TrackWarn: An AI-driven warning system for railway track workers

Authors:

Abstract
This contribution focuses on developing an AI-driven warning device to ensure the safety of railway track workers. Recent studies clearly show that track workers safety has become a major challenge for the railway industry despite many precautionary measures that are implemented. In this regard, many technological solutions have been proposed and developed to warn track workers of the approaching trains. However, the cost and complexity are the drawbacks of these systems. Therefore, we introduce TrackWarn, a low-cost portable smart gadget that detects the sounds of the approaching trains and provides a warning signal to track workers via a phone call. TrackWarn uses a state-of-art Convolutional Neural Network (CNN) that utilizes environmental sounds and spectrograms to classify if the train is approaching or not. This model achieves an average classification accuracy of 92.46%. With the help of Arduino Nano 33 BLE Sense micro controller, the whole system becomes very handy and potable. This paper addresses the design of the TrackWarn and the results obtained with respect to the various test cases. Further, the performance and communication challenges are also described in detail.