基于粒子滤波的系数编码车辆跟踪方法

传统基于生成式的车辆跟踪方法仅考虑了目标信息,忽略了车辆背景信息,降低了目标与背景的表征能力.针对复杂背景条件下视觉导航对车辆跟踪精度的需求,提出了一种基于粒子滤波的系数编码车辆跟踪方法.该方法首先对获取的帧图像进行仿射变换归一化处理,并采用深度去噪自编码器对变换后的图像进行完备特征字典的生成;接着,采用稀疏编码对完备特征字典进行降维处理,消除网络高层目标特征的冗余信息,保留网络底层的高效关联特征;最后,将提取的深度稀疏编码特征应用到粒子滤波的框架内实现车辆的有效跟踪,有效克服了判别式跟踪方法无法处理遮挡问题的缺陷.实验结果表明,在尺度变化、光照变化以及遮挡等复杂环境下,本文方法将跟踪精度提升了17％,每秒处理的帧图像速度提升了64％.

Vehicle tracking method based on particle filter depth sparse coding

The traditional vehicle tracking method which bases on generative model just considers the target information,and ignores the background information of the vehicle,which reducing the representational ability of the target and the background.In order to meet the requirement of visual navigation for vehicle tracking under the complex background,a new vehicle tracking method bases on particle filter for coefficient encoding is proposed in this paper.Firstly,the frame image is normalized by affine transformation,and the complete feature dictionary is generated by using the depth de-noising self-encoder (DDSE) algorithm.Then,the sparse coding is used to reduce the dimension of the complete feature dictionary,eliminate the redundant information of the high level feature of the network,and preserve the high efficient correlation feature.Finally,the extracted depth sparse coding features are applied to the framework of particle filter for tracking the vehicle effectively,which effectively overcomes the bug of the discriminant tracking method which cannot deal with the occlusion problem.The experimental results show that the tracking accuracy is improved by 17％,and the frame image processing speed per second is increased by 64％ in the complex environment such as scale variation,illumination variation and occlusion.