Video crowd counting system and method

Abstract

The invention discloses a video crowd counting system and method. A front-end encoder performs feature extraction on a frame of image in a given video sequence; the common supervision module suppresses interference of a fixed background in a scene by utilizing related characteristics between a previous frame and a to-be-detected frame in a video sequence; the spatial attention module selects an attention area by utilizing shallow structure characteristics, and guides the system to distinguish foreground crowds from background interference; the multi-scale self-supervision module extracts different scale features of the crowd in the image through a multi-branch structure by using deep semantic features, adaptively integrates context feature information through a self-attention mechanism, and autonomously selects the most suitable scale branch feature; and the rear-end decoder is used for fusing the features extracted by the three modules and outputting a predicted crowd density map so as to calculate the number of people contained in the to-be-detected video frame. According to the invention, crowd number estimation is carried out on each contained frame of image according to the video sequence, and dense crowd counting for the video sequence is realized.

Description

technical field [0001] The invention relates to the technical field of counting dense crowds under surveillance video, in particular to a video crowd counting system and method. Background technique [0002] Crowd counting and crowd density estimation are of great interest in computer vision as it plays a pivotal role in applications including physical security, public space management, and architectural design of public places. The dense crowd counting problem aims to estimate the number of people in a specified region of interest (ROI) in an image. In recent years, many classes of methods have been applied to the dense crowd counting problem. Initially, researchers proposed detection-based methods, but detectors performed poorly on images with severe scenes, scale shifts, and severe occlusions. Therefore, methods based on regression tasks have become the mainstream of research. The goal of the regression task is to learn a function to represent the global population cha...

AI Technical Summary

Problems solved by technology

[0006] To sum up, the existing dense crowd counting methods can perform good counting predictions for some targets whose crowd size does not change much or the occlusion is not very serious. No occlusion or less occlusion only accounts for a small part. For small-sized crowds that are far away or large-sized targets that are close to the lens, object occlusion, self-occlusion targets, and isolated clusters of people that occasionally exist in the picture, count The result is not very good, and there is a huge problem of count jitter from frame to frame

Images