The key to realize the technology of automobile unmanned driving is to extract the ground information accurately.At pres-ent,heuristic algorithm and point cloud segmentation are mainly used to realize the ground extraction.In order to improve the ground recognition accuracy and evaluate the efficiency of various machine learning algorithms,this paper uses three common machine learn-ing classification methods,KNN,decision tree and support vector machine,to identify and classify ground and non-ground points in 3D laser point cloud data.In the experiment,the 3D point cloud data set VMR-Oakland-v2 of urban scene with multiple feature markers is used for model training,and the trained model is used to classify and verify the 3D point cloud data of real collection.The results show that KNN and decision tree models have achieved good extraction results.From the perspective of model size and training time,the decision tree model occupies a large space,has low efficiency,and achieves good classification accuracy.
ground point cloud extraction3D laser scanningmachine learningclassification algorithm
万方数据
维普
