Motion classification of pedestrian walking behaviors on the sidewalk

Abstract

This paper proposes a motion classification algorithm of pedestrian walking behavior by fusing radar and vision sensors. Under the situation that the pedestrian is detected by both two heterogeneous sensors, it will be discussed how quickly the intention of pedestrian on the sidewalk, e.g., stop, crossing, and walking along the road, is determined. While most of previous researches use lateral position of pedestrian as a feature for classification, the velocity with respect to lane at the sidewalk is estimated and used as an additional feature. Support vector machine as a classifier is used for motion classification of walking behaviors. Finally, the proposed algorithm will be validated via driving test data.