基于相关滤波器和最小二乘估计的目标跟踪方案

Target tracking scheme based on correlation filter and least squares estimation

针对复杂背景下基于视觉图像的目标跟踪问题,提出一种结合相关滤波和最小二乘估计(LSE)的目标跟踪方案。首先,对已有视频帧进行初始化处理,并构建目标的几何失真图像模板,以此设计一个相关滤波器。然后,当采集到新视频帧时,根据之前的目标位置序列,利用LSE预测出该帧中的目标位置。接着,提取目标图像,并计算其与目标模板之间的相关性。最后,通过阈值来判断是否检测到真正的目标,并进行反馈用来更新目标模板。在GPU上实现该方案,几个视频序列上的实验结果表明,该方案能够准确跟踪目标,且对噪声和光照等环境因素具有很好的鲁棒性。

For the issues that the target tracking based on visual image in complex background, this paper proposed a target tracking scheme combining correlation filtering and least squares estimation （LSE）. First, it initialized some existing video frames and constructed a geometric distortion image template to design a correlation filter. Then, when the new video frame was acquired, it predicted the target position in the frame by the LSE according to the previous target position sequence. Next, it extracted the target image and calculated the correlation between the target image and the target template. Finally, it used the threshold to determine whether a real target was detected and feedback was used to update the target template. Experiments on the GPU and the experimental results on several video sequences show that the scheme can accurately track the target and has good robustness to environmental factors such as noise and light.