A Study Of Pole-like Traffic Facilities Extraction And Classification Based On Moblie Laser Point Clouds

Pole-like traffic facilities are important road basis infrastructures.The rapid acquisition and updating of pole-like traffic facility information is of great significance to the safety of roads.Approaches to extract pole-like traffic facilities like manual measurement or semi-automatic extraction through images are lack of efficiency as well as low degree of automation,and they fail to meet the requirements for rapid extraction and updating of pole-like traffic facility information.High-precision pole-like traffic facility information such as position,inclination,orientation,and attributes plays an important role in road inventory,autonomous driving,and driving assistance.The mobile laser scanning system is one of the main methods for acquiring three-dimensional information on roads and surrounding environment.It can continuously and automatically collect spatial information and spectral information of target objects in the road environment,and is suitable for rapid extraction of pole-like traffic facilities.On the one hand,it has fast surveying speed,high measurement accuracy,and rich information,which is conducive to the rapid extraction and update of the pole-like traffic facilities;on the other hand,it obtains inconsistent density,high data volume,redundancy of mobile laser point clouds as well as occlusion.Those problems have challenged the rapid,accurate,and robust extraction of pole-like traffic facility information from the mobile laser point cloud.To solve the above problems,this paper studies the method of extracting pole-like traffic facilities from mobile laser point cloud and further classifying them.The specific research contents are as follows:1.An object-based method of automatic extraction of pole-like traffic facilities is proposed.First,the original point cloud is preprocessed,and the point cloud is distributed to a two-dimensional grid.In each two-dimensional grid,the ground points are filtered according to the local minimum elevation,and then the non-ground points are Euclidean-distance clustered;after clustering The mixed point cloud are detected by the minimum bounding rectangle,then the iterative minimum cut algorithm is used to segment the mixed point cloud,and pole-like traffic facilities are able to be separated from the mixed point cloud;finally,according to prior knowledge and shape knowledge of pole-like traffic facilities,a filter is constructed to detect pole-like traffic facility clouds from all object point clouds.Two datasets were used to test the effectiveness of the proposed method.The experimental results show that the accuracy,completeness and F1 measure of this method all reach over 90%respectively.Experimental analysis shows that for different density of mobile laser point cloud data,this method also can achieve good results and has certain adaptability.Compared with previous methods,this method has better extraction effect.2.A method of classification of pole-like traffic facilities based on supervised classification is proposed.First,manually select pole-like traffic facilities and further label the training samples.According to the shape and geometric characteristics of the pole-like traffic facilities,two types of features,the shape features and the general geometric features,are extracted;then those features are set as inputs to random forest to train the classify model,after that the trained random forest classifier is used to classify the test samples.The effectiveness of the proposed method is tested on two datasets.The experimental results show that the overall accuracy of our method in the two datasets both reaches more than 95%.The experimental analysis shows that for different density point cloud data,the feature extracted in this paper still maintains high separability;combining two different types of features for classification can achieve higher and more stable classification accuracy than any single type feature.