Multi-camera combined large-scene crowd counting method

Abstract

The invention discloses a multi-camera combined large-scene crowd counting method, and the method comprises the following steps: S1, obtaining a base map and a monitoring image of an observation scene, obtaining a homography matrix corresponding to the base map and the monitoring image, mapping the monitoring image to the base map, and constructing a scene model; s2, preprocessing the monitoring image, obtaining information of all human head point bounding boxes in the monitoring image through processing of a crowd counting neural network, and mapping the human head point bounding boxes to a base map; and S3, according to a non-maximum suppression algorithm, screening the human head point bounding boxes in the base map, deleting the repeated human head point bounding boxes, and calculating the total number of the human head point bounding boxes to serve as the total number of people in the observation scene. The method provided by the invention has good adaptability to the crowd counting application situation of multiple cameras in a large scene, and can obtain the accurate crowd number in the scene in real time.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a multi-camera joint large-scene crowd counting method. Background technique [0002] With the increasing number of surveillance cameras in modern society, the spatial relationship between cameras has become more and more complex. It is a big challenge for surveillance personnel to accurately count the number of pedestrians in the entire scene in complex and overlapping surveillance video data. challenge. At present, the crowd gathering behavior in the entire monitoring scene still lacks effective management and control, and the slow flow of people due to crowd congestion may cause serious group incidents. On the other hand, due to the increase in the number of surveillance cameras and the expansion of coverage, the traditional manual inspection and analysis of surveillance content requires a lot of human resources, which is time-consuming, laborious, ineffic...

AI Technical Summary

Problems solved by technology

[0002] With the increasing number of surveillance cameras in modern society, the spatial relationship between cameras has become more and more complex. It is a big challenge for surveillance personnel to accurately count the number of pedestrians in the entire scene in complex and overlapping surveillance video data. challenge

At present, the crowd gathering behavior in the entire monitoring scene still lacks effective management and control, and the slow flow of people due to crowd congestion may cause serious group incidents

On the other hand, due to the increase in the number of surveillance cameras and the expansion of coverage, the traditional manual inspection and analysis of surveillance content requires a lot of human resources, which is time-consuming, laborious, inefficient, and cannot be analyzed in real time.

[0003] At the same time, due to the complex distribution of surveillance cameras in the real scene, there are overlapping phenomena in the surveillance area. It is difficult to obtain the accurate number of pedestrians in the entire scene by simply importing images from a single surveillance camera and simply counting the number of people in each camera.

Therefore, the existing methods are difficult to meet the needs of large scenes with multiple cameras, and the tasks related to crowd counting under large scenes with multiple cameras still face great challenges.

Images