A Review of Artificial Intelligence Techniques used for Urban Automatic Incident Detection Systems

Abstract

Over the past decades, tremendous research efforts have been proposed to deploy Automatic Incident Detection Systems (AIDSs) onto urban roads. As a result, new challenges to the research community have been introduced to enhance the AIDS performances. To overcome these challenges, and to improve the efficiency and safety of road traffic, attention has been drawn to use Artificial Intelligence (AI) techniques as a solution. However, choosing appropriate AI technologies to process real-time traffic data is an area that needs spotlighting. The aim of this paper is to discuss recent advancements in urban AIDSs based on AI techniques. We carried out a review for twelve previous AIDS study proposed since 2009. We compared these AIDSs in term of traffic data sources, input variables, and type of incident. Also, a comparison of simulators, techniques and the evaluation metrics has been used. The results indicated that hybrid AI techniques reported the best detection performances. In this context, fixed or fusion data sources were also given good results compared to others. This review shows reliable results of AI classification techniques applied in AIDS approaches. This finding urges to continue improve the performances of the proposed AI techniques. In future research, testing others AI techniques with smart technologies may increase the incident detection efficiency.