Research On Localization Of Indoor Robot Based On Depth Vision

Mobile robot localization technology is one of the key technologies to realize autonomous navigation.The location technique mainly relies on the 3D information of an image.Plenty of visual odometry algorithms have been proposed,according to different image data-processing methods.However,none of the existing algorithms can address the location problem of any environment perfectly.Therefore,with the image sequences obtained in multiple environment,the robustness and precision of feature point based method and direct method are analyzed in this article.A method which can handle the environment degradation problem effectively based on the feature of environment is provided.Experiments are carried out to verify the correctness of the algorithm in different kinds of indoor environment.The main contributions in this article are as follows:Firstly,a visual sensing system based on Kinect V2 is designed to collect the information on environnent and attain the data sets of color image and depth image.Comparisons are made among several depth acquisition approaches and simulation experiment are conducted to demonstrate the high efficiency of Kinect V2 based platform To rectify the aberration of camera and get the corresponding relations between the space and image coordinate system,the camera was calibrated and the intrinsic and extrinsic parameters of the camera are obtained simultaneously.Secondly,a robot location model are developed and the motion estimation which bases on the feature points and the direct method of RGB-D visual odometry methods respectively,are realized by selecting data of different environment.Experiment are designed to compare the performance of the two methods.A method based on the feature of environment is provided.By considering the feature points as a threshold,the feature points based method and the direct method are combined.The feature points based method is selected in feature-rich environment;And the robust performance are guaranteed by using direct pixel information based method to realize location estimation in feature lack environment.Finally,in order to verify the effectiveness of the proposed odometer selection algorithm based on the environmental characteristics,three sets of image sequences which contain different feature point information are captured by using KinectV2 based visual sensing system to illustrate the performance of this optimization algorithm.The result show that the optimization algorithm obtains high precision perforance in feature-rich environment.And the robust performance are guaranteed when the feature points within the environment are severely missed.