基于局部最大概率特征和映射模型学习的行人再识别

摘要

进行了行人再识别研究.针对行人自身差异和相机视域差异的存在导致再识别率低的问题,提出了一种基于局部最大概率特征和映射模型学习的行人再识别算法.该算法首先对行人图像提取局部最大概率特征,克服光照变化并提取图像的完整信息;然后学习交叉映射模型,利用学习好的模型进行行人特征变换,从而消除不同摄像机拍摄区域的特征差异;最后进行距离度量和排序.实验表明,该算法合理有效,能够获得较为完整的判别性特征表示,成功地提高了行人再识别的匹配精度.%The technique of person re-identification is studied .Aiming to solve the problem of low re-identification rate caused by pedestrians '' own difference and the difference of camera view , a person re-identification algorithm based on local maximal occurrence features and the mapping model is proposed .Firstly , the algorithm extracts the local maximal occurrence features from pedestrians '' images, so as to overcome illumination changes and get complete image information .Secondly , it learns the cross-view mapping model to convert pedestrian features to eliminate characteristic differences of different camera views .Finally, the features will be sorted by distance metric learning . The experimental results show that the proposed person re-identification algorithm is effective enough to obtain a more complete discriminant feature representation so that it can improve the matching accuracy for person re -identi-fication.