Pedestrian intention estimation from egocentric data

The prediction of pedestrian movements, here referred as intention, is important to improve the autonomous vehicles' safety systems. Recognize it correctly helps to avoid traffic accidents, specially on low visibility areas, non-line-of-sight or hiding. This paper analyzes the use of movement information provided by an accelerometer carried by pedestrians. With real data extracted from a real experiment, It is proved that the use of simple classifiers can detect the crossing-or not-crossing intention over a street, before the pedestrian does action.