Pedestrian detection and tracking method and device

Abstract

A pedestrian detection and tracking method and device are disclosed. The method includes de-noising a video frame of a video sequence, and determining a pre-tracked target box according to the video sequence, detecting the target box and acquiring the detected target, correcting the position of the tracking target according to the detected target, inputting the characteristic sample of the tracking target to the particle filter system, tracking the tracking target by the particle filter system, outputting the tracking result of the tracking target, judging whether the current video frame is the last frame, if so, showing that the tracking is successful. The device comprises a tracking target acquisition module, a detection target acquisition module, a particle filter module and a trackingoutput module, which are used for realizing the pedestrian detection and tracking method. On the basis of satisfying the real-time requirement, the invention can greatly improve the accuracy rate of detection and tracking of complex scenes.

Description

【Technical field】 [0001] The present invention relates to the technical field of computer vision, in particular to a pedestrian detection and tracking method based on pedestrian motion vision and a device applied to the method. 【Background technique】 [0002] Currently, target detection methods for fixed cameras, that is, static backgrounds, can be roughly divided into background model-based detection and classifier-based detection. The classic methods of background model detection include background difference method, inter-frame difference method, optical flow method, image multi-feature fusion method, etc.; classifier-based detection can be realized by a classifier. Among them, the current mainstream classification methods are Neural Networks, Boosting, Decision Trees, Support Vector Machines (SVM), etc. [0003] Target tracking solves the problem of how to quickly and efficiently select a target among candidate targets. Search is generally divided into two categories: ...

AI Technical Summary

Problems solved by technology

Tracking algorithms are complex and diverse, and different algorithms complement each other and improve the target tracking performance in the scene.

[0005] However, there are several problems in the above tracking algorithms. Problem 1: There is no general algorithm suitable for various scenarios

Question 2: In real life, the monitored scenes are more complex, such as nonlinear motion, local or global occlusion, changes in illumination, changes in weather, a lot of clutter in the foreground and background, deformation of non-rigid targets, target scale changes, etc. etc. These factors will seriously affect the effect of target detection and tracking

Question 3: The relevant technologies in densely populated scenes are not mature enough, and there are few related achievements. The domestic video analysis technology for densely populated crowds is still in its infancy, and there is still a lot of room for development and research value

Target tracking technology in dense scenes has high practicability, but there are also severe challenges

Images