Demonstration of a Drone-Based Track Safety Inspection System Using Asset-Based Change Detection

Today’s prevailing methods of visual track inspection tend to be expensive, disruptive to operations, and have potential to be less thorough than preferred. To help address these issues, machine vision technology emerged in the rail sector over a decade ago; however, the technology has not achieved its expected potential due to the uncontrolled nature of rail environments. Recent findings have shown that image-based change detection is able to identify many relevant track conditions while maintaining a detection probability near 100 percent in uncontrolled rail environments. The use of change detection transforms the rail sector machine vision problem into a relevant/non-relevant determination. Findings indicate that this reformulated problem is expected to be handled robustly and cost-effectively by machine learning. The proposed research intends to demonstrate a prototype Drone-Based Track Safety Inspection System in combination with AI-assisted change detection to automatically identify relevant track conditions. If successful, the technology is expected to reduce the cost of visual track inspection with simultaneous improvements in effectiveness, equating to increased operational safety for trains.