Real-time intrusion detection and tracking method for multi-channel camera

Abstract

The invention provides a real-time intrusion detection and tracking method for a multi-channel camera, which comprises the following steps: firstly, acquiring a plurality of video streams through multi-channel image acquisition equipment, then, acquiring a first frame image of each video stream, sequentially selecting a specified forbidden area by a user, and splicing multi-channel video stream image frames into a large image frame; then, using YOLOv4 to detect personnel in the multiple paths of image frames, judging whether the personnel in the forbidden area invade illegally or not; if the illegal invasion is found, giving an alarm for an illegal invasion behavior, and carrying out cross-camera continuous tracking on the personnel. According to the invention, through splicing multiple paths of camera pictures, the balance between time cost and computing power cost in a multi-path picture tracking task is achieved; meanwhile, multiple camera images are monitored and tracked in the same algorithm, the cross-camera tracking function can be efficiently achieved, high pertinence is achieved for forbidden zone intrusion detection and invader tracking tasks, and the accuracy is high.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a multi-target detection and tracking method applicable to multi-channel cameras. Background technique [0002] Intrusion detection and tracking is an intelligent automatic monitoring technology based on traditional security cameras, which can effectively solve the problem that manpower cannot always observe the monitoring screen and discover intrusion behaviors in time; and it is difficult to follow up after an intruder is found, greatly saving bans The manpower and material resources required for monitoring in the area. [0003] Traditional intrusion detection methods generally adopt the following two schemes when dealing with multiple cameras: 1. Parallel scheme, starting several algorithm models for several cameras. The advantage is that the algorithm accuracy of each camera is high. This model has higher requirements on the computing power of the computing ...

AI Technical Summary

Problems solved by technology

[0007] At the same time, the present invention uses a neural network to model the apparent characteristics of the detection target, which can not only cooperate with the database to realize the identification of the identity of the intruder, but also measure the similarity of the apparent characteristics of the target that failed to be tracked by the Kalman filter during tracking. In this way, the problem of re-recognition of occluded targets and cross-camera targets can be solved, and the actual needs can be solved

Images