摘要： 对现有基于压缩感知的视频目标跟踪系统进行改进，提出一种可实现隐私保护的监控视频目标跟踪系统。在编码端采用结构化随机矩阵，以提高随机采样矩阵的生成速度。在解码端采用GPSR-BB算法，以提高系统抗噪性。利用粒子滤波器算法实现目标跟踪，减少跟踪结果误差对压缩感知恢复算法准确性的影响和分析时间。实验结果表明，该系统在实现隐私保护的同时，提高了系统对光照的鲁棒性，在室内外光照条件下均能准确跟踪目标。与BP和Lasso方法相比，分别可节约30.3%和51.6%的处理时间。

关键词: 压缩感知, 目标跟踪, 粒子滤波, 结构化随机矩阵, 梯度投影, 基追踪

Abstract: The proposed system of privacy-enabled object tracking is improved based on the primary one. After using the structurally random matrices at the encoder, the generation speed of random sampling matrix is improved. After using fast methods such as GPSR-BB in reconstruction and particle filtering in analysis at the decoder, the noise resistance of system is improved. It uses particle filtering algorithm for target tracking, reduces the tracking results error influence on compression perception recovery algorithm accuracy, as well as the time needed for analysis steps. Experimental result shows that the proposed framework enables the privacy in tracking and at the same time increases the robustness to illumination condition. It also avoids the error of tracking results affecting the accuracy of CS reconstruction algorithm, which is faster than the original one and the performance of tracking is excellent. The process time of this system is saved 30.3% and 51.6% of the BP and Lasso method.

Key words: compressive sensing, object tracking, particle filtering, structured random matrices, gradient projection, basis pursuit

中图分类号: 

• TN919.8