Foreground extraction algorithm based on confidence weighted fusion and visual attention

Abstract

The invention discloses a foreground extraction algorithm based on confidence weighted fusion and visual attention. Firstly, confidence coefficients are given to sample color values and texture valuesin a background model; respectively counting the sum of the color-level distance between the current pixel and the sample value and the confidence coefficient of the sample of which the texture distance is less than or equal to the distance threshold of the current frame during classification; and then endowing the sum of the two confidence coefficients with different weights for addition, and when the sum is greater than or equal to a judgment value, taking the current pixel point as a background, otherwise, taking the current pixel point as a foreground. Adaptively updating the confidence coefficient and the weight; secondly, averagely dividing the video sequence into M sub-sequences, taking a foreground detected by a previous frame in the sub-sequences as a region of interest R of static foreground detection, and calculating color saliency and texture similarity of the region R; and circularly detecting whether the R region is a static foreground or not until the last frame of thesub-sequence or the R region is detected as a background. The algorithm disclosed by the invention can effectively overcome the color camouflage problem, and has good robustness for static foregrounddetection.

Description

technical field [0001] The invention belongs to the technical field of computer vision, specifically relates to foreground detection, and can be applied to intelligent security video monitoring in public places such as schools and squares. Background technique [0002] The general steps of the foreground detection based on the background modeling algorithm are to compare the current frame data information with the background model to extract the foreground target, and then update the background model. The difficulty of background modeling lies in how to overcome the problems of color camouflage and sudden stillness of moving targets to extract complete foreground targets. Algorithms proposed so far include pixel-based and region-level methods, as well as background modeling methods based on color information and texture features. These methods have their specific advantages and ensure real-time performance, but most of them cannot overcome color camouflage and target Sudden...

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: Aiming at the problem that the background subtraction method cannot overcome the color camouflage problem to extract the complete foreground and cannot detect the static foreground target, the present invention proposes a foreground extraction algorithm based on confidence weighted fusion and visual attention, through weighted fusion of color and texture Confidence overcomes the problem of color camouflage, and uses the visual attention mechanism and saliency detection to extract static foreground targets, optimizes the foreground detection method, and thus can extract accurate foregrounds with higher robustness

Images