Multiple intention tracking by a generalized potential field approach

Fully automated vehicles and mobile robots operate in a shared environment with pedestrians. To minimize the risk for pedestrians, it is very important to track them in a precise way. As cameras are often installed in surveillance situations, they are used for tracking pedestrians in a shared environment. To improve the accuracy of the tracking, it is necessary to include all available context information in the fusion process. One important information source is the intention of the pedestrian. A generalized potential field is used, which can be modeled using pedestrian movements. When the intention of the person is unknown, different hypotheses for the intention of the pedestrian are considered. A Multi-Hypotheses tracking filter fuses the intention information and the pedestrian position measurements of a camera, whereby the tracking accuracy is improved. The proposed approach is evaluated using real camera data from a simple scenario in Edinburgh Informatics Forum. All results are evaluated in dependence of the measurement quality and the frame rate of the camera. The Multi-Hypotheses based tracking outperforms the simple Kalman filter over the whole range of frame rates and standard deviations.