Feature detection network training method, enhancement of existing, virtual and real registration tracking method and occlusion processing method

Abstract

The invention relates to a method for feature detection network training, enhancement of present-virtual-real registration tracking and occlusion processing, and the method comprises the steps: obtaining a monocular image training sample containing annotation information, the annotation information comprising an actual key point heat map and an actual depth map corresponding to each monocular image in the monocular image training sample; determining a value of a loss function of the feature detection network according to the actual key point heat map and the actual depth map; and according to the value of the loss function, adjusting parameters of the feature detection network to meet a convergence condition, and determining a completely trained feature detection network. According to the method, the feature detection network is utilized, the monocular image can serve as input, camera pose solving and depth map prediction are carried out at the same time, the processing flow efficiency of the augmented reality system is improved, and the requirement of the system for hardware is reduced.

Description

technical field [0001] The present invention is in the technical field of augmented reality, in particular to a method for feature detection network training, augmented reality virtual-real registration tracking and occlusion processing methods. Background technique [0002] With the development of augmented reality technology, its applications in various industries such as manufacturing, education, medical care, and entertainment are becoming more and more extensive. At the same time, with the development of hardware devices, such as the popularity of head-mounted glasses and mobile tablets, the application of augmented reality technology has also been promoted. This makes the augmented reality technology put forward the requirements of wider adaptability, higher timeliness and higher accuracy. The existing augmented reality registration tracking and virtual and real occlusion processing often need to be processed separately through different algorithms, and rely on the co...

AI Technical Summary

Problems solved by technology

[0004] In view of this, it is necessary to provide a feature detection network training, augmented reality virtual reality registration tracking and occlusion processing method, in order to overcome the problem that the augmented reality application method in the prior art is not efficient enough and accurate

Images