In this paper, we address the problem of wildlife recognition for road accident prevention, where a rate of 63.17% per year of road accidents is noted in Senegal. Given that the movement of animals is unpredictable in spite of road signs, this constitutes a handicap in preventing wild animals that may cause an accident. The solution proposed in this paper allows real-time detection of wild animals crossing roads, especially in non-built-up areas. It is based on computer vision using deep learning with the Yolov4 approach which allowed the categorisation of the three types of wild animals chosen in this paper: cows, donkeys and goats. The choice of these three types of animals is justified by the fact that most wildlife-related road accidents are caused by these types of animals. To achieve this, we first collected a set of images of the three types of animals. These collected images are sorted before the model is created. The evaluation of the model was carried out using test images and also videos taken on the Niague road, more precisely in the suburbs of the Senegalese capital, with accuracy rates of 94, 32%, 98.85 and 99.96%, respectively, for cows, goats and donkeys.
Automatic Recognition of Wild Animals for Road Accident Prevention Using Deep Learning with Yolov4
Signals, Communication Technology
International Conference on Signal Processing and Information Communications ; 2023 ; Singapore, Singapore February 25, 2023 - February 27, 2023
6th International Conference on Signal Processing and Information Communications ; Chapter : 1 ; 1-11
2023-11-11
11 pages
Article/Chapter (Book)
Electronic Resource
English