Deer-Vehicle Collisions Prevention using Deep Learning Techniques

Animal-vehicle Collison is a serious problem in the modern world. Approximately 1.5 million deer-related car accidents, over 1 billion dollars in vehicle damage, and around 175-200 fatalities and 10, 000 injuries reported annually by Federal Highway Administration Research and Technology. The current anti-collision technology does not work well in preventing deer-vehicle collisions. The progress in cloud computing makes it possible to apply machine learning algorithm on vehicles that prevent this type of collisions. This research aims to reduce the Animal-vehicle Collison. In specific, the researchers propose algorithm can effectively prevent the collision between vehicles and deer. A dataset containing 4, 209 images of deer and vehicle in roadway to develop a Machine Learning algorithm based on Deep Neural Network. This algorithm can distinguish deer from different kinds of vehicles and other objects. In this paper, both traditional Machine Learning models like Support Vector Machine (SVM), Random Forest, and Deep Learning networks like Convolutional Neural Network (CNN) were used. The best model, a 16-layer CNN returns accuracy of 95.42% on testing data and 97.90% on validation data.