Figure outline quick detection and tracking method based on depth image

Abstract

The invention relates to the technical field of real-time figure outline detection and tracking of a depth image, in particular to a figure outline quick detection and tracking method based on the depth image. The figure outline quick detection and tracking method based on the depth image is characterized by comprising a figure outline quick detection step and a figure outline mask real-time tracking step; the figure outline quick detection step comprises the steps that S1, a background region is determined through movement of a figure; S2, update is performed through a background model update strategy according to the size and the direction of a pixel value change of the depth image in the background region; and S3, a figure outline region is obtained through background deduction, and background update is performed again depending on the figure outline region; and the figure outline mask real-time tracking step comprises the steps that S4, quick partitioning is performed on a figure outline mask region based on depth information; S5, an outline obtained after partitioning is tracked based on outline mask superposition; and S6, a re-detection strategy is adopted.

Description

technical field [0001] The present invention relates to the technical field of real-time human contour detection and tracking of depth images, and more specifically, to a method for rapid detection and tracking of human contours based on depth images. Background technique [0002] Nowadays, human-computer interaction technology is developing rapidly, and the posture analysis of human body is an indispensable part in the field of human-computer interaction. Extracting human body contours from depth images is of great significance for posture analysis. The contour information of people in depth images is conducive to the analysis of human skeleton points, and then the secondary development of applications such as somatosensory games and virtual fittings. At present, commercial algorithms in this field include the KinectV2 SDK developed by Microsoft and the NITE2 algorithm developed by Primesense, both of which can extract the outline of a person in the depth image and calculat...

AI Technical Summary

Problems solved by technology

However, both of them have deficiencies when the scene is complex. The human body mask extracted by KinectV2 SDK is prone to noise when the person is close to the depth camera (<1.5m), and the person touches objects in the scene, such as chairs, Tables, etc., are prone to mask mutation loss or growth, and character ID information is easily confused when exchanging positions between characters.

Moreover, the KinectV2 SDK has high requirements on computing resources, requiring GPUs after DirectX 11.0. Many notebooks cannot run without NVIDIA GeForce and AMD Radeon external GPUs.

The NITE2 algorithm also has mask mutation loss or growth when people approach or touch objects in complex scenes, which will affect the recognition of human bone points

Images