Level Crossing CCTV Eye Tracking Study
Network Rail
Background
There have been a number of fatal accidents at remotely controlled level crossings. These are crossings where CCTV images are relayed to a signal box where a signaller makes an assessment about whether the crossing is ‘clear' or not. Any obstruction on the crossing resulting in the approaching train being stopped before arrival at that crossing.
The final check undertaken by the signaller is a visual one using the CCTV monitor. Well documented incidents have occurred where crossings have been judged to be ‘clear' when an obstruction could be clearly seen from the screen shot.
Project
This research programme examined whether eye-tracking techniques could be used to improve operator performance in the visual search task. It looked for the underlying reasons for these failures and this information then used to improve the training of operators.
Implementation
A series of video records were taken of various ‘obstructions' on a variety of level crossings. These included wheelchairs, child-sized mannekins and vehicles. Records were also made of the crossing being actually clear.
The video records were edited into a randomized sequence which could be presented to subjects. Following calibration of the eye-tracking equipment subjects were shown crossings with and without targets on them and asked to judge whether the crossing was ‘clear' or not. The responses were then compared with the actual status of the crossing as presented to the subject. Two groups of subjects were used railway staff and naïve subjects.
Outcome
Significant differences in patterns of eye-movement were identified and it was clear that there was no universal strategy being adopted for the scanning of the image for targets. Correlations between eye-movement and target detection were not established though the pattern of eye-movement when targets were missed could be analysed.
There was some evidence to suggest that the location of the target on the level crossing, visual distractions in the actual image and a lack of thoroughness could all contribute to failed detections.
For more information
Network Rail