Comparison of Image segmentation, HOG and CNN Techniques for the Animal Detection using Thermography Images in Automobile Applications

Abstract

Animal Vehicle Collision is an inviolability concern that comes with the cost of both humankind and animals. It has popularly resulted in millions of deer-vehicle collisions claims and fatalities. The only way to prevent the above-saddened statics is to drive wildlife safely away from roadways due to morbidity and injuries. This paper undrapes the optimal comparative study between edge-based image segmentation and CNN-HOG for self-acting animal detection. As the fatal crashes peaks during night-time, night vision image detection is focused on this paper with the mounted camera in the vehicle. Edge-based image segmentation is applied to the intelligent animal detection system to demonstrate the prowess of animal detection. The intelligent system processes thermographic images and feature extractions used for the object existence prediction. Deer is the overly populated animal and most commonly spotted animal used as the subject of detection in this research. The animal detection is done using the Histogram of Oriented Gradient (HOG) transform, whereas optimization is demonstrated using image segmentation. Image segmentation helps in precise animal detection by extending the continuity of the images, which is crucial for image processing during detection. The results vividly conclude the contribution of image segmentation accuracy to the existing HOG-based intelligent system with 91% accuracy using the wide roadsides of San Antonio, TX, in the USA.