Foreground detection method based on motion vector division

Abstract

The present invention provides a foreground detection method based on motion vector division, characterized in that discrete motion vectors are used for dividing a movement trend to perform foreground detection, when the movement trend of a foreground and the movement trend of a background are distinguished, dense optical flow is used as a distinguishing basis, quantitative processing is performed for a calculated optical flow field to obtain a discrete motion vector matrix, then directional division is performed for the matrix, and connected domains are searched finally, so as to find a foreground object of which motion direction or speed is different from the motion direction or speed of the background, thereby completing a search process for the foreground object. A continuous vector quantification method and a division method in the present invention can well perform analysis processing for an optical flow vector field. The foreground detection method of the present invention which obtains the foreground object through comparison of the areas of the connected domains is a method for quickly capturing the position of the foreground in the case of uncertain foreground object.

Description

technical field [0001] The invention relates to a foreground detection method, in particular to an image content retrieval method based on dense optical flow and direction division. Background technique [0002] With the coverage of video surveillance systems and the popularity of handheld mobile camera equipment such as mobile phones, video data is growing at an alarming rate, and how to extract valuable information from a large number of video files has become an urgent problem to be solved. [0003] Background modeling and object tracking are the foundation of video analysis. Background modeling distinguishes and recognizes the foreground and the background based on whether the background and the foreground are different in motion, motion direction, and color, and the target tracking realizes the real-time tracking of the target based on the characteristics of the target and the difference between the front and rear frames of the video generated by its motion. track. ...

AI Technical Summary

Problems solved by technology

[0007] Wang Yongzhong, Liang Yan and others from Northwestern Polytechnical University are engaged in the research of target tracking. The main achievement: In view of the shortcomings of the traditional mixed Gaussian background modeling method that is sensitive to non-stationary scenes, a new mixed Gaussian model is proposed.

Images