Unsupervised pose and depth calculation method and system

Abstract

The invention discloses an unsupervised pose and depth calculation method and a system. The unsupervised pose and depth calculation method mainly adopts the following modules: a pose prediction network model TNet, a depth estimation network model DMNet, a visual reconstruction model V and an error loss function module. The method comprises the steps of calculating a forward motion relative pose and a backward motion relative pose; calculating the depth estimation result of the image and the corresponding depth of the image, summing the reconstruction error, the smoothing error and the twin consistency error to obtain a loss function, performing iterative updating until the loss function converges, and finally calculating the relative pose and the predicted depth map of the camera accordingto the trained model Tnet and model DNet.

Description

technical field [0001] The invention belongs to the fields of SLAM (Simultaneous Localization And Mapping) and SfM (Structure from Motion), in particular to an unsupervised pose and depth calculation method and system. Background technique [0002] In recent years, the algorithms of monocular dense depth estimation and visual odometer VO (Visual Odometry) based on deep learning methods have developed rapidly, and they are also key modules of SfM and SLAM systems. Existing studies have shown that VO and depth estimation based on supervised deep learning achieve good performance in many challenging environments and alleviate performance degradation issues such as scale drift. However, to train these supervised models in practical applications, it is difficult and expensive to obtain enough ground truth labeled data. In contrast, unsupervised methods have the great advantage of only requiring unlabeled video sequences. [0003] Deep unsupervised models for depth and pose esti...

AI Technical Summary

Problems solved by technology

However, the existing technologies rarely take into account the following key issues: the timing of VO, and ignore the shortcoming that the unmanned driving data set only has a single direction of motion. The model can only handle one-way motion and does not use forward and backward motion constraint

The existing model does not consider the complexity of the model, and the number of parameters is large, so it is difficult to apply to the actual application scenario of VO

Images