Machine Learning for Real-Time Traffic Incident Detection

Abstract

As more vehicles and roadways are being used, innovative traffic management systems are needed to control, regulate, and improve urban mobility. Among the currently existing traffic management systems, real-time traffic incident detection has become a proactive research area because road traffic incidents significantly affect time saving, emissions reduction, and road safety when moving people or transporting goods. To restore traffic to its pre-incident flow with minimal impact, incident detection should occur within minutes, simultaneously comply with the principle in WSN-based transportation systems, and confirm the vehicle type within large-scale transportation networks on urban roads.

Recently, due to the increase in traffic complexity and transportation requirements, the WSN-based transportation systems have evolved, supporting a set of functions from large-scale medium-level transportation consisting of trips between transportation zones and medium geographical scales, up to micro-level transportation focusing on trips between cells. The transitions of traffic scenarios from the macro to micro level have made the need for high-performance, automated, and intelligent systems even greater. Consequently, an advanced transportation support system, i.e., a suitable ICT, is needed to achieve smart city goals and manage operational traffic efficiently. One way to improve the traffic management system is to enhance the performance of incident detection in real time in the transportation system. Given the limitations of poor system performance, low detection accuracy, narrow adaptability, and complexity of the urban transportation stage, this paper discusses the need for various applications of real-time traffic incident detection.