Crowd Tracking of Electric Wheelchair using RGB-D Camera with Median of Candidate Vectors Observer

Electric wheelchair in a crowded place must avoid many pedestrians and drive safely. In complex environments, the tracking algorithm of a pedestrian is sometimes more effective rather than automatic driving. However, a target pedestrian is lost unexpectedly in crowded places. In contrast, a robust tracking method has been studied to follow pedestrian crowds that naturally form in crowded places. Although the method generates a target point of a wheelchair from a cluster of pedestrians who move the same direction, it is known that the sensing noise sometimes generates outliers in cluster classification significantly. In this paper, a crowd tracking of an electric wheelchair using RGB-D camera is proposed by using a median of candidate vectors (MCV) observer. A target point is generated from clusters of pedestrians heading in the same direction by clustering pedestrians’ group. The outliers when recognizing pedestrians can significantly degrade the accuracy of clustering of the group. Therefore, we design MCV observer to deal with outliers in clustering of the tracking group. Finally, the effectiveness of proposed crowd tracking is evaluated in the fundamental experiments.