A Train Identification Method Based on Sparse Point Clouds Scan Dataset

When monitoring the railway environment through trackside equipment, lidar data could be affected by number of factors: such as the equipment installation location, the different shapes of freight and passenger trains, the low reflectivity of train windows and others. The lidar point cloud data is always sparse and irregular, thus the train is usually identified incorrectly. To solve this problem, this paper proposes a train identification method based on point cloud projection to filter the lidar data. In this method, the groundpoints was segmented by the algorithm of patchwork++, the point of track was extracted based on the geometric position analysis. Then the point cloud is rasterized into x and y directions, and use the connected component analysis to calculate the area of missing-groudpoints in the current frame. There are two conditions will be used to judge whether the object is a train: the positional relation of the non-ground points and the track, the difference in the size of missing part between the current frame and template. At the same time, the time taken for train to pass through the lidar's view also is considered as an auxiliary judgment condition. The experimental results showed that this algorithm has promise application values with its short consum ing-time and high recognition accuracy. And it has been successfully applied in the trackside detection equipment in the specific department of railway maintains.