Research Of Recognition And Classification Algorithm Of 3D Point Cloud Of Railway Scene

With the construction and development of high speed railway, the railway transportation with the characteristic of long distance, high speed and high carrying capability has become the most important and indispensable basic means of transport in our country. The safety of railway is getting more and more attention. Objects intruding the railway clearance, such as, the abrupt natural disasters, pedestrians crossing the tracks, vehicles through the level crossing and other intrusions into the railway clearance, do great harm to the railway safety. In the paper, an algorithm for 3D point cloud recognition and classification in railway scene was researched. It is more accurate and intuitive than 2D image. It has important and practical significance to guarantee the safety of railway transportation.In the paper, the overall plan of 3D point cloud recognition and classification algorithm Is proposed. It is divided into three steps:point cloud acquisition, point cloud segmentation, point cloud classification and recognition. Firstly, a 3D point cloud acquisition method is designed based on 2D radar. It is obtained with the pitch of 2D radar driven by a digital servo.3D point cloud of different regional scope was realized by controlling the radar’s horizontal scanning range and the pitch range. Point cloud of different density can also be given by parameter of scanning intervals as needed. Secondly, the points beyond the user’s setting range are removed from the 3D point cloud by a pass filter, and the sparse noise points are removed using statistical outlier filter. The quality of point cloud was improved by these two filters. By comparing the experiment result of different segmentation algorithms, the segmentation algorithm based on region growing is chosen as the most effective segmentation method. After segmented with this method, point cloud clusters of different single objects are setup. It was verified that point cloud of same object has similar Viewpoint Feature Histogram (VFH) characters. In the paper, the VFH feature library of different objects was built. The characteristics matching with the cloud point to be identified was searched by the kd-tree nearest neighbor searching algorithm in the feature library. Thus the recognition and classification algorithm of 3D point cloud was realized.The field experiment of railway scene shows that the 3D point cloud acquisition, segmentation, recognition and classification method can correctly detect the foreign objects intruding into the railway clearance.