Research On Visual SLAM Algorithm Based On Graph Optimization In Indoor

In recent years,Simultaneous Localization and Mapping(SLAM)technology is a hot research direction in the field of robotics.It is considered as a key technology to solve the problem of mobile robots implementing autonomous navigation in an unknown environment.SLAM technology refers to an object carrying a specific sensor,estimating its own motion during the movement in an unknown environment,and establishing a model of the environment.Therefore,the mobile robot through SLAM technology not only can obtain the location information of its own in the unknown environment in real time but also can construct the model of the environment,which lays a foundation for autonomous navigation.Compared with other types of sensors,vision sensors have many advantages such as large amount of information,wide application range,and high cost performance.Therefore,the visual-based SLAM technology is very popular among researchers at home and abroad.This article selects Microsoft's Kinect 2.0 camera as a visual sensor to investigate the problem of simultaneous positioning and mapping in an indoor environment.The main research contents of this paper are as follows:(1)The structure and function of the Kinect 2.0 camera,the principle of depth measurement,the image acquisition method,and the model of the camera were studied.Using the Zhang Zhengyou calibration method to calibrate the Kinect 2.0 camera,the internal and external parameters of the camera were obtained to lay the foundation for the subsequent research work.(2)The problem of information loss in depth images acquired by Kinect 2.0 camera has been deeply studied.Based on the characteristics of the depth image's own data,it is proposed to use the method of guided filtering to repair the depth image.Experiments show that the method effectively solves the problem of depth image information loss and reduces the pose estimation error of the system.(3)The techniques of feature point extraction,feature point matching,motion estimation and key frame selection in visual odometer are studied.Aiming at the large number of mismatches in feature point matching results,this paper improves the feature point mismatch culling method based on the research of minimum threshold method and ransac algorithm,which effectively shortens the time required for feature point matching.Improve the efficiency of the algorithm.(4)Mathematical modeling of the visual SLAM problem is performed to transform the SLAM problem into a nonlinear least squares problem.The graph optimization model of visual SLAM problem is expounded,and the commonly used graph optimization solving tool g2 o is introduced,which provides theoretical support for the back pose optimization.(5)In this paper,the performance of feature point extraction and feature point matching algorithm are compared by using the public data set.At the same time,the effect of the guided filtering algorithm is verified.The experimental results show that the poses calculated by the visual SLAM system after filtering are all Square root error decreased by 4.8% compared to before filtering.Finally,the visual SLAM in the indoor environment is realized by holding the Kinect 2.0 camera.