为了提高校园监控系统对目标人物的识别准确度,从人脸检测算法与人脸识别算法两个模块对系统进行改进.一方面通过替换检测网络颈部结构中的常规卷积层为DO-Conv,并改进IOU损失函数,提高YOLOv5人脸检测算法的检测精度;另一方面通过引入h-swish激活函数与SE注意力机制,提高Mobile Face Net人脸识别算法的识别效果.实验证明,所提改进人脸检测算法,在保持YOLOv5算法较快运算速度的情况下,提高了算法的精确度,检测精度达到91％,综合性能最佳;所提改进人脸识别算法的识别精度同样在保证较小的参数量与较快的运算速度的同时,提高了算法的识别精度,相较于原始Mobile Face Net算法与SphereFace算法,提高了 0.11％和0.16％,具有更好的识别结果.基于改进人脸检测算法与改进人脸识别算法搭建的校园安全监控系统,对目标人物的识别准确率较高,平均准确率达到913.31％,符合校园安全监控需求,值得进一步研究和推广.
Smart campus surveillance video system construction and face recognition technology
In order to improve the recognition accuracy of the target person,the system is improved from two modules:face de-tection algorithm and face recognition algorithm.On the one hand,the detection accuracy of YOLOv5 face detection algorithm is im-proved by replacing the conventional convolution layer in the detection network neck structure with DO-Conv and improving the IOU loss function.On the other hand,by introducing h-swish activation function and SE attention mechanism,the recognition effect of Mobile Face Net face recognition algorithm is improved.Experiments show that the proposed improved face detection algorithm can improve the accuracy of the algorithm,and the detection accuracy reaches 91％,while maintaining the fast operation speed of YOLOv5 algorithm,and the comprehensive performance is the best.The recognition accuracy of the proposed improved Face recogni-tion algorithm is also improved by 0.11％and 0.16％compared with the original Mobile Face Net algorithm and SphereFace algo-rithm,while ensuring a small number of parameters and a faster operation speed.The campus security monitoring system based on im-proved face detection algorithm and improved face recognition algorithm has a high recognition accuracy of target people,with an aver-age accuracy of 913.31％,which meets the needs of campus security monitoring and is worthy of further research and promotion.
万方数据
维普
