Pedestrians’ Intention Recognition Method using Hidden Semi-Markov Model: The Case of Crossing the Crosswalk

Abstract

It is very important to ensure that elder people can perform safe outdoor activities, especially crossing the crosswalk. In this paper, we propose a novel system that can recognize intentions of the elder pedestrians in the vicinity of traffic lights to support the safe crossing. In order to recognize the intention, we applied Hidden Semi-Markov Model (HSMM), which is the most widely adopted method in this field of research. Our system consists of three functions: spatial context identification, HSMM-based learning, and intention recognition. To implement our system, we used GPS data collected from sensors embedded in the elder pedestrians’ smartphone, traffic lights data collected through Open API, and pre-classified activity data for activity learning. In the experimental section, to evaluate the performance of our system, we conducted experiments to find optimum k of k-prototype clustering and to determine the number of hidden states. The key contribution of this paper is to recognize the intentions from the pedestrians’ point of view for the safety of the pedestrians, not the intention of the driver for safe driving of the car.