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

Using the opportunities offered by the increasing number of sensors embedded into transport infrastructures and the development of their digital twins, a French research team has developed a method aiming at managing animal-vehicle collisions. The goal is to map the collision risk between trains and ungulates (roe deer and wild boar) by deploying a camera trap network.

Led by Sylvain Moulherat and Léa Pautrel, from OïkoLab and TerrOïko, France, the study is published in the open-access journal Nature Conservation.

The proposed method starts by simulating the most probable movements of animals within and around an infrastructure using an ecological modelling software. This allows the assessment of where they are most likely to cross.

After identifying these collision hotspots, ecological modelling is used again to assist with the design of photo sensor deployment in the field. Various deployment scenarios are modelled to find the one whose predicted results are most consistent with the initial simulation.

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

Finally, the processed data are fed into an abundance model, which is another type of ecological model. It is used to estimate the probable density of animals in every part of a studied area using data collected at only a few points in that area. The result is a map showing the relative abundance of species and, therefore, the collision risk along an infrastructure.

This method was implemented on an actual section of railway in south-western France, but it can be applied to any type of transport infrastructure. It may be implemented not only on existing infrastructures but also during the conception phase of new ones (as part of the environmental impact assessment strategy).

Such a method paves the way for the integration of biodiversity-oriented monitoring systems into transport infrastructures and their digital twins. As sensors collect data continuously, it could be improved in the future to provide real-time driver information and produce dynamic adaptive maps that could be ultimately sent to autonomous vehicles.