Pose Estimation Method Based On Salient Information Fusion For Depth Cameras In Low-texture Environment

Simultaneous localization and mapping(SLAM)technology is the foundation of autonomous mobile robot,and has broad application prospects.Robots could get perception of unknown environment and localize themselves using SLAM technology,and complete tasks such as automatic driving and autonomous navigation.Visual odometry(VO)is an important part of visual SLAM.Usually,cameras are used as sensors to obtain image input information and camera motion between frames can be estimated,which is also known as camera pose estimation.Camera pose estimation plays an important role of SLAM technology,because it affects the real-time performance and accuracy of SLAM systems.The study of pose estimation method is of great significance to the development of robot technology.In recent years,with the rapid development of computer vision technology,the problem of camera pose estimation has been preliminarily solved.However,in lowtexture environment,pose estimation is still difficult due to the small number of feature points and low signal-to-noise ratio.Pose estimation method based on feature points is put forward in early years,which extracts the feature points in the images and matches the points between frames to estimate camera pose.However,in low-texture environment,due to the small number of detected feature point,the feature point method is prone to the situation that the number of matching feature points is not enough to estimate the pose,which results in the interruption of camera pose tracking.On the other hand,the pose estimation method based on the direct method assumes that the intensity is invariant,and then the optimal pose is obtained by minimizing the intensity residual between all or part of the pixels in the image.However,the direct method is easily affected by the low signal-to-noise ratio of low-texture environment thus the accuracy of pose estimation is greatly reduced.In order to solve the problem of pose estimation in low-texture environment,there are some pose estimation methods based on line features to replace feature points.But line feature extraction is complex and it is difficult to compute the descriptors,so using line features is not suitable for the back-end optimization.Other methods increase the signal-to-noise ratio by extracting pixels at the edge position in the image,but the unstable edge extraction and the small number of pixels are likely to cause poor robustness in pose estimation.Also,the environment with less texture is not conducive to mini-mizing intensity error and iteratively optimizing poses,which greatly affects the accuracy of pose estimation.Aiming at the low signal-to-noise ratio of pose estimation in low-texture environment,this paper proposes a pose estimation method based on salient information fusion for depth cameras.This method makes full use of the salient information in the image,which improves the signal-to-noise ratio of pose estimation in low-texture environment and makes the pose estimation more stable.The depth based error minimization method is introduced to use the structural information in the environment and improve the accuracy of pose estimation.In order to test the performance of this method,experiments on the standard dataset are conducted to show the effects of this method.The main work of this paper is as follows:1.The traditional method and the advantages of deep learning are used to extract the edge,and then the extracted edge is used to estimate the pose of the camera based on the direct method.The non-parametric statistical method is used to fit the residuals of edge pixels,and the pixel weights involved in the calculation are given by that.Through experiments on 7 sequences of standard dataset TUM,the superiority of the edge fusion pose estimation method is proved.2.Gray image edge extraction,depth image edge extraction and high residual area extraction are fused as salient information.On the basis of minimizing intensity error,a method of minimizing depth error is introduced to jointly calculate camera pose.Through experiments on sequences of low-texture dataset,the accuracy of pose estimation based on salient information fusion for depth camera in low-texture environment is proved.