Intelligent System for Detection of Wild Animals Using HOG and CNN in Automobile Applications

Abstract

Animal Vehicle Collision, commonly called as roadkill, is an emerging threat to humans and wild animals with increasing fatalities every year. Amid Vehicular crashes, animal actions (i.e. deer) are unpredictable and erratic on roadways. This paper unveils a newer dimension for wild animals’ auto-detection during active nocturnal hours using thermal image processing over camera car mount in the vehicle. To implement effective hot spot and moving object detection, obtained radiometric images are transformed and processed by an intelligent system. This intelligent system extracts the features of the image and subsequently detects the existence of an object of interest (i.e. deer). The main technique to extract the features of wild animals is the Histogram of Oriented Gradient (HOG) transform. The features are detected by normalizing the radiometric image and then processed by finding the magnitude and gradient of a pixel. The extracted features are given as an input to the basic deep learning model, a one-dimensional convolutional neural network (1D-CNN), where binary cross-entropy is used to detect the existence of the object. This intelligent system has been tested on a set of real scenarios and gives approximately 91% accuracy in the correct detection of the wild animals on roadsides from the city of San Antonio, TX, in the USA.