Research On UAV Localization And Mapping Method In Dynamic And Similar Environment

In recent years,UAV has been widely used in military and civil fields due to its mobility and flexibility.In an unknown environment,real-time perception of information and autonomous navigation is the key to the completion of other missions.Based on the above objectives,visual SLAM(Simultaneous Localization and Mapping),was applied.However,there are often dynamic objects in the real environment,which will cause interference to SLAM.In addition,there are often many similar areas in real complex environments,which affect the accuracy of loop detection.To solve the above problems,based on the traditional visual SLAM method,an improved method adapted to the dynamic similar environment is designed in this paper.The main work includes:First of all,based on the practical application background and project requirements,on the basis of basic visual SLAM system framework,system framework is presented,in the visual odometry to join the dynamic characteristics of the filter module,to eliminate the influence of dynamic objects in the environment,at the same time to the loop detection is improved,add pruning function to avoid the error of the loop.The UAV can stably locate and build map in the dynamic similar environment.Aiming at the problem of poor accuracy of localization and mapping in dynamic environment,this paper designs a visual odometry in dynamic environment.The dynamic feature point detection module is added to the traditional visual odometry to detect and segment the feature points extracted from the image frame,filter out the dynamic feature points,and only retain the static feature points for localization and mapping construction.Finally,the open data set is used to experiment.The results show that the proposed algorithm can effectively filter the feature points extracted from dynamic objects,and the average localization accuracy is improved by about 96%.In order to improve the accuracy of visual SLAM loop detection in the face of similar environment,the traditional loop detection algorithm is improved in this paper.The Bo W is used to calculate the similarity between images,and the databases are constructed to improve the computational efficiency.The filter function is used to screen and avoid the detection of wrong loop frames.Finally,experiment is also carried out using public data sets.The results show that in the face of similar environment,the proposed algorithm improves the accuracy by about 29% compared with the traditional algorithm when the recall rate is 80% in the face of similar environment.Finally,based on the quadrotor UAV platform,the experimental verification of the algorithm is completed in the actual dynamic similar environment.The results show that the localization error of the proposed method in dynamic environment is about 0.24d%,which meets the requirements of localization accuracy.Meanwhile,the influence of similar areas on the accuracy of loop detection can be avoided.The average processing time of the system for one image is 39.3ms,which meets the real-time requirement.The results of the project can be applied to the localization and mapping of the intelligent unmanned platform in the dynamic similar environment,which has certain practical value.