Research On Image Feature Processing And Camera Pose Estimation Based On Visual SLAM

The development and application of mobile robot technology increasingly affect people's life and work,and change people's life style.Nowadays,with the rapid development of modern high-tech technology and the continuous improvement of manufacturing level,the development of mobile robot technology is also more and more towards the direction of autonomy and popularization.In the hot research field of mobile robot technology,more and more professionals are engaged in the research of the technology of simultaneous localization and mapping(SLAM),because it is the basis for mobile robots to achieve autonomous navigation and environmental exploration in unfamiliar environment,and also has a very important theoretical significance for mobile robots to achieve real intelligence.Therefore,the research of SLAM technology is very valuable and challenging! The essence of SLAM technology is that when a mobile robot is exploring in a strange environment,it uses its own data sensors to obtain the observation information of the surrounding environment,and then it is convenient to use these sensor data to build an incremental map consistent with the environment.Finally,it uses the built environment map to estimate the current pose of the mobile robot.Therefore,the accuracy of positioning and mapping determines whether the mobile robot can land in the actual application scene.Since the SLAM technology was put forward,many classical algorithms and frameworks for robot positioning and mapping have emerged.These algorithms,combined with the unique advantages of visual sensors,have also improved the application level of SLAM.However,due to the low accuracy and poor robustness of these existing algorithms in the process of visual image processing and pose positioning,there are still many problems to be solved in the research of SLAM technology.Visual SLAM is one of SLAM technology,which uses camera as its data sensor.This paper mainly studies the extraction and matching of image feature points and some classical algorithms of camera pose estimation in vision SLAM,and carefully analyzes the shortcomings of these classical algorithms in the practical application process.Finally,combined with the theory of computer vision and probability theory,these classical algorithms are improved accordingly.Experimental results show that the improved algorithm effectively improves the accuracy and robustness of key frame image processing and pose estimation.The specific research work of this paper mainly includes the following aspects:First of all,this paper analyzes and introduces the current research situation of SLAM technology at home and abroad,then introduces some background knowledge about vision SLAM problem in detail,as well as some classical algorithms about image processing and camera pose estimation.Secondly,according to the existing key frame image feature point extraction,matching algorithm and the camera pose estimation problems of rgb-d vision SLAM,some improvements have been made to the original algorithm.The experimental results show that the improved algorithm is effective.