基于改进词包模型的车型识别算法

摘要

针对基于原始词包模型的车型识别算法识别速度慢、识别率低的问题,提出了一种基于改进词包模型的车型识别算法.首先使用Dense-SURF算法提取图像特征,并通过改进稠密采样策略进一步提高特征提取速度;然后提出特征上下文-矢量量化(FC-VQ)编码算法,并用其对特征向量进行编码,使编码后的特征包含空间位置信息,进而提高识别率;最后采用快速直方图相交核作为核函数,将提取到的特征送入SVM分类器进行训练或识别.实验结果表明:与其它车型识别算法相比,论文算法识别速度更快且识别率更高.%Aiming at the problem that vehicle type recognition algorithms based on original bag-of-words model run slowly and inaccurately,a vehicle type recognition algorithm based on improved bag-of-words model is proposed. Firstly,Dense-Surf method is used to extract features when stratage of dense sample has being optimized to speed up the process;Secondly,a new algo-rithm named feature context-vector quantization(FC-VQ)is proposed to encode features,from which the location information of features can be expressed clearly,and the recognition rate increases synchronously. Finally,fast histogram intersection kernel is used as kernel function of SVM classifier for training and recognizing processes of encoded features.Experimental results show that the algorithm proposed in this paper has higher recognition rate and faster recognition speed compared with other vehicle type recog-nition algorithms.