Research On RatSLAM Closed-loop Detection Algorithm Based On Improved Co-HOG

Mobile robot in the unknown complex environment,through the sensor accurately and quickly update its posture information and actively explore the environment,and finally build the environment experience map,the method is also known as Mobile robot SLAM(Simultaneous Localization and Mapping)algorithm.SLAM algorithms often perform well in a small and static environment,but the sensor of the mobile robot has errors,and the mobile robot will also have errors in the process of moving.With the increasing and changing environment,if the mobile robot only relies on the current environmental information to update its posture,these errors will accumulate and eventually lead to the failure of mobile robot maps.The closed loop detection of SLAM system can eliminate the accumulated error through algorithm convergence.Closed-loop detection algorithm is one of the important parts of SLAM system.A good closed-loop detection algorithm needs to meet the requirements of accuracy,real-time and robustness.The work content of this paper is as follows:The Co-HOG algorithm is improved.On the original Co-hog closed-loop detection algorithm,the common HOG descriptor is improved to GDF-HOG descriptor with Gaussian derivative filter to provide more prominent texture and intensity information.Based on the similarity measurement of closed-loop detection,the GDF-HOG global rough matching was added to reduce the number of frames in the visual template and improve the matching speed of similarity measurement.After the similarity measurement,a Region of Interest(ROI)position matching was added to improve the accuracy of the final closed-loop.This paper algorithm and other algorithms of the same type in the Gardens point and so on four different emphasis on public data sets on the experiment,through the image characteristics,matching accuracy and recall rate,unit algorithm in computing performance,and many other experimental data analysis,the results show that the improved algorithm to extract the characteristics of the texture stronger,finally matching accuracy is higher,The running time of the algorithm is basically unchanged,and the comprehensive unit calculation performance of the algorithm in this paper is higher.The improved Co-HOG algorithm was fused with the RatSLAM algorithm,and the original closed-loop detection link of the RatSLAM algorithm was improved.The RatSLAM algorithm of the improved Co-HOG was fused with the ORBSLAM2 algorithm and the original RatSLAM algorithm for experiments on KITTI public data set.The simulation results show that the proposed algorithm has higher accuracy in closed-loop detection and matching,smaller displacement error and relative rotation error,and the final trajectory graph error is smaller than the other two algorithms.The hardware experimental platform of Turtle Blot3 Burger mobile robot was also built in this paper.The RatSLAM algorithm of improved Co-HOG was combined with the visual mileage calculation method of mobile robot to test in the real laboratory environment.The experience map generated by the algorithm in this paper and the visual mileage calculation method was compared.In the absence of closed-loop detection and correction,the path trajectory of mobile robot will gradually shift,but the algorithm in this paper can perform closed-loop detection and correction of the trajectory,indicating that the closed-loop detection ability of the algorithm in this paper is good and can adapt to different scenarios.