Human Activity Determination
The goal of this project is to classify complex motions (activities) of human bodies. Instead of relying on image data alone, we are using a 3D motion database that contains representative motions. The motions are captured in a MoCap studio and depend on the application of the system. The users then perform similar motions in front of inexpensive video cameras. The system will automatically classify the user's motions into categories and execute appropriate commands (e.g., raise an alarm). Applications include physical security, user interfaces, assistive technologies for handicapped people, remote supervision of physiotherapy, or remote motion training for athletes.
Background & Objective: The insight behind the system is that while simple vision processing may provide incomplete and inaccurate information about the user's movements, with the addition of domain knowledge from a previously captured motion database, plausible classification is possible. Perhaps more surprisingly, this process can be performed interactively, with less than a second of delay between the capture of the video and the classification of the motion.
Technical Discussion: The system combines information about the user's motion contained in silhouettes from several viewpoints with domain knowledge contained in a motion capture database. In our system, the user performs in front of one to three video cameras and the resulting silhouettes are used to estimate his or her orientation and body configuration based on a set of discriminative local features. Those features are selected by a machine learning algorithm during a pre-processing step. Sequences of motions that approximate the user's actions are extracted from the motion database and scaled in time to match the speed of the user's motion.
| Technical Reports: | |
| Learning Silhouette Features for Control of Human Motion | |
Technology Area: Computer Vision
Modification Date: September 12, 2007