A Target Tracking Method Based on Improved Joint Model of Decision and Generation

Abstract

The invention discloses a target tracking method based on an improved joint model of judgment and generation. Firstly, relevant parameters are set; dictionaries D1 and D2 of the initial GM model are generated, the histogram ψ of the initial target is obtained, and the template of the initial DM model is obtained. T; then the affine transformation generates a candidate target set; secondly, the candidate target is processed by the GM model; processed by the DM model classifier; finally, the template is updated, including the update of the target histogram ψ, dictionary D2 and template T; the update of the dictionary D2 , which is updated in the way of incremental subspace. Compared with the prior art, the tracking precision of the present invention is higher.

Description

technical field [0001] The invention relates to an improved target tracking method, which can be used in the fields of safety monitoring, traffic monitoring, human-computer interaction and military affairs, and belongs to the field of computer vision technology research. Background technique [0002] With the rapid development of science and technology, more and more shops, banks, restaurants, schools, stations, airports and other public places, military-related security areas, and even some private residences have installed video surveillance systems. However, the monitoring system in most of the application scenarios can only simply record and save the video. Observation and analysis of subsequent video streams require follow-up manpower to perform. However, this monitoring method has obvious shortcomings, such as high labor costs, timely warning of emergencies, and high probability of missing abnormalities through manual monitoring. For this reason, the research of inte...

AI Technical Summary

Problems solved by technology

[0071] a) In the above SCM algorithm generation model, when the template is updated, only the histogram of the tracking target is updated, but the dictionary D used to calculate the orthogram is not updated, which leads to subsequent tracking errors. It is proposed to update based on the incremental subspace dictionary D, thus improving tracking accuracy

[0072] b) In the generation model of the above SCM algorithm, the position information of the small block in the candidate target is not considered, which leads to the problem of tracking drift in the follow-up tracking. It is proposed to add a module based on the small block based on the existing generation model Local sparse model of location information to improve tracking accuracy

Images