Laser-based tracking of groups of people with sudden changes in motion

This paper presents a method for laser-based tracking of people in a group. A two-layer laser scanner set in an environment provides laser-scanned images. A background subtraction method extracts people´s position data from the images. Data association based on heuristic rules and the global nearest neighbor (GNN) identifies multiple people in crowded environments. Their people are then tracked via a model-based tracker; the interacting multiple model estimator is applied in order to track people with sudden changes in motion (walking, running, starting, stopping, or turning suddenly). People in a group meet a considerable number of occlusions, which often deteriorates tracking performance. To cope with this problem, the proposed method groups people with similar motions. Thus, when people are occluded by other people in the same group, the occluded people are tracked using laser position data related to their nearest neighbor in the group, which improves the tracking performance. The experimental results of 11 people tracked in a small area validate the effectiveness of this method.