基于人工智能的嵌入式人流量检测数据分析系统设计

Artificial Intelligence-Based Designof Embedded Human Flow Detection Data Analysis System

随着计算机视觉和深度学习的高速发展,人流量数据分析在从多场合下得到越来越广泛应用,对人员密集的场所进行数据分析,具有较高的商业价值,同时对类似踩踏等安全问题也起着预防的作用。目前应用人工智能系统的设备成本较高,不利于推广,因此提出设计一个基于边缘计算的嵌入式人流量检测数据分析平台,采用基于Flask的嵌入式WEB服务器以及针对树莓派改进的YOLOv5算法,通过添加新的人脸检测头,使用Stem块替换focus层,修改SPP层中kernel等方式提高YOLOv5对于人特征的识别能力,实验表明,改进的YOLOv5算法在wider-face数据集下,平均正确率较YOLOv5算法提高了5%,基于嵌入式WEB方式实现的人流量检测数据分析系统,既降低了系统成本,又提高人流量数据的平均正确率,有利于进一步推广。

With the rapid development of computer vision and deep learning,crowd flow data analysis is more and more widely used in many occasions.Data analysis of highly crowded places has high commercial value and also has a preventive effect on safety problems such as trampling.At present,the application equipment of artificial intelligence systems is costly,which is not conducive to proliferation.Therefore,it is proposed to design an embedded data analysis platform based on edge computing for people flow detection,using an embedded web server based on Flask and an improved YOLOv5 algorithm of Raspberry Pi,by adding a new face detection head,replacing the focus layer with a Stem block,and modifying the methods of kernel in the SPP layer to improve the ability of YOLOv5 to identify human features.The experiment shows that the average accuracy rate in the wide-face dataset is improved by 5%.The flow detection data analysis system based on embedded WEB not only reduces the system cost,but also improves the average accuracy of the flow data,which is conducive to further promotion.