Proposal of a Method for Wildlife-Vehicle Collisions Risk Assessment Based on Geographic Information Systems and Deep Learning

This work proposes a deep learning and GIS based workflow to assess the influence of highway barriers on wildlife collisions. Our work consists of using Convolutional Neural Networks to classify images extracted automatically from Google Street View to determine the type of barrier, and using geoprocessing tools to estimate parameters as barrier length and location. The method was applied in a real dataset, classifying correctly the barriers in the road-kill points with accuracy of 84.44%. Statistical tests were used to evaluate the influence of each type of barrier on the road-kills.