Robust Pose Estimation In Challenging Environments

Position estimation technology has always been a research hot spot in the field of intelligent mobile robots.It is also the basis of Simultaneous Localization and Mapping(SLAM)technology.In recent years,pose estimation technology has been comprehensively studied and achieved promising results,providing many good solutions for six degrees of freedom(DOF)state estimation,mapping,obstacle avoidance and navigation of mobile robots.However,most pose estimation methods are discussed under "normal scenes",so these methods are difficult to deal with some extreme "challenging" environments in daily life,such as low texture scenes,high-speed high-dynamic scenes,which limits the application of mobile robots.In this paper,we mainly focuse on some "challenging scenarios" in practice,and propose robust pose estimation methods for these scenarios.The main work of this thesis are composed of the following three parts:(1)In order to solve the pose estimation problem of mobile robots in visually degraded environments,a real-time,robust and low-drift depth vision SLAM(simultaneous localization and mapping)method for depth cameras is proposed by utilizing both dense range flow and sparse geometry features from sequential depth images.The proposed method is mainly composed of three optimization layers,namely range flow based visual odometry layer,ICP(Iterative closest point)based pose optimization layer and pose graph based optimization layer.In range flow based visual odometry layer,range change constraint equation is used to solve fast 6 DOF frame-to-frame pose estimation of camera.Then,local drifts are reduced by applying local map in ICP based pose optimization layer.After that,the loop closure constraints are built by extracting and matching sparse geometric features and a pose graph is constructed for global pose optimization in pose graph based optimization layer.Performances of the proposed method are tested on TUM datasets and in real-world scenes.Experiment results show that our front-end algorithm outperforms the state-of-the-art depth vision methods in challenging environements,and our back-end algorithm can robustly construct loop closures constraints and reduce the global drift caused by front-end pose estimation.(2)Traditional 3D reconstruction methods only reconstruct the geometry and texture information of the environments.In this paper,we present a real-time reconstruction system that can not only reconstruct the geometry and texture information of the environments,but also the thermal temperature information which is beyond human eye.This perception ability is not only useful in search and rescue robots,but also in building energy auditing,and fault diagnose of critical equipments in factories.The algorithm consists of three parts:external and internal parameters calibration of thermal camera and RGB-D camera,construction of RGB-DT image pairs,and pose estimation by fusing a depth-ICP algorithm and a direct RGBDT visual odometry method.We carried out several experiments to demonstrate the performance of the proposed system.The experiment results show that our system can achieve large-scale 3D multi-dimensional reconstruction in real-time.(3)In order to solve the pose estimation problems such as "long corridor","high dynamic"and "high speed" faced by autonomous car in high way scene,a error-state based Kalman filter method is proposed to achieve laser and IMU fusion.This method can not only realize pose estimation in geometrically degraded scenarios,but also overcome the bad influence of dynamic objects on laser measurement.The algorithm consists of two parts.Firstly,the motion compensation of each laser point is achieved by calibrating the laser and IMU.Secondly,the data fusion between laser and IMU is realized by establishing Kalman filter equation based on error state.The experimental results show that the proposed laser-IMU fusion method can effectively achieve robust pose estimation in high-speed and high-dynamic scenarios.Finally,The paper is summarized and future research work is proposed.