Research And Implementation Of SLAM Positioning Algorithm In Unmanned Driving

As the development direction of the future automobile industry,unmanned vehicles have attracted more and more automobile and technology companies to research and develop.In order to drive in the environment completely autonomously,in addition to an accurate environment map,accurate positioning is also the basis for unmanned vehicles to navigation and path planning.Existing maps,such as raster maps,are not suitable for large environments due to their large storage space requirements and high related computational complexity.At present,SLAM based on particle filter is the main solution for automatically creating environmental maps,but the number of particles limits the accuracy of algorithm.When the classic GPS positioning system is easily shielded by obstacles such as trees,buildings,bridges and holes in the environment,it will cause signal loss or sudden increase of localization error.The instability of localization will affect the safety of unmanned vehicles.In summary,the main research work of this paper consists of three parts,as follows:(1)This paper research how to use the line segment feature map to represent information of the environment,and research the extraction method of line segment features.The existing line segment feature extraction algorithms include IEPF and Hough transformation?.The IEPF algorithm has two main disadvantages.First,the threshold for line segmentation is difficult to determine.If the threshold is too large,the problem of under segmentation will occur,and if the threshold is too small,the problem of over segmentation will occur.Second,IEPF is more sensitive to noise.Hough transform is computationally intensive and easy to extract similar features,at the same time,the line segment feature parameters extracted are not accurate enough.The line segment feature extraction method proposed in the dissertation is divided into three steps.First,the Hough transform is used to determine the line segment features existed and the points these line segmentspass through.Then the similar line segment features are merged by the method of deep searching the Hough ballot box.Finally,the least square method is used to fit the straight line to obtain the four parameters of line segment feature.The experimental results prove that the method proposed in this paper can effectively extract line segment features and has good anti-noise ability.(2)This paper research how to use SLAM to create a line segment feature map of the environment.In recent years,RBPF-SLAM has been the focus of current SLAM research.Traditional RBPFSLAM is not suitable for the creation of feature maps,and the number of particles directly affects the accuracy of the algorithm.However,more particles means more storage memory.And the amount of calculation increases.In the dissertation,based on the RBPF-SLAM algorithm steps,this article adds the step of constructing a local line segment feature map,at the same time,the map update step makes the local line segment feature map into the global feature map and implement the feature map creation.In the sampling step,the unmanned vehicle odometer data is used to estimate the pose and extract a local feature map near the pose as a template,and use the observation feature sub-map extracted from the sensor observation data to match in the template to obtain the precise pose.Then sample new particles around the pose.Experimental analysis shows that,compared with GMapping,the method this paper proposed can achieve the same accuracy when using fewer particles.(3)This part researches how to locate at the created line segment feature map based on the environment observation data of the sensor.In existing feature map localization studies,EKF based localization methods require data association and provide an accurate initial pose.The Markov-based localization methods cause the problem of error accumulation when the unmanned vehicle moves with a non-integer grid size during the time update period.To solve this problem,a time update model based on the continuous relationship of grid-reliability was proposed in the dissertation..First,the discrete relationship of grid-reliability will be continuous,then make the continuous surface do the same displacement as the unmanned vehicle,and then integrated on each grid to obtain a new reliability of each grid.Through the analysis of experimental results,the proposed method can effectively estimate the position information of unmanned vehicles.