As presented at the 2021 ICOET, Arcadis developed a computer vision model to automatically identify wildlife species from camera trap images to gather insights on wildlife usage of transportation structures. In 2023, Arcadis proposes to join a panel discussion to illustrate how data science and artificial intelligence (AI) can be utilized for both asset management and wildlife protection.
Before a computer vision model for wildlife image recognition can be applied effectively, wildlife vehicle collisions (WVCs) must first be identified. Wildlife vehicle collisions are a serious concern for wildlife protection, but also for driver safety. In the Netherlands WVCs account for wildlife mortality, property damage/loss, human injury and less often driver fatality. In order to prevent this, Arcadis developed a machine learning (ML) model that predict possible locations of WVCs (i.e., hot spots). Logically, one of the main parameters are the road kills but this is ‘just’ a result. Using the model, users can predict WVC locations based on a wide variety of parameters including natural environment, species, time of the year, historical accidents, and manmade protection measures, such as barriers or wildlife crossings. Users can predict where WVCs are most likely to occur – or least the highest risks areas - to inform future management decision.
Our data scientists and transportation engineers have also found that the impact of road kills can extend beyond the WVC itself. Road kills and animal carcasses can also have a significant impact on road surface asphalt due to a chemical reaction between animal blood and bitumen in the asphalt. In order to deal with the impact of wildlife mortalities and other road distresses we’ve applied state-of-the-art computer vision models to automatically detect any road distress on street imagery, ranging from cracks to raveling and potholes. Utilizing neural networks that are fed with decades of manual visual inspection data, the AI can complete faster, more consistent, and more objective road surface inspections and in a safe and reliable manner. The model can also position the distress in geographical space, meaning that the class, shape, size and severity can be calculated, resulting in a risk classification of the road distress itself and the road segment, based on US standards such ASTM and PASER. The technology can also be applied to do road surface analysis on concrete roads, traffic signs, markings and – yes – wildlife.
AI/ML
Computer Vision
Data Science