Railroad tracks need to be periodically inspected and monitored to ensuresafe transportation. Automated track inspection using computer vision andpattern recognition methods have recently shown the potential to improve safetyby allowing for more frequent inspections while reducing human errors.Achieving full automation is still very challenging due to the number ofdifferent possible failure modes as well as the broad range of image variationsthat can potentially trigger false alarms. Also, the number of defectivecomponents is very small, so not many training examples are available for themachine to learn a robust anomaly detector. In this paper, we show thatdetection performance can be improved by combining multiple detectors within amulti-task learning framework. We show that this approach results in betteraccuracy in detecting defects on railway ties and fasteners.
展开▼
