Machine Learning based Obstacle Detection for Avoiding Accidents on Railway Tracks

The wild life population has been endangered by animal collisions with railway tracks. Additionally, the railway infrastructure is also damaged by these accidents. Saving the larger animals, such as elephants, rhinos, and buffalo, is the primary objective of the proposed research study. Numerous research solutions exist, some of which simply rely on sensors and others only use software. However, this research work utilizes both approaches by incorporating Machine Learning (ML) models. Deep learning techniques have proven to be more effective in detecting animals that stand/walk on the track from the captured image. In the proposed method, ESP32 camera is fixed in the train to capture images of obstacles on the railway tracks like elephants, buffaloes, bicycles etc. The captured images are interfaced with Arduino board and the details are fed to Personal Computer (PC) in the train. YOLOv3 Convolutional Neural Network (CNN) pretrained on COCO dataset is used to perform obstacle detection and classification. An ultrasonic sensor fixed in the train also measures the distance between the train and obstacle. Alert message will also be sent to the engine driver and also the distance information gets displayed in Liquid Crystal Display (LCD). The proposed method is also used to prevent train-animal collisions. Finally, the experimental results prove that the YOLOv3 architecture gives better results in terms of average precision and confidence score.