Zircon were awarded funding from the RSSB, who play a major role in improving and regulating the safety of the railway industry, to undertake a feasibility study as part of the Remote Condition Monitoring Challenge. RSSB had identified ten unresolved challenges that prevent the full implementation of remote condition monitoring on the railway.
Zircon received funding to investigate a system that utilises the forward facing CCTV cameras on trains to detect unauthorised human presence within the boundary fence, whilst filtering out authorised human presence. Additionally, we also looked at whether the train CCTV feed could be used to identify the significant movement of the objects within the vicinity of the track that may adversely affect the passage of rolling stock or hinder the visibility of essential rail side infrastructures.
As a feasibility study, we were called on to examine a high number of variables and potential design problems. For example, the system would be expected to function accurately in a variety of lighting and weather conditions, distinguish the difference between rail workers in ‘high-vis’ clothing and members of the public who have strayed inside the boundary of the track, and do all this from a moving platform whilst providing accurate position information.
With regards to detecting human incursion, the focus of the study was on our ability to detect human presence in a multitude of different lighting conditions, similar to those experienced on the daily journey of trains, and the ability to identify and differentiate authorised personnel from unauthorised human presence. During testing we found that the probability of detecting people over a variety of contrast and lighting conditions was surprisingly high, and following the inclusion of pre-configurable ‘high-vis’ parameters the identification of authorised presence was achievable.
Unlike detecting human presence, the ability to detect the movement of objects was slightly more complex and unfamiliar. As there is no way to define what the system should be searching for, the techniques used in previous object detection projects we had done would not necessarily be suitable. Our solution for this problem was to generate an ‘interest map’ of the journey landscape in order to compare CCTV footage from train journeys in order to monitor and identify changes in landscape. Our tests found that generating the map itself was easy, however monitoring changes against the map was less so, but still possible.
Currently, the entirety of the study has been completed off-train without representative CCTV footage, and is awaiting the opportunity for a train mounted trial in order to try and identify real life problems. It is our hope that once this system has been proven, the technology could be utilised to forewarn of further issues which have an impact on the levels of safety and performance of the railway.