Indoor Visual SLAM Based On ROS System Technology Research And Implementation

In recent years,with the rapid growth of the Chinese robot market demand,the country has begun to vigorously develop the robot industry.Among them,autonomous mobile robots have become an important direction in the current robot research and application field,and it is also an inevitable trend in the development of robot technology.Visual SLAM is to realize the autonomous movement of robots.The premise of this is to facilitate navigation and path planning by knowing its specific location on the map.This article mainly researches the indoor visual SLAM technology based on ROS system,and provides related experiments for verification.The main research contents are as follows:Firstly,the imaging principle of the camera model and the principle of RGB-D camera output depth image are studied,and the process of acquiring images by camera motion is transformed into the observation equation and motion equation of the camera,using GN or LM for nonlinear solution,and giving Out of the basic framework of visual SLAM.Secondly,the visual odometer based on the ROS system is studied.Aiming at the problem of the corner point concentration and uneven feature distribution in the traditional feature point extraction algorithm,a quadtree-based ORB feature point homogenization extraction method is introduced;for features After the point extraction,the traditional screening mismatching algorithm still has a lot of mismatch problems.After brute force matching,this paper first uses the traditional screening mismatching algorithm to perform mismatch screening,and then introduces the RANSAC algorithm to further remove the mismatches,and conduct related experimental verification.Then the visual SLAM framework based on the ROS system is studied.The large amount of data collected by the camera will cause image information redundancy.The key frame selection strategy is given.The pose map is used to optimize the initial trajectory of the camera,but the trajectory error still exists.The loop detection module builds a bag of words model,establishes the data association between the current frame and the historical frame,further optimizes the camera motion trajectory,and finally uses voxels to construct an octree map.Finally,the positioning accuracy and mapping function of the visual SLAM were verified through experiments,the software and hardware platform was built for the experiment,the Kinect2.0 camera was calibrated,the TUM data set and the actual scene were used for related experiments,and the APE and The comparison of the RPE trajectory accuracy evaluation standard found that the estimated motion trajectory accuracy of this system is high,and then a globally consistent point cloud map is constructed based on the internal and external parameters on the key frames,and the point cloud map is converted into an octree map.It is found that the map construction is compared Clear and able to meet basic mapping requirements.