Facial Tracking and Animation


Introduction Two weeks ago we submitted our Critical Design Review. At that time we had performed a proof of concept for each of the subsystems in our overall design: Data Acquisition, Point Initialization, Point Tracking, and FAP Generation. Since then we have begun implementing the base functions of each subsystem, as well as started the integration process, though currently just between Data Acquisition and Point Initialization. Data Acquisition With the design constraints nailed down in the Critical Design Review (CDR), implementation of the Data Acquisition System has begun. Phase 1 of the Data Acquisition system has been fully implemented and validated. Phase 1 of the Data Acquisition system contains the functionality to extract and deliver video frame data from a movie file (avi) to any needing functions. The data acquisition system will run at an adjustable periodic rate, from 25 milliseconds to one second. The core of the validation and testing of Phase 1 of the data acquisition system was aimed at the timing of the data delivery. Reliable timing characteristics are crucial to the testing of the point tracking algorithms, which this system is designed to accommodate. The system has built-in testing and validation functionality. The system displays, on the monitor, the current data being passed out on the output buffers. This functionality enables visible confirmation that the system is running, and has good behavior. The system has built in checks for the periodicity of the system. The system checks the period on every loop and counts the number of time the period is outside of a plus or minus 10 percent of the target period. Upon completion, the system displays the number of times the period was outside of its bounds, and in which direction the error occurred (higher or lower than the target period). Details about the errors, such as, the actual period and which iteration, can be found in a delay log file. The results gathered through a series of test runs at different periods indicate that the system is stable and will satisfy the needs of the point tacking and initialization. In a test of 21 runs, the system produced an average of .28 high errors per run, and NO low errors per run. With .28 errors per run, the system has a .08 percent error, which will not compromise the effectiveness of our system. This is acceptable because the system can guarantee a period …

Extracted Key Phrases

Cite this paper

@inproceedings{Belote2004FacialTA, title={Facial Tracking and Animation}, author={Todd Belote and Brad Busse and David T. Brown and Bryan Harris}, year={2004} }