为了解决这一问题,提出了一种结合融合与分离操作的理论来分析视频的方法,将多目标跟踪的问题转化成为一个找寻后验极值的问题。采用图论的方法表示观测结果,利用图像序列中目标运动轨迹和外观相似度的信息,将多目标跟踪的问题归结为找寻图中多个最优路径的问题。该方法采用了滑动窗口框架以便统计固定数目帧的信息。实验结果表明该方法能够应对实际中发生的上述现象,达到了准确识别多目标数量的目的。
Due to the noisy foreground segmentation, an object may be represented by several foreground regions and simi- larly one foreground region may correspond to multiple objects. To deal with this problem, we propose merge and split oper- ations to generate new hypotheses to the measurement graph. The multiple target tracking problem as a maximum posteriori problem is proposed, and a graph representation of all observations over time is adopted. To make full use of the visual ob- servations from the image sequence, both motion and appearance likelihood are introduced. The multiple target tracking problem is formulated as finding multiple optimal paths in the graph. The proposed approach uses a sliding window frame- work, which aggregates information across a fixed number of frames. Experimental results show that the method could han- dle this kind of situations and achieve the goals of accurate recognition of target number in multiple target tracking.
出处
《光学与光电技术》
2012年第3期42-46,共5页
Optics & Optoelectronic Technology
关键词
多目标跟踪
融合与分离
后验极值
识别
multiple target tracking
split and merge of detected regions
maximum a posteriori
recognition
