Autonomous Positioning In Indoor Unknown Environment Using Kinect Point Clouds

With the development of robot positioning,simultaneous localization and mapping(SLAM)has become a hot topic in computer vision.The main research content is that in the unknown environment,the data obtained by sensors are used to complete the mapping of the scene.With the release of the RGB-D camera,a new sensor for the SLAM problem is provided.As one of the RGB-D cameras,Microsoft Kinect has been constantly updated and upgraded so that the use of Kinect devices to realize the unknown environment positioning has become a hot research topic.At this stage,there are several ways to use Kinect camera for positioning: using the data obtained by Kinect to locate through iterative neighborhood point(ICP)algorithm,which is influenced by its own range of scene and Kinect depth image precision.The error caused by ICP algorithm at last registration results accumulated relatively large,relatively low positioning accuracy;generating positioning and map of indoor scene by using the SLAM method,but when the scene is not closed or closed scene range is very large,processing loop detection the result is not very good.In view of the above situation,this paper is based on the ORB-SLAM system,through the ORB feature extraction,feature points of color image sequence images obtained by Kinect in the extraction used,realize image tracking,obtain the relative conversion relationship between adjacent frames of image,to solve the prior information vector machine information matrix,multi feature information update information in vector filtering and augmented state;in depth image,the calibration results using color images and depth images,color image sequence of point cloud data are generated,the generation of point cloud data,complete the feature extraction and face feature points in the scene,according to the feature extraction results,completed the construction of the characteristics of the observation model;based on the observation model and the results of the augmented state,status updates,solving a posteriori information matrix and vector information.To weaken the influence of the error accumulation in the localization process,the translation parameters in the updated state vector are solved,and the Kinect point cloud data is automatically located in the unknown environment.In this paper,2 experimental scenarios are selected by using Kinect sensors,and experimental verification is carried out.The contents include: the data were obtained by using ORB-SLAM system model,realize the image tracking;on this basis,using ICP to obtain the feature point feature and scene construction of multi feature extended information filter model,the experimental data processing.The experimental results show that this method can reduce the error of the cumulative effect of the Kinect data to a great extent,according to the scene 1,compared with data fusion,the positioning accuracy in the range of 0.2m,according to the scene 2,locate respectively by ORB-SLAM method and the method,the positioning accuracy and 0~0.8m respectively.0~0.5m,the positioning errors are respectively 0.48 M and 0.23 m.