Road Centerlines Extraction Based On Tensor Voting Using Airborne LiDAR Data

As an important traffic information,roads are closely related to human life and play a key role in the rapid development of cities.With the development of "smart city" and intelligent transportation,the automatic and effective access to road information is gradually becoming important.Compared with the traditional way of road information collection,airborne Li DAR data has the advantages of high accuracy and high efficiency,and has a broad prospect in road information feature recognition.However,for the problem of road centerlines detection in airborne Li DAR data,the algorithms proposed by domestic and foreign researchers have some shortcomings and defects,such as road type limitations,low efficiency of algorithm operation,and the noise interference problems that are difficult to solve.In this paper,a tensor voting algorithm was proposed to extract road centerlines,which determining the minimum scale factor and using muti-feature road saliency.Firstly,combined with the height,intensity and distance characteristics of 3D laser points,the road point cloud was extracted from the original points step by step using preprocessing algorithms and then road points were transformed into road intensity image;Besides,the minimum scale factor is used to determine the voting field corresponding to the road points.The polarity significance of the road is obtained by the undirected ball tensor voting,and the threshold is set to strip the edge points and noise points of the road firstly.Then the image is voted by the un-orientated ball tensor and the orientated tensor to get the linear significance of the road,extract the optimal road target,and finally refine to generate the accurate road centerlines using morphological thinning algorithm.In this paper,the synthetic data and three groups of airborne Li DAR data sets were used for comparative experiments to analyze the scale factor sensitivity,robust of the improved tensor voting algorithm,preprocessing results of the point clouds,polarity saliency map and curve saliency map of road points,accuracy of final road centerlines,the resolution value of 3D points converted into 2D image,and time complexity of tensor voting.The experimental results show that:(1)The running time of the tensor voting algorithm is directly proportional to the key scale factor,and voting based on scale factor determined by road width can improve the efficiency of the algorithm.At the same time,the algorithm can effectively detect the target roads from the noise background,and the average detection quality of the simulated images with different noise densities can reach more than 88%.(2)The multi-feature tensor voting algorithm based on the polarity and curve feature can extract the smooth target roads,gradually and effectively suppress the influnce of road noise.The detection quality of the road centerlines extracted from three airborne Li DAR data sets by the method in this paper is more than 85%.