Subway Double-Door Anti-pinch Based on RGBD Binary Classification Network

The safe operation of the subway has great significance for the daily order of large cities. Under special circumstances, passengers and various objects may get caught between the subway train doors and the platform screen doors, which poses a safety hazard. We propose a method based on an RGBD dataset and a convolutional neural network binary classification model. By extracting features and using local features for normal and abnormal classification, we can detect in real-time whether there are people or objects caught between the subway double doors. Four sets of experiments were conducted using two classification models, with both RGBD and RGB datasets loaded, to demonstrate the advantages of using an RGBD dataset in improving accuracy. We also found a lightweight and high-accuracy model suitable for this application scenario to be run on edge devices, solving practical problems. By detecting foreign objects in the double-door gap, subway door anti-pinch measures can prevent passengers from getting caught between the subway train doors and the platform screen doors, ensuring passenger safety. This enhances the safety and reliability of the urban subway system, providing assurance for urban development.