Anomalous On-Freeway Driving Behavior Detection Based On Smart Mobile Devices

Traffic safety is of great importance and is tightly bound to domestic economy and social lives of the public. As the stem of the road network, freeway safety is doubtlessly a crucial part of overall traffic safety. Arbitrary lane changes and U-turns are considered major inducements of traffic accidents on freeways. Besides, hard brakes and dodges are often considered signs of existing traffic incidents. Discovering these anomalous driving behaviors and notifying stakeholders in time will make it possible to reduce the adversity of traffic incidents and even to avoid potential accidents.The proposed technique of anomalous driving behavior detection in this paper is based on smart mobile devices which are quite prevailing nowadays. Sensor data from accelerometer, gyroscope, rotation vector and GPS are collected, and further processed by fusion, frame transformation and filtering. Data are then segmented to form candidate driving events, from which many behavior features are extracted. Finally, rule-based classifier are trained to classify these candidate driving events based on the extracted features, hence detect primitive behaviors like turns and hard brakes. Composite behaviors like lane changes and dodges are detected by composing turn events into corresponding candidate driving events, and feature extracted as well as classified in the same way as primitive behaviors.A prototype system named DriveDroid is also designed and implemented based on the proposed technique, with which real world experiments are conducted. The results of experiments show that this driving behavior detection approach has a relatively high accuracy and is promising in real-world applications.