Video multi-target fast tracking method based on joint probability data association

Abstract

Aiming at the problems of video multi-target data association and tracking with a large monitoring range and few target appearance features, the invention provides a video multi-target fast tracking method based on a complex condition of improved joint probability data association (JPDA) by utilizing target movement features, that is, the video multi-target fast tracking method is based on the joint probability data association. The invention comprises the steps of calculating the optimal k joint events of JPDA by adopting simplified Murthy algorithm to greatly reduce the computational complexity; discussing movement conditions of targets according to association probability of JPDA, analyzing the current frame measurement and the data association problem of tracking targets in the event of complicated cases such as new appearance, shielding, disappearance, separation and the like of multi targets, and acquires the multi-target tracking trajectory of the complex movements. The method provided by the invention can achieve the video multi-target fast tracking under large monitoring range and greatly improve the tracking performance.

Description

technical field [0001] The invention belongs to the technical field of video multi-target tracking, and relates to a video multi-target fast tracking method based on joint probability data association. Background technique [0002] In the video surveillance of multiple moving targets, it is necessary to carry out data association according to the detection results to match and identify different moving targets between consecutive frames, so as to realize multi-target tracking. In video surveillance, it is often necessary to monitor a wide range of scenes, and there are often cases where moving objects have similar appearance characteristics or the target area is small. At this time, data association can only be completed by relying on the motion characteristics of the target, while traditional data association methods are applied to There are still many problems in video multi-target tracking. [0003] At present, the main methods to solve the data association problem of mu...

AI Technical Summary

Problems solved by technology

[0003] At present, the main methods to solve the data association problem of multi-moving targets are as follows: 1) According to the nearest neighbor method, calculate the measurements that fall within the tracking threshold, determine the movement of the target, and directly estimate and update the current movement state according to the effective measurement , this method has a small amount of calculation, but when the number of targets is large and the movement is complex, the anti-interference ability is poor, and it is easy to generate false associations; 2) Multi-target data association methods such as Joint Probability Data Association (JPDA) , JPDA), multiple hypothesis tracking (Multiple Hypothesis Tracking, MHT), currently mostly used in infrared, radar small target or maneuvering target tracking systems, although similar to video surveillance tracking, but must meet the constraints of one-to-one association, while in In the video monitoring and tracking system, multi-moving targets often appear, disappear, occlude, separate and other complex motion situations, that is, there are one-to-many or many-to-one associations. Therefore, it is necessary to further study how these classic multi-object data association methods Applied to video multi-target tracking; 3) Using optimization algorithms to analyze the optimal association between the detection area of ​​the current frame and the tracking target, such as graph optimization, weighted bipartite graph and other methods, but such methods usually need to obtain the appearance of occlusion or separation areas When the target area is small and the appearance information is small, it is difficult to obtain the target appearance features, which makes the association accuracy low.

Images