Personnelde-duplication identification method, storage medium and computer equipment

Abstract

The invention discloses a personnel de-duplication identification method, a storage medium and computer equipment, and relates to the technical field of image processing. Personnel distinguishing de-duplication identification can be effectively carried out on persons in a list or strangers not in the list by maintaining a cache queue T, the face frame position overlapping degree is matched with the similarity double threshold value of the face algorithm, on one hand, the problem that people are lost due to video decoding frame dropping or the low detection rate of the face algorithm can be solved, and extra wrong passing records generated when people in the list pass are effectively reduced; on the other hand, the double threshold values of the similarity of the face algorithm are adopted, so that the static target and the non-static target are effectively de-duplicated; and no matter the position overlapping degree of the face frame or the similarity calculation of the face algorithm, the calculation process has low requirement on the calculation power, and the method can be smoothly operated on most equipment.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for deduplication and identification of persons, a storage medium and computer equipment. Background technique [0002] With the maturity and development of face recognition technology, face recognition is more and more widely used in intelligent access control systems, and many corresponding face recognition access control systems have appeared on the market. However, when people walk in front of the camera, they often capture more than one or two face photos, so the face recognition smart access control system will generate many face image records that are not much different, but belong to the same person. To this end, it is necessary to use a deduplication algorithm to minimize such records. At present, the main deduplication algorithms for face images mainly include deduplication methods based on traditional image features (such as color histogram, LBP hist...

AI Technical Summary

Problems solved by technology

However, due to the large difference in light in different video surveillance scenes, the same threshold cannot be used to achieve compatibility with all application scenarios, and it is not universal; the fast deduplication method based on motion matching needs to set the pixel difference threshold and difference ratio. It mainly removes static repeated targets, and cannot handle deduplication of non-static targets; based on tracking deduplication algorithm, three parameters need to be set: tracking frequency, deduplication frequency and retention frequency, deduplication effect depends on detection algorithm and tracking algorithm

If the detection frequency is low or the detection rate of the detection algorithm is poor, it will lead to people being lost; and because of the minimum detection frequency requirements, the machine performance requirements are high

[0004] At the same time, the above deduplication algorithms do not take into account the classification of personnel in the actual access control application (that is, the personnel in the list or strangers), and there is no special targeted treatment; especially, when the personnel in the list just appear When in the camera screen, the captured face may not reach the recognition similarity threshold, and it will be judged as a stranger passing record; when walking to the middle area of ​​video surveillance, the captured face may reach the recognition similarity threshold, and it will be It is also judged as a pass record of people on the list; when the person on the list is about to leave the camera screen, the captured face may not reach the recognition similarity threshold, and it will be judged as a pass record of strangers; and so on. When using a camera, there are two different classifications of traffic records; these are additional wrong traffic records generated without considering the classification of people in the application

Images