Research On GMPHD-SLAM Algorithm Based On Lidar

In recent years,in the field of intelligent robots,industrial unmanned inspection robots and mine blasting robots have become more and more widely used.SLAM technology,as one of the key technologies for robots to realize intelligence and autonomy,has quickly become the focus of attention in the field of intelligent robots.One of the problems that need to be solved in the SLAM problem is the data association problem between measurement data and map features.The SLAM algorithm based on data association and Kalman filter uses data association to solve this problem,and has achieved good results when the associated scene is not complicated.However,when there are many clutters in the environment and a large number of missed detections of sensors,the performance of SLAM algorithms based on data association and Kalman filter is prone to degradation,and problems such as overestimation,underestimation,and misestimation of map features appear.Random finite set is a new theory in the field of target tracking,which can overcome the difficulty of accurate data association in traditional target tracking methods in complex environments,which has prompted scholars to study SLAM algorithms based on random finite sets.The current SLAM algorithm based on random finite set is not perfect in many aspects,and there are many problems.For example,1)The existing random finite set SLAM algorithm deals with the integration of ordinary variables and set variables at the same time,or through RB decomposition Using particle sampling to fit the integral of ordinary variables,the calculation process is very complicated;2)The existing SLAM algorithm based on random finite set rarely considers the separation of moving and static targets,and it is difficult to solve the SLAM problem under the condition of dynamic targets;3)Based on random finite set The SLAM algorithm of Jiji has a short development history and lacks an algorithm test platform that is easy to reuse.In response to the above problems,this paper has carried out the research of SLAM algorithm under random finite set.The main work of this paper is as follows:(1)Under the framework of random finite set,for the traditional random finite set SLAM algorithm,it is necessary to process the integration of ordinary variables and set variables in the form of interactive iteration,or use particle sampling to fit the integration of ordinary variables through RB decomposition,resulting in the algorithm For the problem of complex calculation process,a GMPHD-SLAM algorithm for decomposition of robot pose and map feature density information is proposed.First,in the framework of Bayesian posterior estimation,the SLAM estimation equation of robot pose and map feature density information decomposition is derived;then PHD is used to approximate the density information of robot pose and map feature;finally,Gaussian mixture technology is further used to approximate PHD Finally,the PMD-GMPHD-SLAM algorithm that can be realized by the project is constructed.Experiments have verified the advantages of the algorithm in the presence of clutter and missed detection.(2)On the basis of work 1),for most current SLAM algorithms based on random finite sets,only static targets are considered.In an environment where dynamic targets exist,the performance of the algorithm will be degraded,and it is impossible to complete SLAM while completing dynamic targets.To identify and track the problem,a PMD-GMPHD-SLAM algorithm that separates moving and static targets is proposed.First,the separation of moving and static targets is completed by calculating the proposed measurement contribution index,and then the PMD-GMPHD-SLAM algorithm is used to complete the separation of robots and robots.Positioning and mapping of static targets,and finally tracking dynamic targets through GMPHD algorithm.The effectiveness of the algorithm is verified in the presence of moving targets,clutter and missing detection.(3)Based on the actual measurement platform of the lidar PMD-GMPD-SLAM system,the algorithm proposed in work 1)was tested and verified.First,a verification system was designed based on the proposed PMD-GMPHD-SLAM algorithm,a car equipped with lidar was built,a data acquisition and processing system was built,the on-site data debugging was completed,and the data was finally processed to complete the actual measurement verification of the algorithm.