Improving smartphone based collision avoidance by using pedestrian context information

Abstract

Pedestrians globally comprise 22 % of all road traffic deaths in 2013. Various approaches for reducing accident numbers have already been introduced and are still being researched. Most of these approaches have specific limitations, like requiring line of sight. To overcome these limitations, we propose the Wireless Seat Belt (WSB), a smartphone-based collision avoidance system for pedestrians. Unlike other systems, the WSB uses context information, obtained from a pedestrian's smartphone, not only as additional information but also for using the information to improve the collision detection accuracy. The WSB introduces independent, individual modules for recognizing the pedestrian's direction, position, and speed. We first evaluate the influence of the measurement errors of each module on the missed alarm probability in a typical urban collision scenario using a simulator. Then, the impact of using the pedestrian's context to decrease the missed alarm probability is evaluated. The evaluation is done using the example of a curb detection module. The curb detection is used to infer that the pedestrian has stepped onto the street to correct the pedestrian's position. The results show a decrease of the missed alarm probability by 46.5 % in the scenario considered.