Dynamic Mean shift kernel bandwidth updating method based on compressed domain fusion

Abstract

The invention discloses a dynamic Mean shift kernel bandwidth updating method based on compressed domain fusion. According to the method, compressed domain analysis and a Meanshift tracking algorithm are combined together, namely probability statistics analysis is performed on a motion vector generated in a video encoding process to obtain a size of a target motion, so that a color probability model of a target and a size of a kernel bandwidth are dynamically updated according to the size of the target motion, and a target model is more real. According to the technical scheme disclosed by the invention, the motion tracking precision under a condition that the target size is obviously changed is improved, and the operation efficiency is improved. The scheme is especially suitable for a situation that video encoding and target tracking in intelligent video monitoring equipment are executed at the same time.

Description

technical field [0001] The invention belongs to the technical field of video intelligent analysis, in particular to a method for dynamically updating mean shift kernel window width based on compressed domain fusion. Background technique [0002] The Mean shift algorithm (mean shift algorithm) is a non-parametric density estimation algorithm first proposed by Fukunaga in 1975. As an efficient pattern matching algorithm, it has been successfully applied in the target tracking system with high real-time requirements. Dorm Comaniciu first applied the Mean Shift algorithm to the fields of image filtering, segmentation and target tracking. Bradski proposed a Mean shift target tracking algorithm with color histogram as the target mode. The algorithm first uses the color histogram to obtain the color projection map of each frame image, then adaptively adjusts the position and size of the search window, and takes the obtained optimal center position as the center of the target thro...

AI Technical Summary

Problems solved by technology

[0009] (1) The size of the kernel window width remains unchanged during the tracking process. When the target has obvious scale changes, it may lead to tracking failure;

[0010] (2) The color histogram is a relatively weak description of the target features. When there is similar color interference (occlusion), the algorithm effect is not good;

[0012] In addition, the improvement of the tracking effect of the mean shift algorithm mainly depends on the accuracy of a single color feature to describe the tracked target. If this idea continues, it is difficult to improve the tracking effect of the mean shift target based on the color feature.

Images