低分辨率条件下鞋类的自动分类方法

Automatic Footwear Classification Method under Low Resolution Condition

根据视频监控中行人所穿鞋的鞋型搜索犯罪嫌疑人是公安机关常用侦查技战法。然而在现实案件中很多视频监控分辨率较低,公安民警不能精确识别到具体鞋型,且消耗大量的时间和警力。针对这一问题,提出一种对低分辨率视频监控下的鞋类进行自动分类的方法。参考全国制鞋标准化技术委员会2017年制定的制鞋标准,初步将鞋类分为皮鞋和运动鞋两大类;构建鞋类数据库,包括59853幅皮鞋和47878幅运动鞋图像;进而基于卷积神经网络,设计一种适用于鞋类自动分类的鞋类识别网络模型。实验结果表明,鞋类自动分类模型在测试阶段对鞋分类的准确率达到了95.7%,可见基本能准确识别皮鞋和运动鞋两类鞋。

In video monitoring,it is a common investigation technique for public security organizations to search criminal suspects based on the shoe shape of pedestrians.However,in real cases,many video monitoring images are fuzzy,and the police cannot accurately identify the shoe type,which consumes much time and police force.To solve the problem,an automatic classification method of footwear under video monitoring was proposed.Referring to the year 2017 standards of China’s national technical committee on shoe making,all footwear could be systematically classified in general into two major categories:leather shoes and sneakers.A footwear database was constructed,including 59853 leather shoes and 47878 sneakers.Then,based on the convolutional neural network,a footwear-recognition network model was designed suitable for automatic footwear classification.The experimental results show that the success of the automatic footwear classification reached 95.7%in the test stage,and thus the method is able to identify leather shoes and athletic shoes accurately.