论文部分内容阅读
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 influence of the target occlusion and overlap in the scene and the segmentation error caused by the detection, the same target is often identified as a number of motion regions, or multiple targets are identified as the same motion region, resulting in a wrong identification of the target number. In order to solve this problem, a method of video analysis based on the theory of fusion and separation is proposed. The problem of multi-target tracking is transformed into a problem of finding posterior extreme. The method of graph theory is used to represent the observation results. By using the information of the target motion trajectory and the appearance similarity in the image sequence, the problem of multi-target tracking is reduced to finding the optimal paths in the graph. The method uses a sliding window frame in order to count a fixed number of frames of information. Experimental results show that this method can deal with the above phenomena in practice and achieve the purpose of accurately identifying multi-objective quantities.