An Animal Detection and Collision Avoidance System Using Deep Learning

Abstract

All over the world, injuries and deaths of wildlife and humans are increasing day by day due to the huge road accidents. Thus, animal–vehicle collision (AVC) has been a significant threat for road safety including wildlife species. A mitigation measure needs to be taken to reduce the number of collisions between vehicles and wildlife animals for the road safety and conservation of wildlife. This paper proposes a novel animal detection and collision avoidance system using object detection technique. The proposed method considers neural network architecture like SSD and faster R-CNN for detection of animals. In this work, a new dataset is developed by considering 25 classes of various animals which contains 31,774 images. Then, an animal detection model based on SSD and faster R-CNN object detection is designed. The achievement of the proposed and existing method is evaluated by considering the criteria namely mean average precision (mAP) and detection speed. The mAP and detection speed of the proposed method are 80.5% at 100 fps and 82.11% at 10 fps for SSD and faster R-CNN, respectively.