Research On Indoor Positioning And Map Construction Based On Machine Vision

With the continuous improvement of people's living standards and the continuous development of artificial intelligence technology,indoor mobile robots have become a better choice for daily cleaning and pet care in most homes.The main purpose of the simultaneous positioning and map creation(SLAM)technology research is to allow mobile robots to locate and map in real time in an unknown environment.This article explores SLAM technology,and takes the indoor environment as the research scenario to build a monocular vision indoor positioning and map construction system to achieve indoor positioning and mapping functions.In this paper,the common ORB-SLAM2 has the problems of slow positioning speed,and the map cannot be used for path navigation.To solve these problems,the visual odometer and map building module are improved separately.The indoor positioning and positioning based on machine vision are designed and developed.The map construction system better meets the needs of indoor positioning and mapping.In the visual odometer module,this article uses a semi-direct method to implement the visual odometer,replacing the feature point method in ORB-SLAM2,reducing the calculation of a large number of feature points,greatly reducing the matching time and improving the positioning speed.In the mapping module,because the sparse roadmap map used by ORB-SLAM2 cannot be used for path navigation,and common point cloud maps take up too much storage space,this paper chooses to build a three-dimensional map based on the octree structure through experimental comparison.Maps not only support subsequent path navigation and obstacle avoidance,but also take up much less storage space than point cloud maps.In order to verify the function of the system,the system and ORB-SLAM2 were simulated on a standard data set.The experimental results show that the indoor positioning and map building system constructed in this paper is roughly the same in positioning accuracy as ORB-SLAM2,and the running time is significantly less than that of ORB-SLAM2,and some data sets can reduce the running time by 60%.The improved positioning speed of the visual odometer is significantly higher than that of ORB-SLAM2.The average number of image frames processed per second can be increased by up to 162%,and the average time required to process each frame image is reduced by up to 61% compared to the original system.Compared with the sparse roadmap in ORB-SLAM2,the octree map selected in this paper has the role of path planning and supporting subsequent navigation.In terms of storage space,the octree map can save about 90% of storage compared to the point cloud map which is more suitable for intelligent systems with limited storage space and computing power.