Mitigating animal-vehicle collisions with field sensors, artificial intelligence and ecological modelling

Taking advantage of the opportunities offered by the growing number of sensors embedded in traffic infrastructure and the development of their digital twins, a French research team has developed a method aimed at managing animal-vehicle collisions. The goal is to map the risk of collisions between trains and ungulates (roe deer and wild boars) by deploying a network of camera traps.

The study, led by Sylvain Moulherat and Léa Pautrel from OïkoLab and TerrOïko in France, is published in the open access journal nature conservation.

The proposed method first uses ecological modeling software to simulate the most likely movements of animals in and around infrastructure. This can assess where they are most likely to travel.

After identifying these collision hotspots, the ecological model was again used to assist in the design of on-site light sensor deployment. Model various deployment scenarios to find the one with predicted results that are most consistent with the initial simulation.

Once the sensors are deployed, the data collected (in this case, photos) is processed through artificial intelligence (deep learning) to detect and identify species near the infrastructure.

Finally, the processed data are fed into an abundance model, another ecological model. It is used to estimate the likely density of animals in each part of the study area using data collected at only a few points in the area. The result is a map showing the relative abundance of species and thus the risk of collision along the infrastructure.

The method was implemented on a real stretch of railway in southwest France, but it can be applied to any type of transport infrastructure. It can be implemented not only on existing infrastructure, but also during the ideation phase of new infrastructure (as part of an environmental impact assessment strategy).

This approach paves the way for the integration of biodiversity-oriented monitoring systems into transportation infrastructure and its digital twin. As the sensors continue to collect data, they can be improved in the future to provide immediate driver information and produce dynamic adaptive maps that can eventually be sent to autonomous vehicles.