Hydra is a real-time system for detecting and tracking multiple people when they appear in a group. We describe the computational models employed by Hydra to track multiple people before, during and after occlusion. Hydra combines a silhouette-based shape model, a motion model, and correlation-based matching methods to classify whether or not a foreground region contains multiple people, and to segment the region into its constituent people and track them. Experimental results demonstrate robustness and real-time performance of the algorithm.