Traffic Warning System for Wildlife Road Crossing Accidents Using Artificial Intelligence
Ikenna D. Uwanuakwa, Ugochukwu G. Isienyi, John Bush Idoko, Shaban Ismael Albrka Ali
Abstract
Wildlife-vehicle collision (WVC) is a major problem associated with regions with high-density wildlife. Urban designers have in the past introduce overpasses, underpasses fence, reflectors, and sensors to aid safe wildlife road crossing, but these have not been able to reduce the wildlife-vehicle collision. This research focused on the automated warning system to vehicle users to minimise wildlife-vehicle collision which could integrate computer vision in the detection of features on the road together with the location-time information feed. The proposed system was trained using AlexNet, GoogelNet, ResNet-50, and VGG-16 algorithm on a deep convolutional neural network (CNN) using 20,964 images of 25 variables consisting of 21 animals and four different vehicle body type. The dataset was divided into training and validation set. The results show that CNN algorithms could identify objects on real-life traffic data with noise background at a reliable accuracy. The GoogelNet, ResNet-50, and VGG-16 model outputs were found to have a better prediction accuracy than the AlexNet model in detecting the object features on the traffic images.