低清晰视频的“白化-稀疏特征”车型分类算法

The“whitening-sparse coding”vehicle classification algorithm for low resolution video

车型识别分类,对低/高速行车道划分、流量统计,特别是超长/重、危险品车的识别具有现实意义.实验室曾提出的基于尺度不变特征转换SIFT、方向梯度直方图HoG视频检测方法抗干扰能力弱,在因道路环境差、网络拥塞随机造成图像模糊时,往往误判.为此,在机理上,分析比较了上述分类算法与特征白化、稀疏编码算法的局限或优势,提出了适应低清晰度视频的"白化-稀疏特征"车型分类算法.该分类算法采取PCA白化技术特征数据预处理、超完备基的凸优化迭代,从而获得稀疏编码特征数据.经与SIFT-SVM算法的现场图像检测比较,其在图像模糊条件时的分类准确率也能达到90%,一般优于93%,均耗时约0.04s.

Vehicle Classification makes great significance in the division of different lanes and traffic statistics.Videobased vehicle classification detection has a wide range of potential applications.Currently,the approaches based on Computer Vision may incur huge errors in the case of blur videos caused by the external environment.So after the discussion of the Whitening Preprocessing technic,we propose a＂Whitening-Sparse Coding＂vehicle classification algorithm which can adapt to low resolution videos.We compared the SIFT-SVM vehicle classification algorithm with the＂Whitening-Sparse Coding＂algorithm by theoretical analysis and experiments.By the use of the NingHuai Freeway we find that The＂Whitening-Sparse Coding＂algorithm can classify more quickly,be more adaptive to all kinds of environment and has a higher accuracy.