Detection and Classification of Animal Crossings on Roads Using IoT-Based WiFi Sensing

Road traffic accidents involving animals cause great health, environmental and monetary costs every year, specially on rural areas. Current animal detection systems suffer from either cost, scalability or accuracy issues, which prevent their effective use in a more extensive manner. To provide an accurate and cost-effective detection system, we employ WiFi sensing using low-cost IoT devices to train a modern deep learning model - the Transformer network. Our Transformer network detects road crossings with an accuracy of 97.1 percent and exhibits low false positive and false negative rates. In particular, our model accurately distinguishes vehicles from small and large animals, which enables scalable and economical accident prevention over large distances in rural roads. Finally, our model is on par with state-of-the-art predictive models, outperforming recent AutoML mechanisms and evidencing the suitability of the Transformer network for WiFi sensing-based tasks.

I. Introduction

Road traffic accidents account for more than 1.35 million deaths every year and are the leading killer of people aged 5–29 years [1]. A significant portion of these accidents is caused by collisions between drivers and wildlife, incurring great health, environmental and monetary costs to local com-munities and the natural ecosystem surrounding them. In the United States, it is estimated that animal-vehicle collisions cause 26,000 human injuries, 365 million vertebrate animal deaths and over 8 billion dollars in damages every year [2]. Over 89% of these accidents involving animals occur on rural, two-lane roads [2], which often have a limited budget for infrastructure due to their extensiveness and low amount of traffic compared to highways closer to big population centers. Building wildlife fencing and underpasses to avoid incidents over such an extensive network is often not eco-nomically feasible. An alternative method of reducing the likelihood of accidents is by using animal detection systems to monitor animals crossing the roadway and alert incoming drivers by lighting up warning signs along the road. Many technologies have been employed for this purpose, including LiDAR sensors [3] and imaging cameras [4], but they are not scalable and cost-effective enough for covering large distances. Fig. 1:

WiFi sensing layout for (a) animal/pedestrian crossing detection and (b) vehicle detection. Graphic elements from Freepik (by starline) and Flaticon (by Freepik and Good Ware).