Critical areas detection and vehicle speed estimation system towards intersection-related driving behavior analysis

A large number of serious traffic accidents occur at intersections due to the unsafe driving behaviors. In this paper, we propose a smartphone-based system to provide important information for driving behavior analysis at intersections. Our proposed system consists of two parts: (1) a deep convolutional neural network based model to detect traffic lights, crosswalks, and stop lines. (2) a Long Short-Term Memory (LSTM) neural network based model to estimate vehicle speed using accelerometer and gyroscope embedded in the smartphone. Important objects detection in traffic scenes and real time vehicle speed estimation are crucial for driving behavior analysis. We performed a thorough evaluation of our system, including analysis of the effectiveness of the proposed algorithm itself and the comparison with other methods as well. Our experiments exhibit the robustness of our system in various traffic scenarios.