Cross-scene video crowd understanding two-channel network method based on aerial view angle

Abstract

The invention discloses a cross-scene video crowd understanding two-channel network method based on an air view angle, which belongs to the technical field of image recognition counting, analyzes the relationship between people and background and determines a crowd number under bird's-eye view crowds by establishing a two-channel crowd counting network model and fully utilizing inter-frame information and input through RGB images of two-path streams. According to the method, the receiving range of the convolution kernel is expanded by the hollow convolution layer, so that the spatial context information in the element is enriched, the relationship between the person and the background is better analyzed, the optical flow and frame difference information is extracted by the double-flow crowd counting model to serve as additional branches, the generalization ability of the model in different scales and time is improved, background noise generated in cross-scene testing is reduced, the number of crowds is automatically known from visual data collected by the unmanned aerial vehicle, negative effects in the shooting process of the unmanned aerial vehicle are reduced, and the screening and observation capability of the number of crowds is improved through optimization of a network structure.

Description

technical field [0001] The invention belongs to the technical field of image recognition and counting, and in particular relates to a dual-channel network method for cross-scene video crowd understanding based on an aerial perspective. Background technique [0002] Drones can be used for dynamic traffic monitoring, object detection and tracking, and other vision tasks, but the variability of shooting locations brings some tricky challenges to these functions, such as variable scale, unstable exposure and scene shift. [0003] With the development of crowd counting, it is found that among the influencing factors (occlusion, scale and perspective, etc.), scale variation has the largest negative impact, and many crowd counting algorithms have focused on scale variability in recent years. [0004] Cross-scene and cross-domain crowd counting has attracted the attention of researchers in recent years due to the complex data annotation required for crowd datasets. In this task, mod...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to propose a dual-channel network method for cross-scenario video crowd understanding based on aerial perspective in order to solve the problem that data labeling performance affects cross-scene and cross-domain crowd counting

Images