运动背景下多目标跟踪的小波方法

Wavelet method for multi-target tracking in a moving background

为了能从运动背景中检测其中的运动目标,并进行跟踪,提出一种基于小波变换的分层匹配跟踪算法。利用小波分解的多层子图进行分层匹配,估计整个背景的运动矢量;利用差分算法从运动背景中检测出多个运动目标,计算出多个动目标的形心坐标,绘出各动目标的运动轨迹。该算法与传统的块匹配算法相比,滤除了原图像的高频噪声,防止了在含噪原图像上进行块匹配不准确的缺点;另外,在低频分量图像上N×N范围进行块匹配,相当于在原图像上2nN×2nN的范围进行匹配搜索,搜索速度快。当相邻两帧背景运动向量小于10个像素,运动目标相对背景的运动向量小于5个像素时,实验结果证明了此算法的有效性和可行性。

In order to detect and track the moving target in a moving background, a tracking algorithm based on multi-layer block matching of wavelet transform is proposed. The moving vector of the whole background is estimated by means of multi-layer matching for multi-layer sub-images of wavelet decomposition and multi-moving targets are detected from the moving background by difference algorithm. The centroid coordinates of multi-moving targets are calculated and the moving traces for various moving targets can be drawn out. Compared with traditional block matching algorithm, the high-frequency noises of original image are filtered and the fault of inaccurate block matching performed with original images with noises is prevented. In addition, block matching performed in N×N range of low frequency sub-image corresponds to block matching performed in 2nN×2nN range of original image, the searching speed is very fast. Experiment results show that when two frames of adjacent background moving vectors are less than 10 pixels, the moving vector of the moving target for the relative background is less than 5 pixels, the algorithm is effective and feasible.