Dynamic Object Detection Method Based on Video Stream Codebook Model

Abstract

The present invention relates to the technical field of moving target detection in video processing, in particular to a dynamic target detection method based on a codebook model of a video stream; the technical problem to be solved is: to provide a dynamic target detection method based on a codebook model of a video stream Method, this method makes full use of the correlation between the pixels of the coding block, avoiding time-consuming operations such as reconstruction and loop filtering in the decoding process; the technical solution adopted to solve this technical problem is: the codebook model is based on the pixel coding block Establishment, by analyzing the information such as the macroblock type, prediction mode, quantization parameter, motion vector residual and pixel residual of the coded block of the video stream, two types of codewords, intra-frame prediction and inter-frame prediction, are respectively established, and through analysis Whether the coded block of the video code stream matches the background code word is used for dynamic target detection; the method of the invention is applied to the dynamic target detection in the field of video processing.

Description

technical field [0001] The invention relates to the technical field of moving target detection in video processing, in particular to a dynamic target detection method based on a video code stream codebook model. Background technique [0002] Dynamic target detection has always been a hot issue in the field of computer vision research, and has a wide range of applications in real life. Intelligent transportation, intelligent video surveillance, object tracking, etc. all require the effect of dynamic target detection as a guarantee; usually dynamic target detection The purpose is to detect the changing area of ​​the target in the sequence image and extract the dynamic target from the background image; traditional dynamic target detection algorithms include frame difference method, optical flow method and background subtraction method, among which background subtraction method is widely used The key to the background subtraction method lies in the establishment of the backgroun...

AI Technical Summary

Problems solved by technology

[0003] The algorithm of the traditional Codebook model models a single pixel in the RGB space, and judges whether the pixel belongs to the foreground or the background according to the two information of color and brightness. The essence is to split a complete image into many isolated small pixels. Points are used to judge, and after a certain algorithm is judged, the pixels are recombined to form a detection result screen; this method has imperfections: (1) Since the screen is dispersed into discrete pixels for judgment, a pixel is ignored The connection between them; the dynamic target in the picture generally occupies the pixel of a certain area, so it often happens that a single pixel is detected as a dynamic target, but none of the surrounding eight pixels detects a dynamic target; (2) The codebook algorithm uses the RGB color space to determine the color distance, and at the same time adds the brightness determination condition, which causes the detection sensitivity to be different between high brightness and low brightness; (3) when the light changes, especially when there is a sudden change , matching failures will often occur, resulting in large-scale false detections, insufficient adaptive ability to light changes, and low detection accuracy; (4) Complicated calculations, high time overhead, and difficult to implement in real time

[0004] Surveillance video data is data compressed by video coding algorithms, rather than single pixel data; currently, the applicable international standards for video compression mainly include H.264 and H.265, and the domestic standard is AVS, which all use block-based hybrid Coding algorithm; the traditional dynamic target detection algorithm based on the Codebook model processes each pixel in the image, and needs to fully decode the video data before processing each pixel. The calculation of the massive received video data is large and complex , it is difficult to achieve real-time processing of dynamic targets

Images