【摘要】 针对公共安全场所中如何通过非接触性的方法识别犯罪可疑分子，成为目前研究的热点之一，其中步态识别技术在公安视频侦查工作中具有良好的应用前景。现阶段，步态识别技术在公开数据集CASIA-B的正常行走状态下识别正确率已经达到了96%，但是行人在穿外套和携带包裹等有遮挡行走状态下的识别正确率效果不太理想。针对此问题，采用联合Triplet和Softmax两个损失函数的方法，同时在联合损失函数中加入L2正则化，从而优化网络步态模型，通过训练提高行人的步态识别准确率。研究结果表明，正常行走条件下的步态识别准确率维持在合理范围内，穿外套和携带包裹行走的步态识别准确率均有明显的提升，分别提升至93.03%和81.03%。更多还原

【Abstract】 For how to identify criminal suspects in public security places through non-contact methods, it has become one of the hot spots of research nowadays. Gait recognition technology has good application prospects in public security video investigation work. At present, the gait recognition technology has reached 96% in the normal walking condition in the public data set CASIA-B, but the correct recognition rate of pedestrians wearing coats and carrying parcels in the covered walking condition is less satisfactory. In order to solve the practical problem, this paper adopts the method of combining the loss functions of Triplet and Softmax and adding L2 regularization to the joint loss function to optimize the network gait model, so that the gait recognition accuracy of pedestrians can be more effective through training. The research results show that the accuracy of gait recognition under normal walking conditions is maintained within a reasonable range, and the accuracy of gait recognition for walking with jacket and package has been significantly improved to 93.03% and 81.03%respectively.更多还原