Crowd density estimation system and method under multi-camera condition

Abstract

The invention relates to a crowd density estimation system and method under a multi-camera condition. The system comprises a multi-camera wide-view-field scene image acquisition module, a saliency region extraction module based on saliency detection, a face feature detection module for a saliency region and a saliency region crowd density estimation module based on a long-time and short-time deeplearning neural network. The system is mainly used for carding and analyzing macroscopic features such as crowd density features of scene crowds and estimating density features of dense crowds. A usercan analyze and obtain a key area where a crowd is located according to a wide-view-field image obtained under the condition of multiple cameras, and feature information such as crowd density is obtained through analysis from the key area.

Description

technical field [0001] The present invention relates to a crowd density estimation system and method under multi-camera conditions, specifically a crowd density estimation and early warning system and method in a wide field of view panoramic monitoring mode for special alarm warning, which belongs to monitoring and early warning security field. Background technique [0002] At present, the regional scene data information collected by panoramic vision is mainly used to extract visual features to complete the tasks of target recognition, fast tracking, and autonomous positioning. However, the method of using panoramic images for face recognition and crowd density monitoring has not been put into practical use, and the advantages of panoramic images with wide field of view and large viewing angle have not been well utilized. [0003] For feature extraction in panoramic images, it can be mainly divided into two categories, global visual features and local visual features. The ...

AI Technical Summary

Problems solved by technology

[0005] In addition, most of the current applications of computer vision use various operators to extract visual feature information in images. Although this process can effectively describe the characteristics of image data, it does not fully comply with human perception of the world. The way of understanding, how to effectively use the existing research results in the field of target recognition, combined with the target recognition based on omnidirectional vision and even the understanding and analysis of the semantic level of omnidirectional vision images on the human-like level, this is the current use of omnidirectional vision One of the development directions to complete the autonomous monitoring of the target scene problem visually

Images