Detection and identification of object on Tram Track (iOS Application)

Abstract

Even till today, trams are still a valid public transport in the Hong Kong Island. One of the biggest difference between trams and normal vehicles such as bus or van is that trams are required to run and stay on their own track. Purpose: Obstacles or trash might get stuck in the track from time to time for various reasons. In that case, it will be important to monitor the tram track and make sure trams can run safely on it. Currently, there is no known solutions for automatic detection system of objects on tram tracks in Hong Kong. Moreover, it is difficult and inefficient to have workers keeping their eyes on the track constantly. Therefore, I am developing an iOS application that can detect and identify objects on tram track automatically in this project. Method: I conducted this project with object detection algorithm YOLOv3 and implemented it to an iPhone with Swift and Xcode. I trained my own custom detection dataset which focused on detecting small objects that might get stuck in tram track and made comparisons of it with the default COCO dataset. Result: I built a prototype tram track model for testing my application and the results were satisfactory. The application could run in real-time and could detect small objects such as coins and pens that likely to get stuck in track. When objects were detected on track, there would be an alarm alerting the tram driver about it. I have tested my detection model and it is proved to have a mAP(mean average precision) of 68.69%. Conclusions: This project successfully detects and identifies foreign object on tram track in front of a tram in real time during tram operation using iOS application.