Task 3: Activity Recognition

Calibration-free View Augmentation for Semi-Autonomous VSAMs

Activity recognition is an important component of this project for three reasons: (1) It serves to focus attention, to alert the VSAM to important phenomena and to allow it to ignore unimportant ones (such as wind-blown foliage), (2) activities are of interest to remote commanders and observers, and communicating their type and properties is more efficient than sending live images and (3) it can be used for local decision-making, allowing VSAMs to cooperate with each other and to make decisions based on sensed activities. Past work has enabled us to recognize ``natural clutter'', or temporal textures (wind-blown foliage, fire, water), and repetitive activities like walking.

In Year 1, we shall proceed to the recognition of ``impulsive" or one-time actions like throwing an object or opening a door; that is motion events that have definite structure, but unlike walking of running, do not repeat.

Year 2, we develop the recognition of motion events termed ``transitions", which represent significant changes in status, or the formation of relationships involving one more moving objects, like the starting, stopping, or turning of people or vehicles,

In option research in years 3 and beyond, we shall use contextual information to enhance motion recognition abilities. Context can be spatial, temporal, or situational, or some combination.

On To Next Task

This page is maintained by Mike Van Wie.

Last update: 11/11/96.