Trans-camera target tracking method and trans-camera target tracking device on basis of dynamic path trees

Abstract

The invention provides a trans-camera target tracking method and a trans-camera target tracking device on the basis of dynamic path trees. The trans-camera target tracking method includes carrying out single-camera target tracking on specified targets in the visual angle range of each optional camera in each preliminarily constructed camera network until the specified targets leave the visual angle range of the current camera; acquiring currently required-to-be-searched camera lists according to target search region switching algorithms; detecting objects for video data of all cameras in the camera lists and screening suspicious objects by the aid of visual similarity matching algorithms; processing the suspicious objects according to dynamic path tree models and inferring movement tracks of the specified targets. The similarity of the visual appearance of the suspicious objects and the visual appearance of the specified targets is higher than preset thresholds. The trans-camera target tracking method and the trans-camera target tracking device have the advantages that accurate trans-camera target tracking can be carried out under the condition of the possibly minimum computational resource expenditure, and paths can be accurately predicted.

Description

technical field [0001] The invention relates to the technical field of video analysis, in particular to a method and device for cross-camera target tracking based on a dynamic path tree. Background technique [0002] At present, most of the traditional target tracking technologies are still aimed at the problem of single-camera target tracking. However, with the increasing scale of various intelligent monitoring systems in recent years, the limitations of single-camera target tracking in practical applications have become more and more obvious. The demand for cross-camera object tracking technology is increasing. [0003] Compared with single-camera target tracking, cross-camera target tracking is more challenging, which is reflected in the following three aspects: First, the visual appearance of the same target under different cameras is different due to factors such as shooting angle, shooting distance, and light conditions. There is a large difference, so the difficulty ...

AI Technical Summary

Problems solved by technology

[0002] At present, most of the traditional target tracking technologies are still aimed at the problem of single-camera target tracking. However, with the increasing scale of various intelligent monitoring systems in recent years, the limitations of single-camera target tracking in practical applications have become more and more obvious. The demand for cross-camera object tracking technology is increasing

[0003] Compared with single-camera target tracking, cross-camera target tracking is more challenging, which is reflected in the following three aspects: First, the visual appearance of the same target under different cameras is different due to factors such as shooting angle, shooting distance, and light conditions. There is a large difference, so the difficulty of cross-camera target tracking is much greater than that of single-camera target tracking; second, because the target leaves the perspective of one camera, it is actually difficult to find a similar target in other cameras based on the level of similarity in appearance Just determine whether it is really the original tracking target, so it is very likely that the target will be lost during the tracking process; Third, with the increasing scale of today’s surveillance camera network, the resolution of the images captured by the camera is getting higher and higher. The processing and analysis of these video data consumes more and more computing resources, and the simple and rough method of traversing all of them is usually not feasible in actual scenarios.

[0004] Cross-camera target tracking based on vision technology mainly includes three core algorithms: object detection, single-camera target tracking and cross-camera target similarity matching. Cross-camera target tracking problem, so we also need to make special optimizations for the two problems of target trajectory estimation and excessive computing resource overhead

Images