Research On Loop Closing Based On Three-Dimensional Laser Point Cloud In Indoor Environment

Simultaneously Localization and Mapping is the key technology for mobile robot to realize autonomous navigation.The robot obtains the data in the unknown environment through their own position estimates and sensors,while creating the environment map and the robot autonomous positioning and navigation.Because the SLAM process is affected by sensor error and environmental conditions,loop closure detection is used to determine whether the robot has returned to the area that has been explored,to provide conditions for global correction,and to improve the quality of the map building.Because visual data is sensitive to light conditions,and it is difficult to extract effective features in the weak feature area,the laser point cloud data can make up for these two problems,three-dimensional laser point cloud data is used for loop closure detection research.Considering the idea of closed-loop detection as a classification problem,In this paper a three-dimensional laser point cloud loop closure detection algorithm based on machine learning was proposed.(1)Based on the study of the existing geometric characteristics of three-dimensional laser point cloud,the geometric statistical characteristics of the point cloud are integrated,and the calculation formula of each feature component is deduced.And a range histogram is introduced to construct the features.Then the laser point cloud feature vector with rotation invariance is constructed.Samples for classifier training and prediction are generated based on the features.Using C + + in the Visual Studio platform to write feature extraction code,and Matlab is used to complete the feature component correlation analysis experiment.(2)A loop closure detection algorithm based on indoor three-dimensional laser point cloud is proposed,considering the idea of closed-loop detection as a classification problem.According to the idea of AdaBoost algorithm,the final classifier is linear combined by a set of weak classifier.The decision tree is used as the weak classifier,and the decision tree is constructed by CART algorithm.The samples are generated according to the three-dimensional laser point cloud features to conduct the classifier training and testing experiment.(3)Experiments are carried out to demonstrate the efficiency of the algorithm.Open source dataset is used to demonstrate the efficiency of the algorithm.First,the correlation among the features components should be analyzed to remove the features components with high correlations with other features components.The optimal number of iterations is selected.Then,the samples are generated according to the characteristics after the treatment.The real data is used to verify the algorithm with the optimal number of iterations,and compared with the current three-dimensional laser point cloud loop closure detection algorithm.The detect rate,wrong rate and lost rate are used to evaluate the results of the experiments.In addition,the time complexity of the algorithm is analyzed and the direction of improvement is proposed to achieve the purpose of real-time closed-loop detection.Loop closure detection,as an indispensable part of the SLAM process,plays an important role in the quality of the final plot,providing control for the global correction of the map.In this paper,through system research of the three-dimensional laser point cloud loop closure detection,hoping for playing a role in real-time loop closure detection research.