Video abnormity detection method

Abstract

The invention relates to a video abnormity detection method. Firstly, overall motion estimation is completed by using a gray projection algorithm according to the characteristic of local consistency of video jitter, in combination with a blocking thought; then a frame sequence which can represent the video jitter is selected to serve as an effective frame according to the longest path algorithm, jitter parameters such as the jitter rate, jitter frequency and jitter amplitude of the video jitter are computed according to an overall motion parameter of the effective frame, and an influence factor is set for the jitter parameters; and at last a video jitter degree is computed via a weighted average method. According to the method provided by the invention, limitation of existing video jitter detection technologies is overcome, image jitter detection and jitter degree estimation can be effectively achieved, and meanwhile the method is well applicable to the videos with local jitter.

Description

technical field [0001] The invention belongs to the field of computer vision and is used for a system for detecting abnormality of monitoring video shaking. In particular, it relates to a video anomaly detection method. Background technique [0002] Video jitter is manifested as the discontinuity of the correlation between adjacent frames. There are roughly two types of causes: 1. When the installation position of the surveillance camera (usually a pole type) is impacted by a short external force (such as strong wind or ground vibration), the surveillance video The picture may appear that does not conform to human visual habits, and the imaging area as a whole may appear frequent detours in random directions - which can be regarded as video jitter; It will cause another video jitter phenomenon. In this case, the jitter is mostly manifested as irregular and fast jumping, flickering, deformation, and color distortion and overlapping of the picture. Video jitter will cause th...

AI Technical Summary

Problems solved by technology

Commonly used motion parameter estimation methods include block matching, optical flow method, feature point matching, grayscale projection, etc. Among them, block matching and optical flow method have high computational overhead, low efficiency, and cannot meet real-time requirements; feature point matching requires feature points to have Representative, it is difficult to be extended to video surveillance systems; the grayscale projection algorithm has a small amount of calculation, fast speed, and meets real-time requirements, but its accuracy is low

[0004] The jitter detection algorithm has high requirements for various comprehensive indicators of business applications, and the above methods cannot meet practical applications

Images