Construction And Optimization Of Improved ORB–Based Maps

With the rapid development of the robot industry,more and more service-oriented robots have entered people's lives,and people's requirements for intelligent robot service levels are getting higher and higher.The key condition for improving the service level of robots is to realize the autonomous operation of robots.Achieving robotic autonomous operation involves two core issues: “Where am I?” and “What is the surrounding environment?”.At present,the mainstream method to solve this problem is robot real-time positioning and map construction technology,which combines visual sensor technology and image processing technology to solve the problem of robot positioning and environment map construction in indoor environment.ORB-SLAM is a monocular SLAM system based on image features.It uses visual sensors to obtain environmental image information,then analyzes and processes the image,constructs an environmental map,and optimizes the constructed map.This technology is very robust and can be used to build real-world maps.In this paper,the ORB-SLAM positioning technology is deeply studied and explored,and in the process of feature point extraction,due to the uncertainty of indoor illumination and environmental information,it is easy to cause less feature point extraction when the image contrast is insufficient,which ultimately leads to the failure of the slam process.The jitter during the operation of the robot will cause the acquired image information to be blurred,and the extracted feature points are less,so that the tracking process fails.When matching between adjacent images,the exhaustive method is adopted,and the matching efficiency is relatively low.During the map construction process,as time goes on,the matching points will increase accordingly.The quality of the matching points will greatly affect the matching accuracy,thus affecting the clarity of the constructed map.Aiming at the above problems,this paper puts forward the corresponding solutions,the main work of this paper is as follows:(1)In the process of feature extraction,ORB-SLAM technology uses FAST operator to extract feature points of images,while FAST feature points do not have scale invariance,and in the process of feature point matching,the final matching success rate is lower if the image size changes or the image contrast is low.In this paper,the original feature point extraction algorithm is improved by combining the idea of SURF feature point extraction,and the improved algorithm has scale invariance,and the improved algorithm has good robustness to image feature point extraction when the image contrast is low due to the change of environmental information or the image blur caused by the jitter of the robot during motion.(2)ORB-SLAM technology in the image matching process,that is,after obtaining the descriptor,the exhaustive method is used to calculate the Hamming distance between the feature points of two adjacent images,and the time complexity is high.This paper presents an idea of classifying feature points first and then matching them.Classification of image feature points using Hamming distance based on ORB descriptors.In the matching stage,the Hamming distance between the feature points is calculated to determine which classification the feature point belongs to,and then all the feature points under the classification are traversed,and finally the matching feature points are found,which improves the efficiency of the algorithm.(3)In the process of matching point rejection,ORB-SLAM technology uses the RANSAC algorithm to randomly select the feature points for model fitting,and the fitting results are more susceptible to the initial value.In this paper,an improved RANSAC algorithm is proposed to record the hamming distance between the matching point and its pair during the matching process;the matching points are sorted from small to large according to the hamming distance,and extracts the first n match points to carry out the RANSAC operation,thus ensuring the quality of the initial data.(4)The improved feature extraction and matching algorithm is applied to the ORB-SLAM composition process,and the Open data set is used for experiments.Experiments show that,compared with the initial ORB-SLAM algorithm,the improved ORB-SLAM is well improved in terms of map matching speed and map building sharpness.