Detecting and Tracking Multiple Moving Objects from a Moving Platform

Research is proposed to investigate the feasibility of a novel tracking algorithm that uses spatial context to reason about long term intentions. While it is hard to fully anticipate a person's behavior, we can generate a set of reasonable hypotheses about it by selecting a number of high-level goals for known persons (e.g., a civilian goes to a store or home, a soldier searches for cover, a biker stops at a red light). This innovative technique will fit into a larger effort to detect and track heterogeneous moving obstacles (e.g., cars, humans, and aircraft) in crowded and cluttered environments where multiple occlusions occur. In general, Phase I research will seek to enable: 1) Extraction and tracking of moving objects from moving stereo images in an environment full of moving objects; 2) Tracking of moving objects using short-range prediction techniques; 3) Deduction of a spatial context from moving stereo images; and 4) Tracking of frequently occluded moving objects using long-term behavior reasoning — the essential innovation. Phase I testing will evaluate the performance of the algorithm both on the ground and in the air.