Prediction of Pedestrian Crosswalk at Distraction Reflecting Walking Characteristics

Abstract

This paper proposed a method to predict crosswalk of pedestrians at distraction, reflecting walking characteristics. The method uses the acceleration and the angular velocity data before they cross roads. In the method, sensors are attached on the pedestrian’s head, both shoulder, right wrist, and waist. From the acquired acceleration and the angular velocity data of five body parts, the method acquires pedestrian’s walking behaviors as waveform data. According to our experiments, when a pedestrian is distracted, the movement of the right wrist and the waist changes. On the other hand, when he try to cross a road, some changes appear in the head and the both shoulder. Acquiring changes in these five body parts as walking characteristics, the method judges whether a pedestrian is distracted when he tries to cross a road. A machine Learning method, Random Forest, applied to the method has been confirmed to predict distracted pedestrians at crossing a road with about 75% accuracy