Object Identification Using Two Stages Deep Learning as a Component of Train Collision Avoidance System

• Main Author: Galih Pranajati, Ihsansyah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42552
• Institution: Institut Teknologi Bandung

Train is now one of the modes of public transportation which is widely used by the citizens of Indonesia. The number of train passengers are increasing every year, this problem demands train service providers to issue additional trains. To overcome the congestion of railway traffic, a communication-based train control (CBTC) system, that uses efficient signalling is used. But this CBTC system has a weakness. This system can’t detect other objects that block the railroad tracks and can be potentially dangerous for the train. Therefore there is a need for additional collision avoidance system that can analyze hazardous conditions in the surrounding environment. In this study, an image processing system with stereo vision is used to detect conditions that happen in front of the train. The method used by the system is to compare objects in real time camera captured images with trained graphic model to determine objects accurately. The result of distance measurement using stereo vision shows error below 6% for objects tested on distance range used in the experiment. To improve the model performance in object detection, a two stage deep learning for object detection is used. On the dataset used in this experiment, the model with two stage shows mean Average Precision (mAP) improvement compared to the model with one stage from 0,7218 to 0,8006. This image processing system is applied to trains miniature. This image processing system can also be used to determine the distance of objects in front of it by analyzing the object position in the frame. Then the system can provide control signals to reduce train speed and avoid collisions. With this research, it is expected that image processing system can improve train ability in avoiding accidents if there are objects blocking the railroad tracks.