一种尺度自适应的长时目标追踪算法

A Scale Self-adaptive Long-term Target Tracking Algorithm

在长时目标追踪中,传统的核相关滤波算法受到目标尺度变化和环境因素的影响,追踪效果会有所下降。为解决这一问题提出了一种尺度自适应的长时目标追踪算法。首先,为了实现追踪过程中追踪器尺度的自适应,在核相关滤波算法中加入尺度因子池,通过不同尺度下候选目标的响应值判断目标的最佳尺度;其次,为了提高追踪的准确度,通过扩大候选目标的搜索范围,对追踪不准确的目标位置进行重新检测;最后为了提高追踪效率,根据追踪的稳定性决定是否对追踪模板进行更新,从而提高追踪速度,减少过多错误信息的学入。实验结果表明,所提算法相较于其他追踪算法在精确度上提高了15.3%,在成功率上提高了17.1%。

In long-term object tracking, the tracking accuracy of traditional kernelized correlation filter algorithm decreases due to the effects of the scale changes of targets and environmental factors. To solve this problem, a scale self-adaptive KCF algorithm for long term object tracking was proposed. Firstly, in order to achieve the scale self-adaption in the tracking process, a scale factor pool was added to the kernel correlation filter algorithm, and the optimal scale of the target was judged by the response value of the candidate target at different scales. Secondly, in order to improve the tracking accuracy, the search area of the candidate targets was expanded and the positions of inaccurate targets were re-detected. Finally, in order to improve the tracking efficiency, the tracking template was updated according to the stability of the tracking. By this way, the tracking speed was improved, and the learning time of excessive error information was reduced. The results indicate that, compared with other tracking algorithms, the proposed algorithm improves the tracking accuracy by 15.3% and the rate of tracking success by 17.1%.