Research On Simultaneous Localization And Mapping Using Kinect

In this paper,the robot simultaneous localization and mapping(SLAM) is studied in theory and experiment using Microsoft’s 3D camera Kinect as the environmental sensing sensor.This paper mainly completes the following work:(1) The camera model, coordinate system conversion, Kinect hardware structure and function are described in detail.Three methods of robot to obtain environmental depth are introduced:binocular parallax, time of flight (ToF) and Kinect.This paper also introduces data acquisition process of OpenNI and obtains the depth images and RGB images by related components.(2) The intrinsic and extrinsic parameters of Kinect are obtained using the Zhang Zhengyou’s camera calibration method. Meanwhile, the experiment process is analyzed and improved. According to the extrinsic parameters.this paper conducts the registration experiment of depth images and RGB images.(3) The noise sources of Kinect are analyzed in this paper. Influenced by the Kinect sensor, the measurement environment and object properties, the noises of depth image are dealt with the average filter, median filter, Gaussian filter.And this paper proposes filtering algorithm based on depth image of Kinect, bilateral filter. It better smooths the noise and protects the edge of depth image, but the efficiency is low. So the fast Gaussian estimation method is used to accelerate the algorithm.(4) RGB images of Kinect are used to carry out FAST corners, Harris corners, SURF features extraction experiments.And the performance of the three methods are compared. The paper introduces feature matching and region matching. Then the algorithm based on SURF features matching is proposed, which is based on SSD(Sum of Squared Difference). And the RANSAC algorithm is applied to eliminate the error matching.(5) In this paper,the SLAM problem based on the Extended Kalman Filter (EKF) is studied. This paper builds the EKF-SLAM system model and carries out the EKF-SLAM simulation experiment.Then the Kinect is used to collect the RGB and depth data stream.The research of the EKF-SLAM based on the data stream is carried out,and the EKF-SLAM simulation experiment is extended to three-dimensional model.Finally,the location of the Kinect and the construction of feature points map are completed.