Development of Computer Vision based Robust Approach for Joggled Fish Plate Detection in Drone Images
Aradhya Saini, Dharmendra Pratap Singh · 2018
Periodic inspections of the railroad tracks and their components are necessary to ensure safe transportation. Manual inspections involve the railroad personnel to either inspect the tracks by walk or ride a hi-rail vehicle at comparable speeds. However, manual operations cannot be carried out regularly in remote areas or terrorist affected areas. The unmanned vehicles such as drones are best suited in these cases. In addition, automated computer vision based approaches lead to more accurate as well as frequent condition monitoring of components. The joggled fish plates are track components which provide with emergency repairs to fractured rail sections while maintaining the steadiness and the continuity of the train. Their monitoring is essential for smooth passage of trains. Therefore, in this paper, we have described a novel template matching based approach for detection of joggled fish plates in the drone imagery. Image based statistics and Non-maximum suppression techniques have also been used to reduce the false detections. The specificity or the detection accuracy value achieved is 95%. The proposed approach is robust to noise and works well under various lighting and environmental conditions.