Method for accurately obtaining two-way pedestrian flow through image algorithm

Abstract

The invention relates to a method for accurately obtaining two-way pedestrian flow through an image algorithm, and the method comprises the following steps: 1, modifying a YoloV5 network, and adding four input channels and six output channels; 2, collecting and marking training and testing data sets, and training the network model; 3, deploying the trained network model to an FPGA chip; 4, reading a real-time video through a camera, inputting the real-time video into a network model, reasoning to obtain the detection frames and predicted values of the next frame position of each detection frame, and obtaining personnel IDs corresponding to the detection frames; 5, generating a corresponding personnel movement track according to the personnel ID and the center movement coordinate queue of the detection frame; and 6, according to the generated personnel movement track, determining the personnel movement direction, and carrying out statistics on the two-way people flow. According to the method, the accuracy of people flow detection can be improved, two-way people flow statistics is realized, and the method has low requirements on chip computing power and can be deployed on edge computing equipment for use.

Description

technical field [0001] The invention relates to image recognition technology, in particular to a method for accurately obtaining two-way traffic flow through an image algorithm. Background technique [0002] Obtaining information on the flow of people at the entrances and exits of shopping malls and parks is of great value to managers. There are many existing methods for counting people flow, but the detection accuracy is generally not high, and most of them can only be in one direction. For detection, it is impossible to deal with the detection of two-way flow of people at the same time, and the technology that can detect two-way flow of people at the same time is generally very costly to implement. For example, the Chinese invention patent application with application number: 201510066231.7 discloses a visual analysis of flow of people based on laser detection lines Statistical method and equipment, the method uses a network camera to collect video of people entering and e...

AI Technical Summary

Problems solved by technology

In practical applications, the computing power requirements of pedestrian re-identification trackers increase linearly with the number of people, and it is difficult to achieve real-time processing effects when the flow of people is large, which indirectly increases the requirements for hardware and increases production costs.

Images