Research On Simultaneous Localization And Mapping Based On Jetson TX1

As the demographic dividend decreases,service robots have received widespread attention in recent years.In the face of an indoor unstructured environment like home,SLAM(Simultaneous Localization And Mapping)technology appears important to robots Particularly.In the early SLAM application,LIDAR was mainly used as environment-aware sensor.With the further research in the field of computer vision,scholars use camera as the data source and used SFM(Structure From Motion)Structure)to solve the SLAM problem.That not only brings the solution of observational data association,which enables the system to carry out loop detection and also reduces the cost of robot SLAM technology application,and promotes the application of SLAM technology.The research work in this paper mainly includes the following four aspects:(1)Described the components of visual SLAM.Compared the performance parameters of LIDAR and depth camera,the feature points method based on image local feature is mainly described in the front part,and orb feature points extraction and matching experiments are carried out.Described the application of RANSAC algorithm in the process of basic matrix optimization.The back end compares the filter method with the nonlinear optimization algorithm derived from the SFM problem.The commonly used word bag method in loop detection was described.(2)Aiming at the problem of using PC as a mid-plane in the existing SLAM scheme experiments,a SLAM experimental robot with embedded development board Jetson TX1 as a mid-plane was set up to obtain data information through remote login and control robot.Much more suitable for real application environment.In the aspect of hardware,CUDA,a parallel computing programming model supported by Jetson tx1,are being described.In software,the advantages of ROS platform,communication mechanism,and the tools that needed in SLAM experiment in ROS are described.At the end of the experiment,the SLAM experimental platform built can well accomplish the robots synchronization positioning and mapping tasks.The data processing speed can meet more than 5 frames per second real-time requirements.(3)The calibration of depth camera was described.The advantages of using Kinect2 camera in SLAM based on depth camera are being list,and the main part of calibration experiment is described.According to the camera imaging model,the calibration principle is described and the process of solving the camera's internal and external parameters is obtained.Finally,Calibrated the Kinect2 depth camera,achieved the internal and external parameters of the depth camera and analysis verify the correctness of the parameters obtained.(4)Two-dimensional and three-dimensional SLAM experiment have being done in the use of Kinect2.The paper analyzed the input and output,information processing flow,environment configure process and solutions of the three SLAM schemes:gmapping,ORB SLAM2 and RTAB MAP,and analyzed the problems encountered in the operation scheme of the mobile test platform.Finally,Three SLAM programs were being compared.