Research on risk analysis of pedestrian and vehicle flow in community scenes based on machine vision

Surveillance video is vital to community emergency management and risk control. Existing surveillance video analysis has limitations of low utilization, low sharing rate and low data correlation analysis rate. Based on these shortcomings, this paper reports a framework of community risk identification and warning using machine vision to analyze surveillance video. The prototype is based on the YOLOV4 video analysis network, which extracts and records spatio-temporal coordinate data of key categories of people, vehicles and objects related to community risks. The data are computed to obtain key state information about the direction of flow, speed and range of activity. By analyzing the logical, inclusive and spatio-temporal relationships among objects, the process makes risk judgment in real-time and provides early warning of visualization results. In addition, by accumulating historical data to discover potential dangers, the level of community services are optimized and the construction of smart communities are accelerated. This paper introduces a community risk management and prevention framework based on machine vision, which focuses on networking, sharing and exchanging of video resources as well as multi-dimensional association of video big data. Through real-time surveillance and early warning of risks, analysis and research, the study aims at the visualization, monitorability and early warning of smart community operation.