Research On Methods Of Simultaneous Localization And Mapping (SLAM) Based On Correlation Filter

Nowadays with the development of machinery industry,robots are widely used in every field of agriculture,manufacturing industry,service industry,medical industry,and military industry.As more and more attention has been paid to topics like autonomous driving and UAV navigation,the demands for ability of robots to localizate and map an unknown environment,is getting overwhelming expected to an unprecedented extent.Thearfore,the reaserch on simultaneous localization and mapping(SLAM)has great practical significance.This thesis firstly introduces the history of SLAM,and analyzes why SLAM hits a bottleneck in recent years.Then discusses the feasibility of applying some target tracking methods to SLAM interdisciplinarily and the possibility brought by the upgrading of the hardware to build SLAM systems on new data resource.A SLAM simulation platform based on KITTI database is proposed,which is able to generate intuitive images and data for this study.The SLAM simulation platform will be used in comparative experiments,feasibility studies,accuracy analysis.Secondly,in the basis of this simulation platform,studies of SLAM using the point cloud data sampled by lidar in KITTI database are carried on.By noticing the fact that data between adjacent samples in SLAM have a feature of ignorable rotation,scaling and changes of light and shade,we propose a Simultaneous Localization and Mapping(SLAM)method on base of correlation filter.We design an algorithm to calculate every key parameters in the process,and prove the feasibility of this correlation filter SLAM method through experiments.Comparing with the other common methods,we validate the superiority of correlation filter SLAM.Finally,on the basis of the correlation filter SLAM method,we find the main factor that determines the time complexity of the process of correlation filter and suggest a multi-scale segmentation method.This multi-scale segmentation method of correlation filter passes the tests on the SLAM simulation platform,and shows high feasibility for SLAM problem.