Video target detection avoidance system and method

Abstract

The invention discloses a video target detection avoidance system and method. The system comprises a model adaptability grey box training module, a patch distance self-adaption module, a multinomial loss function calculation module and a digital world patch fitting module. The adaptability training module is used for detecting targets pasted with evasion patches, extracting human body detection number and confidence, and getting rid of limitation on actual detection model types and parameters; the patch distance self-adaption module is used for carrying out distance self-adaption updating on the patches based on a threshold value specified by a user or preset by a system, so as to ensure the protection performance of the patches under different distances; the multinomial loss function calculation module and the digital world patch fitting module are used for realizing clothes wrinkle simulation, physical world color transformation, picture training loss constraint and the like of the patches in the digital world to ensure the robustness of transferring the patches to the physical world. The method does not aim at a specific model, can be effective for different models, has good physical world robustness, and meets the privacy protection requirement of a user side.

Description

technical field [0001] The invention belongs to the technical field of anti-sample protection target detection privacy leakage technology in computer vision, relates to a video target detection avoidance system and method, in particular to a general-purpose human body target detection privacy protection system and method. Background technique [0002] Behavior tracking, intelligent monitoring and many other fields are developing rapidly. Target detection and recognition, as its core technology, not only provides people with many conveniences, but also brings huge security challenges and risks to personal privacy. Users often use disguise methods such as wearing glasses, hats, and masks to avoid personal privacy leakage, but these methods cause inconvenience to users when traveling, and cannot completely avoid video target detection technically. [0003] The main reason why target detection and recognition technology poses a great risk to personal privacy security is that the...

AI Technical Summary

Problems solved by technology

However, most of the existing adversarial samples are generated based on a single model of a white box, and the actual target detection model is complex, and the adversarial samples are difficult to meet the actual privacy protection needs of users, and there are weaknesses such as not easy to carry, easy to be detected, and long-distance attack failure; how to improve The ability of samples to interfere with various target detection models, to ensure the validity of samples in the full range, and to improve the naturalness and portability of samples are the challenges that need to be solved urgently in the current video target detection jamming technology based on adversarial samples

Images