Target analysis method and device based on intensity image and range image

Abstract

The invention provides a target analysis method and device based on intensity image and range image. The device comprises a main part detecting unit, a foreground segmentation unit and a moving part detecting unit, wherein the main part detecting unit detects the main part of a target from the intensity image of the target; the foreground segmentation unit calculates the intensity threshold on the basis of the intensity value of the detected main part, uses the intensity threshold to convert the intensity image to a binary image and uses the binary image to mask the range image to obtain a mask range image; and the moving part detecting unit detects the moving part of the target from the mask range image.

Description

technical field [0001] The invention relates to a video object analysis method and equipment. More specifically, the present invention relates to a target analysis method and device based on intensity images and depth images, by which each component of a video target of interest can be accurately detected from a video stream to obtain target action information. Background technique [0002] With the wide application of 3D animation, games, human-computer interaction interface, etc., in the field of computer vision and pattern recognition, video object analysis technology is being extensively studied. In order to perform animation simulation, it is first necessary to detect the various parts of the target from the video in order to obtain the target's motion information [0003] To detect parts of objects, bottom-up methods and up-bottom methods are usually used. In the former method, the target part is firstly detected by its features, such as detection by skin color, bar...

AI Technical Summary

Problems solved by technology

The background of color images is generally very chaotic, so foreground segmentation based on color images is very computationally expensive and imprecise

Although a foreground segmentation method based on depth images has also been proposed, this method is not accurate

[0005] Once the foreground segmentation is inaccurate, it will make the subsequent target analysis very difficult, and the target analysis results will be disturbed by background noise with similar characteristics

In addition, there are often abundant edge features in the chaotic background, and these edge features become noise, which affects the edge feature analysis results in target component detection or hypothesis evaluation

Images