실내 방범 로봇에서 YOLO와 VMD를 활용한 실내 침입자 탐지

초록

Today, security has become an important issue in all areas of life. Recently, the intruder response method is changing to using CCTV or thermal imaging equipment to search for intruders with a small number of people and respond to security companies when they are caught. However, the fixed CCTV has the disadvantage that it can respond to intrusions only in a limited area and there may be blind spots. Therefore, in this paper, we propose an indoor intruder detection system through YOLO and VMD based on an indoor crime prevention robot capable of autonomous driving to solve the shortcomings of the existing fixed CCTV. The autonomous driving of the robot proceeds by sequentially moving the coordinates designated by the manager through the multi-point cruise function. Using the YOLO v3 Tiny model that can be used even in low-spec environments among various YOLO models, if an object corresponding to the Person class is detected, it is determined as the first intruder detection. VMD detects the movement of an object by comparing the three frames, and when a movement is detected, it is finally determined as an intruder. An experiment was conducted to compare the mAP and FPS of YOLO v3 and YOLO v3 Tiny, and it was confirmed that the FPS of YOLO v3 Tiny was 223, far superior to YOLO v3.

키워드