A low-resolution multi-view hand reconstruction method based on a human hand parametric model

Abstract

A low-resolution multi-view hand reconstruction method based on a human hand parameter model comprises the following steps of acquiring a multi-view image of a whole human body; extracting a mask, two-dimensional joints, converting into three-dimensional skeletal points; carrying out the motion deformation of the human hand parameter model and obtaining the human hand motion deformation model; optimizing the shape parameters of the human hand parameter model, and obtaining the human hand shape deformation model; reversely projecting skeletal points and models onto multi-view images, and optimizing and updating the existing two-dimensional skeletal points and masks to obtain the final hand model; splicing the mannequin. The whole process of the invention is automatically completed, the gesture movement and shape characteristics are accurately expressed, the reconstruction precision is high, the hand model and the human body model are spliced and recombined, and a more accurate and complete model effect is obtained.

Description

technical field [0001] The invention relates to the field of three-dimensional reconstruction of human hands in computer graphics, in particular to a low-resolution multi-view hand reconstruction method based on a parameter model of human hands. Background technique [0002] High-quality human body 3D reconstruction technology generally needs to be based on complex multi-view camera systems or depth sensor systems. However, whether it is a color camera or a depth sensor, when a person's gestures are slightly complicated, under this reconstruction system, there will be missing fingers or a lot of hand noise. This is an important challenge for completely collecting the human body model. While not affecting the collection settings of the human body model, the reconstruction accuracy of the human hand model is greatly improved, and the high-quality fusion of the human hand model and the human body is restored. [0003] For color cameras, multi-view human body reconstruction req...

AI Technical Summary

Problems solved by technology

[0003] For color cameras, multi-view human body reconstruction requires image-based feature point matching. When the human body appears in the entire picture, the hand area often only accounts for a "small" proportion; and the low resolution and local Similar hand features will affect the matching results of their feature points

For the depth sensor, when the human body is completely captured by it, according to the difference in the height of the human body and the difference in the field of view of the sensor, the person needs to be more than 3 to 4 meters away from the sensor. Common depth sensors, such as Kinect, only have an effective detection range of 0.5 to 4.5 meters. range; at the same time, the geometric structure of the hand is complex and delicate, and a slight deviation will result in missing fingers and introducing a lot of noise

The joints of the human hand are flexible and the movements are complex. The reconstruction process based on vision will be affected by the occlusion of the hand itself and the mutual occlusion of the hands, which in turn increases the difficulty of high-precision human body reconstruction with complex gestures.

Images