Britain’s 20,000-mile rail network requires reliable monitoring of lineside vegetation. Together with the UK Centre for Ecology & Hydrology (UKCEH), Keen AI has developed an innovative solution to remotely monitor biodiversity using a high resolution camera system.
Hamzah Reta from Keen AI shares his experiences having worked on the project and plans for the future to further develop the camera system.
What Did You Do?
The project was undertaken as part of Network Rail’s Biodiversity Action Plan and it was our task to develop a camera system to help monitor flora along the rail-side. We developed a solution that can be mounted to the front of a train and captures high resolution imagery of lineside vegetation while the train is in transit.
We worked closely with suppliers to make amendments to an existing camera housing box so it was able to accommodate the new camera system. With the help of CAD drawings for the redesign process, our new design made sure of a securely mounted camera system and a side-facing window, to allow us to capture imagery whilst the camera is facing perpendicular to the track and motion of the train.
The camera system consisted of a number of components:
- Mounting for a variety of cameras we were trialling
- A small compute device to control the camera and receive remote commands
- A GPS logger, kept in the train cabin to accurately log the image location within 10 meters
- A power bank to power the camera and compute device
We were able to control the system remotely by sending commands to the compute device.
What Challenges Did You and Your Team Have?
The biggest challenge we faced was building a system robust enough to function reliably when mounted on a train. Initial trials highlighted extra factors, such as spotty mobile and GPS coverage, which affected the reliability of the system when the train is in motion. The process of trialling different cameras on a train was time-consuming and therefore costly.
We also encountered GPS positioning drift and moved the GPS from the mounted camera housing into the train cabin. It is important to have an accurate GPS location of the images within 10 metres, as this accuracy enables Network Rail to identify where action is required to protect and manage lineside vegetation.
What Are You Planning To Do Next?
In the coming months, we will work further with Network Rail and UKCEH to improve the camera system as well as deploying AI to classify tree species and habitats. Currently, taking one image every 1 second means missing chunks of the rail-side and where species are. We are developing a solution with a global electronic shutter to increase our coverage of the lineside significantly.
Furthermore, we aim to create models to accurately identify five key tree species in the UK. Keen AI data scientist, Petar Gyurov, has already successfully trained one model to identify species of Ash in winter and we hope to train four further models to identify other tree species.