Track inspections using drones and machine vision

The desire for continual safety improvements and increases in economic efficiency are the catalysts driving research into alternative approaches to conduct visual track inspections. This paper discusses the feasibility of using a small (under 7kg), unmanned, remotely piloted aircraft (drone) to capture track images for the automatic identification of track defects using machine vision technology in a specific and challenging geographic environment. The drone was equipped with a high-resolution camera and a precision GPS system, and followed the track geometry using predetermined waypoints. Images were taken at different heights above the track and at different times of the day (to account for shadows). The machine vision system used was trained to recognise missing sleeper clips from these track images. This paper presents the research results with a discussion on the machine vision approaches applied and camera specifications. Broader issues associated with using drone technology to conduct track inspections in desert regions are also discussed