摘要： 为提高基于行人属性学习的行人再识别算法识别精度,提出一种改进的行人再识别神经网络模型。该模型基于属性之间的异质性,在神经网络中设计不同的识别方法对各类属性进行识别,以提高行人属性识别的准确率。针对不同属性识别方法损失度量算法的不一致,给出异质属性损失度量函数,使得不同识别方法能在同一个网络模型中进行训练和学习,实现网络参数的最优化。实验结果表明,该模型在Market1501数据集、DukeMTMC数据集和DukeMTMC数据集上的首位准确率分别达到88.13%、74.96%和77.64%。

关键词: 行人再识别, 异质性, 深度学习, 属性分类, 回归预测, 多分类

Abstract: In order to improve the recognition accuracy of pedestrian re-identification algorithm based on pedestrian attribute learning,an improved pedestrian re-identification neural network model is proposed.Based on the heterogeneity between attributes,different recognition methods are designed in the neural network to identify various kinds of attributes to improve the accuracy of pedestrian attribute recognition.Aiming at the inconsistency of the loss measurement algorithm of different attribute recognition methods,the heterogeneous attribute loss measurement function is given,so that different recognition methods can be trained and learned in the same network model to optimize the network parameters.Experimental results show that the first accuracy rate of the model on the Market1501 dataset,DukeMTMC dataset and DukeMTMC dataset respectively reaches 88.13%,74.96% and 77.64%.

Key words: pedestrian re-identification, heterogeneity, deep learning, attribute classification, regression prediction, multiple classification

中图分类号: 

• TP18