Real-Time Traffic Accident Detection for an Intelligent Mobility in Smart Cities

Traffic accidents are a major cause of death around the world. Traffic accidents lead to traffic congestion and occur frequently in urban scenarios. In this, traffic accidents are one of the most common causes of death all around the world. A quick response to traffic accidents is crucial to saving a life. With connected vehicle technology, vehicles generate data such as velocity, acceleration, position, etc. which can be periodical. These data can be collected using connected vehicle technology such as cellular, 802.11p, etc. In this chapter, with the growing popularity of smart cities, we focus on the detection of traffic accidents using connected vehicle data under various constraints. After accident detection, steps to send an ambulance, tow trucks, etc. can be taken to save lives and clear roads for traffic flow. The collected data is aggregated to detect accidents in a time- and resource-efficient manner. Here, two approaches are utilized: (i) time aggregation and (ii) position and time aggregation. In time aggregation, the vehicle data for 10 s is aggregated, whereas, in position and time aggregation, all vehicle data within the 50 m range for 10 s is aggregated. In this work, classification modeling algorithms such as support vector machine (SVM), linear and nonlinear kernels, and gradient tree boosting (GTB) are used to detect accidents. Various data features such as vehicle ID number, vehicle type, position, speed, accelerations, and lane change information of vehicles are considered to detect accidents on the road. Finally, comparative results are demonstrated in terms of accuracy, precision, recall, and F-score parameters to validate the performance of the models.