基于红外与可见光的鲁棒压缩感知跟踪方法

Target Tracking with Robust Compression Sensing Based on Infrared and Visible Frames

为了解决复杂背景下遮挡、光照变化或噪声干扰时跟踪的效果容易受到干扰的问题,提出了一种基于红外与可见光的双通道鲁棒的压缩感知目标跟踪方法。该方法提取红外和可见光双通道的多个特征,采用压缩感知的稀疏采样特性,去除稀疏跟踪算法中非负性假设。提出了一种粒子滤波框架下的压缩感知的目标跟踪算法,同时给出一种目标模板根据Bhattacharyya系数自适应更新方法。实验采用复杂环境下多组图像序列,结果表明该方法与3种优秀的跟踪算法相比,具有更强的鲁棒性与更高的跟踪精度,同时减少了数据计算量,实现了复杂环境下图像目标的稳定跟踪。

Tracking of target is easy to be disturbed due to sheltering, changing of illumination, or interference of noise under complex background. To solve the problem, a robust target tracking method based on binary channels of infrared and visible frames was proposed. Multiple features of the two channels were extracted, and the sparse sampling feature of compressed sensing was used to eliminate the non-negative hypothesis in sparse tracking algorithms. A new algorithm of compressed sensing was presented based on particle filter framework, and an adaptive method for target template updating was given, which was based on the Bhatta- charyya coefficients. Experiments were made using multiple groups of image sequences with complex terrain. The results show that： the proposed algorithm has better robustness, higher precision and lower computation burden than the other tracking algorithms, which can achieve a stable tracking of the image target in complex environment.