Research On Environment Perception Feature Extraction And Data Association Method Of SLAM

Localization is a key component in the autonomy study on mobile robot.In the process of navigation with inertial navigation and dead reckoning,errors will be accumulated to be divergence over time,which can usually lead to localization failure and make robot lose.However,simultaneous localization and mapping enable robot to correct estimated pose and reduce navigation error by matching characteristic environment description features which are extracting from the surrounding environment.SLAM makes it possible for mobile robot to autonomous navigation without prior environment information.In order to meet the requirements for high precision and fast speed,feature extraction and data association as necessary component of SLAM will be the main reseach aspects in this paper.Firstly,the overall SLAM framework is built,including coordinate system,kinematic model of mobile robot,environment map model,environmental feature model and sensor measurement model.Secondly,for the environment detecting information of 2D laser range finder,an effectve segment feature method is proposed.The SICK LMS200 laser range finder drived by designed interface program is used to collect perception information in structured indoor environment.For the initial segmentation,a breakpoint detector is used to separate the disconnected data clusters in raw data.Furthmore,a proposed dynamic threshold Splite-and-Merge is applied to each of the connected data clusters for the second segmentation.The classical least square is proposed based on the orthogonal least square method.It will extract linear parameters and parameter covariance from the segmentation region with lower computational complexity.In the end,endpoints' projections on line are choosed to become another two parameters of segment feature.Thirdly,for solving the contradiction between precision and computational efficiency during ICNN and JCBB data association procedure,a hybrid adaptive data association method based on the basic association rules is proposed.Working under the constraints of basic guidelines for data association,it sets the ICNN data association as the preliminary operation.When ICNN is evaluated and judeged to be unreliable,JCBB will be excuted for the second association which would improve efficiency and accuracy of association.Besides,considering the robot pose and sensor measurement range,the hybrid adaptive data association method defines the association area as a half circle to reduce calculation complexity.Experiment results that the proposed adaptive data association method based on association rules is a good choice to account for efficiency and accuracy at the same time.Finally,a segment-based EKF-SLAM is designed,supported by the extended kalman filtering theory.It merges the segment feature extraction method and hybrid adaptive data association method proposed in this paper.And the validity is verified by the experiment.