Multi-camera identification personnel behavior track analysis method

Abstract

The invention discloses a multi-camera identification personnel behavior track analysis method, which is suitable for a monitoring area where multiple cameras are installed, and the number of the distributed cameras is enough to cover the monitored area. The method comprises the following steps: after a plurality of cameras acquire video images of pedestrians from different angles, extracting feature points of each captured frame of image by using a Gaussian noise elimination method and SIFT features; sharing and sending the extracted feature points to all monitoring systems connected with a central server; analyzing and finding out the optimal feature point, finding out a target person subjected to mobile tracking through clustering, generating position information of the person, and sending the position information to a central server in a coordinate form; and enabling the central server to generate a moving track of the target person according to the current coordinates. According to the invention, the efficiency and utilization rate of feature extraction are greatly improved, and the real-time performance and collaboration of recognition and monitoring are improved, so that anaccurate personnel behavior track is formed. According to the method, deep learning and feature recognition are applied to the monitoring system, and the method has very high practical value.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and a device for identifying human behavior trajectories with multiple cameras. Background technique [0002] Recognizing object tracking plays an important role in high-level analytics such as activity analysis, robotics, video surveillance, etc., and in many real-time applications. In the current trend, different tracking methods have been developed, such as data association strategy methods, and methods based on correlation filtering. Data association strategy approaches are highly focused on most traditional detection-based tracking work in visual object tracking, using an object detector to extract a detection set from each frame. Facing complex challenges, such as interactions between objects, unknown number of objects, frequent object occlusions, and complex motions, existing state-of-the-art algorithms yield good results. Nevertheless, it remains an op...

AI Technical Summary

Problems solved by technology

There are some problems. In an unconstrained environment, noisy and uncertain measurements are the main challenges for target tracking solutions. For example, the classifier has a large impact on receiving samples, and changing lighting conditions lead to false positive detections and partial occlusions. detection loss

Images