A Real Time Abysmal Activity Detection System Towards the Enhancement of Road Safety

Nowadays, road accidents are a serious worry particularly in the developed countries where the rules and regulations are not maintained properly. The majority of traffic accidents are caused by the drivers' error or hostile behavior against the drivers inside the vehicle. In a nutshell, any type of abysmal action occurring inside the vehicle can result in a traffic accident. If these behaviors can be detected in real time, unexpected accidents can be reduced in a significant manner. Due to the increasing frequency of road accidents over the years, it has attracted the researchers' curiosity, and they are attempting to mitigate the abysmal activities inside the vehicle. In this regard, the deep learning algorithms can present a real-time abysmal activity detection system for inside use in a vehicle with the goal of reducing road accidents and thus increasing the passengers' safety. In this investigation, a dataset of abysmal activity inside a vehicle has been created, consisting of 10400 diverse photos divided into 9 classes. To the best of our knowledge, it is the largest dataset in this regard. Then, a real-time activity detection model called You Only Look Once (Yolov4) has been successfully employed, which obtained an outstanding accuracy of 99.402%.