Development And Implementation Of A Self-localization Algorithm For Warehouse Autonomous Inspection Robot

The warehouse self-inspection intelligent inventory robot is a technical means to effectively reduce the financial risk of pledge financing.Its purpose is to enable it to inspect the pledged items 24 hours a day,to ensure the authenticity of the pledged items,thereby reducing the warehouse receipt pledge financing risks caused by “false pledge”and “repetitive pledge”.In order to enable the inspection robot to conduct inspections 24 hours a day,the self-positioning algorithm of the inspection robot itself needs to be able to operate 24 hours a day,but the current Visual Simultaneous Localization and Mapping(VSLAM)algorithm used as the robot self-positioning core algorithm still exists problems that need to be solved,including insufficient lighting,lighting changes,and positioning failures affecting inspection tasks.In order to solve these problems,the main work of the thesis is as follows:a)Using infrared images taken by Kinect2 as the input images of VSLAM algorithm,and preprocess infrared images by Gamma Transform to make infrared images can satisfy the requirements of VSLAM algorithm.b)Using IMU information to predict the pose of images.If VSLAM algorithm locate some frame failed,using the predict pose by IMU as the pose of failed frame,then building map continuely by using depth information of the failed frame,that obtained by Kinect2,to keep VSLAM algorithm locating.c)Using AGAST algorithm to detect key points in images,and using LATCH algorithm to descript key points.Additionally,in order to make key points more robust,this paper add a corner filter based on corner determination method to remove key points which corner detector may treat noise as.d)All method proposed in this thesis is implemented based on ORB-SLAM2 algorithm.This thesis test the implementation on open datasets,and equipped the implementation to our patrol robot to test.