A Railway Simulation Framework for Key Scenarios Construction with Motion Distortion Correction of Virtual LiDAR Sensors

The simulated method has been demonstrated as the effective tool for providing a large amount of labeled data for applications of deep learning model in self-driving, robot control, and train perception systems. In the train domain, the simulated method is demonstrated as adoptable tool for the generation of test datasets of algorithms in railway, for the reason that it is strictly restricted in the realistic railway infrastructure to obtain information from kinds of sensors. However, the key railway features and the extreme conditions are not emphasized in current researches, which are of great significance in train perception system algorithm validation for emergency. In this paper, we propose a simulation framework which emphasizes the key railway features and the extreme conditions in railway scenarios. Moreover, the LiDAR correcting model is applied in our method to produce data more similar to the physical counterparts. The foreign object intrusion is simulated in our framework, which is highly similar to the realistic both in scenario and sensor data, revealing the validity and availability of our method.