CrowdSafe: Detecting extreme driving behaviors based on mobile crowdsensing

With the popularity of vehicles, high traffic accident frequency has become a serious social problem in many countries. Thereby, it is of great value to detect driving behaviors and forecast dangerous situations. Specifically, with the recent surge of smart phones, there have been researchers who attempt to deal with this issue based on smart phone sensing. However, these existing studies have neither considered the phone's relative positions in the vehicle nor the phone's placements. In this paper, we propose CrowdSafe, which leverages the aggregated power of passengers to enhance the detection of extreme driving behaviors in public transports. First, we propose a multi-sensor fusion approach that can automatically locate passengers in a vehicle. Second, we investigate the impact of different in-vehicle locations on the performance for different extreme driving behavior detection. Finally, group decision making strategies based on the Bayesian voting theory is proposed to deal with the situations when there are conflicts among the reports from different passengers. Experimental results show that passenger positions and ways of carrying mobile phones have significant influence on the detection of extreme driving behaviors, and the improved voting method can achieve an accuracy of about 90%.