Research Of Dense ORB-SLAM Based On RGB-D

Simultaneous localization and mapping(SLAM)technology in virtual reality,augmented reality,UAVs,service robots and other fields has broad application prospects and has been a hot research topic in China and abroad.With the invention of RGB-D cameras,a new research idea has emerged in Visual SLAM(VSLAM).In SLAM based on RGB-D sensors,RGB-D camera can provide RGB information and depth information at the same time,however,there are problems such as limited vision,narrow measurement range,and large noise.The whole SLAM algorithm needs to balance the contradiction between real-time performance and accuracy,the contradiction between dense maps and real-time performance,and so on.In this dissertation,SLAM based on RGB-D sensor is studied in detail.SLAM method to improve the shortcomings of RGB-D camera and balance real-time performance,accuracy and dense map is proposed.Improved ORB algorithm to meet the needs of SLAM is given.And based on it,a dense SLAM framework algorithm based on RGB-D data is proposed.The main work of the dissertation is as follows:First of all,a large number of related literature is reviewed,the concept,significance and research status of VSLAM are briefly described,and the advantages and disadvantages of several SLAM methods based on RGB-D are summarized.Secondly,the RGB-D sensor used in this dissertation is analyzed,and the conversion relationship between spatial point and pixel point is described.Then the VSLAM with camera mode is analyzed in detail,and the core idea of this method-nonlinear optimization(Graph Optimization)is clarified.Then,in order to solve the contradiction between real-time performance and precision,the ORB feature is improved for the front-end feature extraction part of SLAM.The ORB algorithm can meet requirements of real-time and accuracy in a compromise way.But it lacks adaptability to scale changes and the features extracted are clustered and uneven.That can impact feature extraction,matching and accuracy of SLAM system.Therefore,a multi-scale spatial pyramid is built in the feature detection section to improve the adaptability of the algorithm to scale changes and through the division of the grid to get evenly distributed features.Then,the features are described and matched,and the features suitable for SLAM are finally obtained.Test experiments show that compared with the ORB algorithm,the improved ORB algorithm has a uniform feature extraction while satisfying real-time requirements.Finally,an improved RGB-D SLAM framework based on improved ORB is proposed.Firstly,the method of determining the key frames that this article relies on is given.Next visual odometry(VO)is implemented to get the motion relationship between frames.Then implement loop detection and design pose diagram optimization to ensure global consistency,and finally output a dense map.This paper uses the TUM data set which is proposed by Computer Vision Group of Faculty of Informatics of Technical University of Munich to test and contrast the proposed method.The experimental results show that the SLAM algorithm based on RGB-D using improved ORB proposed in this paper not only can meet the real-time requirements,but also output dense maps,and maintains balance between real-time performance,accuracy,and dense map.