Development of Animal Detection System for Traffical Collision Avoidance

One of the main causes of road accidents in Malaysia is animal-vehicle collisions. This is mainly due to the rapid modernization of the country which affects animal habitat which increases the chance for animals to be caught in roads and highways across the country. Thus, this project is conducted to study the performance of an animal detection system to help solve and reduce AVCs. This project is driven by the fact that as of now, Malaysia does not have a specific system designed to avoid AVCs. Therefore, hopefully the implementation of an animal detection system can help reduce or probably solve the case of AVCs. In this project, a hardware system comprised of a Raspberry Pi 3, used as a microcontroller, and a webcam is used to implement a system to detect animals. The webcam is used to stream live video footages, where the live video feed will then be filtered through an image processing system to detect the presence of any animal. Any image detected in the camera will then be analyzed through a real time deep learning object detection system. This system will utilize a pre-trained Convolutional Neural Network classification system to detect our object of interest. The classification model is pretrained and stored in a dataset which will be used for the detection system. Any image which contains an animal throughout the video feed will be detected and shown in the camera and the image will then be captured and the images will then be saved. The result of the detection can be seen in the live video footage as it will display the type of animal which is detected in real time.