Virtual viewpoint synthesis method based on GPU acceleration

Abstract

The invention discloses a virtual viewpoint synthesis method based on GPU acceleration, and a high-quality virtual viewpoint image is synthesized in an accelerating manner by utilizing a depth image based rendering (DIBR) technology and a GPU multithreading parallel characteristic. The method comprises the following steps: firstly, performing smoothing processing on depth images of a left reference viewpoint and a right reference viewpoint; secondly, restoring two-dimensional plane coordinates of various pixel points in the depth images, after the smoothing processing, of the left reference viewpoint and the right reference viewpoint to a three-dimensional space through 3D mapping; then projecting coordinates in the three-dimensional space to a virtual viewpoint imaging plane by using graphics API rendering so as to obtain a synthetic virtual viewpoint image; and, finally, performing hole filling of the synthetic virtual viewpoint image. The method can utilize a powerful parallel computing ability of a GPU, accelerate a traditional virtual viewpoint synthesis algorithm, effectively reduce rendering time, and provide a great reference for real-time virtual viewpoint synthesis.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a virtual viewpoint synthesis method based on GPU acceleration. Background technique [0002] TV technology was born in the 20th century, and has experienced the development process from black and white to color, from analog to digital, and from standard definition to high definition. While the definition continues to improve, three-dimensional stereoscopic television (3DTV) is also developing continuously, becoming another development direction after high-definition television. In recent years, the continuous emergence of three-dimensional movies has also promoted the development of three-dimensional stereoscopic television and stereoscopic display technology. Multi-view TV is a development direction of stereoscopic TV. Compared with traditional 3D TV, multi-view TV provides images from multiple viewpoints, enabling people to obtain better stereoscopic experi...

AI Technical Summary

Problems solved by technology

[0005] (2) The academic community has proposed many synthesis algorithms and image repair algorithms to improve the quality of virtual viewpoint synthesis, but these algorithms are often complex and time-consuming to run, and are difficult to apply to occasions with high real-time requirements

[0006] (3) Most of the existing virtual viewpoint synthesis algorithms are implemented based on CPU, and the synthesis of virtual viewpoint images is relatively slow

Images