Identification of Traffic Accident Black Spots on Suburban Highways Based on Smartphone Sensors of Drivers

One of the important issues related to road safety is the continuous monitoring of road conditions with the aim of preserving and maintaining the quality of roads. Considering the increasing use of smart phones, a practical solution based on smart phone sensors is proposed in this article to control the safety status of roads. This solution includes the implementation of a centralized traffic information system that monitors the dynamic behavior of vehicles while moving on the roads and collects trip-related information for further processing. To evaluate the system, a 42-kilometer route on a highway in Iran was monitored. A total of 7 parameters comprising speed, three-dimensional instantaneous acceleration and acceleration changes were examined. An event classification approach was adopted to detect accident-black spots based on the pattern of those parameter changes. The classified dataset was trained and modeled using two types of neural network models namely, Multilayer Perceptron (MLP) and Radial Basis Function (RBF). These two neural networks models were trained, tested and validated using MATLAB software and the collected dataset. The predicted error rate was obtained for 700 samples for each output. The mean square error index for RBF and MLP neural networks was obtained as 0.0066 and 0.1399, respectively, indicating acceptable prediction accuracy.