Pedestrian Flow Monitoring System for Commercial Areas Based on Deep Learning and QT

This paper investigates the multi-target detection and system implementation problems of pedestrian flow for commercial areas. First, a deep network is trained based on the PaddleDetection framework and the FairMOT model, and it is fine-tuned for the scenarios of small, multi-target, and high-density targets to improve the accuracy of pedestrian flow counting. Then, an approach to estimate pedestrian density by setting predefined values is proposed. In terms of system implementation, it is developed using C++ language and QT framework, which uses OpenCV to capture the video stream and analyze and process the pedestrian flow in the frame images. In addition, the system is designed with some useful features such as historical pedestrian flow graphs and over limit alarms.