Research On Indoor SLAM Based On Improved ORB And RD Polar Algorithm

With the wide application of intelligent mobile robot in people's life,the autonomous navigation ability of robot is constantly improving.The key technology of autonomous navigation is simultaneous localization and map creation,which is SLAM technology.The depth camera represented by Kinect developed by Microsoft can not only obtain the color image of the environment,but also obtain the depth information corresponding to the color image,which speeds up the data processing process.The research of SLAM Based on depth camera is an important research direction in the field of robot autonomous navigation.In this paper,the traditional rgb-d SLAM method is improved on the basis of the original rgb-d SLAM method,aiming at the problems of high error rate and low efficiency of feature point matching between adjacent image frames,bad effect of error matching removal,large error between camera estimation trajectory and real trajectory,and so on.A rgb-d SLAM method with better accuracy,real-time and robustness is proposed.The specific research contents of this paper are as follows:First of all,the algorithm of orb feature extraction is studied.Aiming at the problems such as too high dimensionality of feature point descriptors,low accuracy of feature matching,and need to calculate the direction of descriptors,an improved orb algorithm is proposed.Four groups of images with different illumination,rotation angle,scale and illumination,scale and rotation are used to simulate the improved orb algorithm,Experiments show that the improves the correct matching rate by 5% to 15%,reduces the time consumption by 50%,improves the speed of image frame matching,and meets the requirements of real-time and accuracy.Second,for RANSAC algorithm can't deal with the problem of data estimation without model very well,and it's difficult to get the correct result when the proportion of outliers is very high;and GTM algorithm can't remove the problem of the wrong matching point pair when there are the same neighboring points between the wrong matching feature point pairs.In this paper,the correct matching point depth image is the same,and the local feature point pair is used According to the spatial consistency relationship between them,a new feature matching algorithm,RD polar(region depth polar constraint),is proposed based on the consistency of region adjacent points.Finally,three algorithms are simulated in the indoor environment.By processing the test data sets of different proportion of mismatches,it is proved that the other two algorithms have better accuracy and robustness.Finally,the rgb-d SLAM algorithm before and after the improvement is simulated on the standard data set of tum,and the estimated trajectory is evaluated and analyzed.It is verified that the real-time performance and accuracy of the improved algorithm are improved compared with the original algorithm.At the same time,the Kinect camera is applied to the mapping experiment of the laboratory environment,and the results are evaluated to verify the effectiveness and feasibility of the improved method.