A Multi-factor Suspicious Person Identification Method Based on Video Feature Learning

Abstract

The invention discloses a multi-element suspicious person identification method based on video feature learning, which includes: screening surveillance video based on perceptual hash mapping key frames, separating foreground information including behavior subjects, and locating the area where key parts are located; The processing results are combined with static and dynamic feature extraction: for static surveillance videos, local features such as head posture and hand posture are extracted, as well as overall features such as abnormal walking and abnormal clothing; The probability of dense crowds and the average stay time of landmarks in the monitoring environment; according to the evaluation criteria combined with the idea of ​​trusted computing, the suspicious personnel identity credibility index is calculated; finally, according to the corresponding threshold, suspicious personnel are dynamically screened and discriminant information is output. The invention can identify suspicious persons more accurately and efficiently under controlled and uncontrolled environments, and has good scientificity and higher practical application value.

Description

technical field [0001] The invention belongs to the field of information technology, in particular to a multi-element suspicious person identification method based on video feature learning. Background technique [0002] With the improvement of monitoring technology, the monitoring video becomes clearer, which provides a good hardware foundation for accurately identifying suspicious persons with criminal motives in a short period of time through video monitoring. At the same time, it also makes suspicious identification change from a controlled environment to an uncontrolled environment, and the identification effect is also uncontrollable. Generally speaking, the reasons for the unsatisfactory recognition effect are roughly divided into the following two aspects: [0003] First of all, the environment used for identity recognition is no longer controlled, and there are dynamic transformation problems such as occlusion and posture when collecting the information of the subj...

AI Technical Summary

Problems solved by technology

[0003] First of all, the environment used for identity recognition is no longer controlled, and there are dynamic transformation problems such as occlusion and posture when collecting the information of the subject. In addition, the identity information of the subject does not necessarily exist in the identity database, which may lead to extraction The feature information cannot meet the task requirements of suspicious person identification;

[0004] Secondly, the amount of surveillance video information is huge, and the processing time is long, which seriously affects the real-time performance of the system.

Images