Visual odometer method based on end-to-end semi-supervised generative adversarial network

Abstract

The invention discloses a visual odometer method based on an end-to-end semi-supervised generative adversarial network, and relates to a visual odometer calculation method. The method comprises: constructing a feature generation network; constructing a discrimination network; performing adversarial training; solving dynamic scene problems. The method specifically comprises: marking point positioninformation and extracting feature descriptors through an SIFT feature algorithm, and obtaining related frames and matching feature points by randomly generating a homography matrix so as to generatecorresponding training labels; the generation network can input an original image to generate corresponding feature point positions and corresponding depth descriptions; and the discriminant network combining the semantic-geometric consistency loss function, the feature point cross entropy loss function and the discriminant loss function to form an adversarial with the generative network. Throughtraining of the GAN, the generation network can generate point position information and depth description which cannot be distinguished by the discrimination network, so that manual design of local features is avoided.

Description

technical field [0001] The invention relates to a visual odometry calculation method, in particular to a visual odometry method based on an end-to-end semi-supervised generation confrontation network. Background technique [0002] In the past few decades, the field of mobile robotics and autonomous driving technology has attracted extensive attention from researchers all over the world, and significant progress and breakthroughs have been achieved. At present, mobile robots can perform complex tasks autonomously. For example, the robot dog developed by Boston Dynamics Engineering Company has been able to imitate humans to do backflips and open doors. Major breakthroughs have also been made in autonomous driving technology, and autonomous vehicles are expected to be mass-produced within two years. Both mobile robots and self-driving technologies require autonomous navigation in complex and dynamic indoor or outdoor environments. In order to be able to navigate autonomously,...

AI Technical Summary

Problems solved by technology

At present, this supervised learning method has the following difficulties: First, it is extremely difficult to obtain data with a large amount of geometric information annotations. One hundred thousand images are tens of millions or even hundreds of millions of points, so it is very difficult to label such a labor-intensive

Therefore, some researchers, such as Kendall et al. [2] Using Visual SFM with Transfer Learning [3] Marking the image pose saves a lot of time and labor costs, but the accuracy is also limited by the Visual SFM algorithm

Secondly, if you only learn the pose of a single image corresponding to the current reference coordinate system, then the usage scenarios will be greatly limited

Images