基于图像处理的二阶段人流检测系统研究

Research of Two-stage Human Flow Detection System Based on Image Processing

提出了一种基于传统机器学习和深度学习计算的人员计数方法。所提出的方法主要使用在特定公共场所,例如:公共汽车候车亭或广场等,用来分辨人流高峰或离峰的估算信息,其优势在于有效节省管理成本与智能运输应用的升级。本系统基于边缘计算的架构设计,分成前台系统和后台系统,包含下述的两种方法。第一阶段主要进行人流量的概数计算并简单分级,后台仅处理第二阶段的行人识别,在最后的实验结果可以发现,本系统能有效地降低计算复杂度与整体的时间花费,其中,前台人流分级的正确率为94.28%,后台人数计数正确率为94.04%。

In this paper,a personnel counting method based on traditional machine learning and deep learning computation is proposed.The proposed method is mainly used in specific public places,such as bus shelters or connecting squares to distinguish the estimation information of peak flow or off peak flow.Its advantages are that it can effectively save management costs and upgrade intelligent transportation applications.The system architecture design based on edge computing is divided into foreground system and background system,including the following two methods.The first stage is mainly used to calculate the approximate number of people flow and simply grade it.The background only processes the pedestrian recognition in the second stage.The final experimental results show that the system can effectively reduce the computational complexity and overall time cost.Among them,the correct rate of people flow classification in the foreground is 94.28%,and the correct rate of people counting in the background is 94.04%.