Research Of SLAM And Path Planning Method For Indoor Search And Rescue Robot Based On RGBD

With the development of robot technology,search and rescue robots are more widely used in postwar and post-disaster scenes.Using search and rescue robots to perform search and rescue tasks can minimize the damage to search and rescue personnel caused by secondary disasters.Moreover,since search and rescue robots can work continuously,it is of great significance to improve search and rescue efficiency.Simultaneously Localization and Mapping(SLAM)and faces target for search and rescue robot path planning technology can effectively improve the degree of independence,reduce the artificial control personnel and make the search and rescue resources fully used.However,in the complex scene of collapse of indoor walls,tables and chairs after the disaster,how to further improve the real-time performance and positioning and mapping accuracy of search and rescue robot SLAM,and then carry out goal-oriented path planning an important research direction to improve the autonomy of search and rescue robots.This paper improves the existing efficient and open source ORB-SLAM2 algorithm to form a set of RGBD SLAM algorithm based on depth camera,which improves the real-time performance and positioning and mapping accuracy of SLAM under special conditions.Through the improvement of the existing path planning algorithm,our method makes the obtained path closer to the shortest path in the real and continuous environment,and improves search efficiency of path planning.Combining the RGBD SLAM algorithm and path planning algorithm proposed in this paper to improve the search and rescue efficiency of the search and rescue robot,the following results have been achieved:(1)A new key frame selection method is proposed.Through pair-to-pair matching of the current frame and the two key frames adjacent to the history,the number of feature matching pairs obtained after matching is compared and a repetitive scene detection method is designed.The key frames are selected in combination with the inter-frame motion time method,and redundant key frames caused by repeated camera motion are eliminated in the process of drawing construction.Through the use of TUM RGBD data sets and its own evaluation tools,we compared the key frame selection algorithm advanced by this article and the original algorithm.The experimental results show the improved key frame selection method reduces about 20% key frames compared with the original algorithm,reduces about 4.7% of the SLAM system running time.Additionally,the Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)of camera trajectory reduce about 7.2% and 7.0% respectively.These results indicate our approach can effectively improve the real-time performance and positioning and mapping accuracy of SLAM,which in turn can enhance the search efficiency of search and rescue robot.(2)A new method for loop closure detection is proposed.By making full use of the Depth information of the image and comparing the Depth information of the current frame and the selected frame after the loop closure,a loop closure detection method combining RGB information and Depth information is designed to eliminate the error loopback caused by similar scenes and avoid the serious impact of the error loopback on the localization and mapping effect of SLAM.To validated the effectiveness of the proposed new loopback detection method,we collected the self-made similar scene data set through the scene arrangement.Further,we evaluated the performance of our algorithm on the our collected dataset and TUM RGBD dataset using EVO evaluation tool.The experimental results show that compared with the original algorithm,this improved loopback detection method eliminates the errors produced by similar scenario loopback,and can still achieve recall rate 50% under 100% accuracy.Meanwhile,in the fr1-room sequence,the RMSE of this improved loopback detection method decreased by 24.1% and the maximum error decreased by 14%.In conclusion,our approach improves the loopback detection accuracy,and the precision of positioning and mapping of SLAM.(3)The SLAM mapping module is extended.The 3D point cloud generated by SLAM mapping is published through the topic in ROS,and then received by RVIZ.In RVIZ,the 3D point cloud map is converted into Octomap and 2D occupied raster map in real time,which reduces the memory consumption of search and rescue platform,and provides map prior information for subsequent path planning technology.(4)The path planning technique based on Lazy Theta* algorithm is improved.This paper introduces the principle and shortcomings of the direct search algorithm A*algorithm for planning the shortest path in the existing static road network,and analyzes the principles,advantages and disadvantages of the improved algorithms Theta* and Lazy Theta*.Aiming at the shortcomings of the optimal algorithm Lazy Theta*,an improved algorithm with weighting factor is proposed.The experimental results prove that the improved Lazy Theta* algorithm in this paper reduces the path distance obtained by the A* algorithm by 9.3% and the path search time by 49.4%,the improved Lazy Theta* algorithm reduces the path distance obtained by the original Lazy Theta*algorithm by 2.2% and the path search time by 8.4%,which shows that the improved path planning method in this paper reduces unnecessary inflection points in the path,the obtained path is better and the path search efficiency is improved.Meanwhile,through the improvement,the algorithm can load local maps for path planning,which lays a foundation for the fusion and use of SLAM mapping module in this paper.To sum up,this paper has carried out research on RGBD-based indoor search and rescue robot SLAM and path planning method.Aiming at the similar scene caused by the collapse of indoor walls,desks and chairs after the disaster,and aiming at meeting the needs of search and rescue tasks,a feasible 3d drawing and path planning system has been formed in the indoor scene.First of all,a new keyframe selection method is designed to eliminate redundant keyframes caused by repeated camera movement.Through the open data set,it is verified that this method can improve the real-time performance,positioning and mapping accuracy of SLAM.Secondly,a new loop-loop detection method is designed to eliminate the influence of error loopback on the mapping accuracy in similar scenes.Through the self-made similar scene data set and the public data set,it is verified that this algorithm can eliminate error loopback in similar scenes and improve the localization and mapping accuracy of SLAM.Additionally,the mapping module was extended to convert 3d point cloud map into Octoamp and 2D map in real time,saving memory consumption of the system and providing prior information for subsequent path planning technology.Then,the weighted factor is added to improve the Lazy Theta* algorithm,and the design experiment comparison verifies that the improved algorithm reduces unnecessary inflection points in the path,the obtained path is better and the path search efficiency is improved.Finally,the above algorithm is combined with search and rescue platform,and the effectiveness of this algorithm is proved through field verification.The results of this paper improve the real-time performance,localization and mapping accuracy of SLAM,and make the goal-oriented path planning more smooth,and the path search is more efficient,laying a foundation for improving the efficiency of search and rescue.