Deep network-based multi-strategy global crowd analysis method

Abstract

The invention provides a deep network-based multi-strategy global crowd analysis method. The method comprises the following steps of: firstly, monitor area modeling: drawing a global map schematic diagram, establishing a layer for a direction and a range corresponding to a global map in a camera monitor area, and waiting for import of crowd density data; secondly, for a monitor scene of each camera, obtaining a space visual angle mapping graph of a displayed monitor image through perspective transformation, namely, overlook visual angle mapping from side-looking visual angle of the camera to the ground; obtaining image features through a VGG16 transfer learning method, mapping pre-blocks input into the images to a feature layer through strides, carrying out SWITCH judgement on the image features of each block, and selecting to carry out density estimation or pedestrian detection operation on the image through an R1 density estimation network of an R2 pedestrian detection network; and integrating the pedestrian detection or density estimation result of each block to form a density map, and mapping the estimated density map onto the layer through perspective transformation so as to conveniently carry out accurate supervision on the global crowd condition.

Description

technical field [0001] The invention relates to a crowd counting and density estimation method, in particular to a deep network-based multi-strategy global crowd analysis method, which belongs to the technical field of machine vision and artificial intelligence. Background technique [0002] With exponential population growth and deepening urbanization, the number and frequency of large-scale gatherings have increased dramatically, such as scenic spot tourism on statutory holidays, sports games, political rallies, public exhibitions, etc. For better management, to ensure environmental safety and personal safety, it is necessary to analyze the crowd, and pedestrian detection and crowd counting are the current research focus. The current detection and counting methods mainly include: [0003] 1. Methods based on individual statistics [0004] Through the top view of the camera, the human head is detected to effectively resist occlusion; the human body is detected through the...

AI Technical Summary

Problems solved by technology

[0010] The technical problem to be solved by the present invention is to provide a deep network-based multi-strategy global crowd analysis method to overcome the background interference of complex scenes and pedestrian occlusion problems, and then realize the crowd density analysis in the scene. an accurate estimate of

Images