Real-Time Model-Based Visual Tracking

This paper describes a robust approach to real-time model-based visual tracking and servoing for rigid polyhedral objects moving with 3 degrees of freedom. It is recognized that low-level feature segmentation is not infallible. Uncertainty measures enter into all computations to account for varying degrees of confidence in the measured object features. Robust tracking and servoing has been demonstrated in the presence of partial occlusion and cluttered backgrounds at tracking speeds of up to 1.2 rad/s (69 deg/s). This algorithm has also been used for vehicle following where a vehicle is driven autonomously by following the path of another vehicle.