Research On Localization Of Indoor Mobile Robot Based On RGB-D Camera

This article focuses on the problem of locating of indoor mobile robots based on RGB-D cameras.In view of the map drift problem caused by the mapping algorithm in the large environment,this paper proposes that the EKF integrates the visual odometer and the wheel odometer,and the fusion odometer is used as the input of the Gmapping to reduce the map drift.For the case where the initial location is not known,the efficiency and accuracy of the Monte Carlo global locating algorithm are greatly reduced.In this paper,the visual dictionary method is used to determine the initial localization,which improves the robustness of Monte Carlo localization.At the same time,an indoor mobile robot locating system including software and hardware was built.Its main work is as follows:First of all,this paper builds a mobile robot indoor locating R&D platform based on ROS architecture.With reference to the design of excellent mobile robots at home and abroad,the mobile robot design including the upper computer and the underlying motion control platform has been completed,which can complete locating,navigation and obstacle avoidance functions.The lower computer is a motion control platform and is mainly responsible for receiving the upper computer speed command and returning the odometer information.The upper computer performs verification of the locating navigation algorithm and selects Kinectl as the main visual sensor,which reduces the overall cost while ensuring the locating effect.Then,this paper proposes an EKF fusion method using wheel odometers and visual odometers to improve the accuracy of single odometer motion estimation.This paper elaborates the principle of motion estimation for wheeled odometers and visual odometers.At the same time,the advantages and disadvantages of these two types of odometers and the improvement of the motion estimation accuracy of the two odometers are described.Then,the fusion odometer is used as the input to the map created by the Gmapping algorithm to obtain more robust map creation results.Finally,this paper proposes a method of visual dictionary to improve the classical Monte Carlo localization.The classical Monte Carlo localization is under the condition that the initial localization is unknown,its locating accuracy and efficiency are greatly reduced,and even the result of locating failure appears.When the mobile robot performs global locating in a large scene,there will be cases where the number of particles cannot cover the entire map,and increasing the number of particles will cause an increase in calculation time.This article uses the visual dictionary method for initial locating in the global map.By saving the key frame localization information on the construction path in real time,a global map of key frames can be established.When the mobile robot performs global locating,similarity scores are calculated through the visual dictionary through the current visual observation and the key frame map.The localization of a key frame with the highest score is considered as the current localization of the robot.Through the initial locating,the effect of Monte Carlo global localization is greatly improved.