Computer Vision-based Applications in Modern Cars for safety purposes: A Systematic Literature Review

Abstract

Human error, fatigue, and negligence cause the majority of road accidents. Modern automobiles are outfitted with advanced driver assistance systems (ADASs) to help drivers and other vehicle occupants improve safety, enforce the law, and provide comfort. The purpose of this paper is to identify research gaps by highlighting the challenges of computer vision-based application techniques in modern automobiles for safety purposes. This study will also highlight publicly available datasets that can be used for research purposes. As a guideline, the study uses the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA 2020) protocol. Our study drew on seventy sources of literature, fifty of which focused on modern car applications (Lane, Pedestrian, and Traffici sign detection) and 20 on publicly available datasets. Using search criteria, the literature was mined in Google Scholar and IEEE Explore. The boolean operators and keywords listed below were employed. The inclusion and exclusion criteria used in the study are detailed in Section II. To understand the research gaps between the presented applications and the availability of public datasets, a comparison analysis was performed. Deep learning techniques are more accurate and robust than traditional computer vision techniques, according to the results. The results also show that there are available public datasets. The study, however, was restricted to English papers, lane, pedestrian, and traffic sign applications. Other languages and applications could be future research topics.