SBIR-STTR Award

A System for Automated Reduction of MPEG-1 Encoded Facial Videos
Award last edited on: 7/22/2022

Sponsored Program
SBIR
Awarding Agency
DOT
Total Award Amount
$100,000
Award Phase
1
Solicitation Topic Code
-----

Principal Investigator
Martin Miller

Company Information

M-Vision Inc

651 Reed Street
Santa Clara, CA 95050
   (925) 997-7298
   N/A
   www.m-vision.com
Location: Single
Congr. District: 17
County: Santa Clara

Phase I

Contract Number: ----------
Start Date: ----    Completed: ----
Phase I year
1999
Phase I Amount
$100,000
M-Vision Inc, along with University of Michigan, proposes to design a system for automated reduction of human face videos that are included using MPEG-1. The proposed system will achieve the data reduction in four steps (they both operate directly on the In M. MPEG-1 encoded video stream, and do not required decoding the stream into its spatial domain counterpart.) * first, it will temporally segment MPEG-1 encoded video of the human face. This segmentation will be based on automatically detecting those frames of the video stream with a frame-to-frame motion of the face is significant. * next, it will ensure that the facial pose within each segment is indeed homogeneous, and that the first step did not introduce any errors * third, it will determine the overall movement of the face as a result of motion between the temporal segments * finally, it to classify the facial pose within each segment by analyzing the face using the within-segment and between-segment information. For pose classification, a non-parametric eigenface approach will be attempted. For comparison, and does and approach with inherent value, a neural network will also be designed for pose classification in the spatial domain. Anticipated results and potential commercial application of results If the Phase I study is successful, then it would demonstrate the technical feasibility of directly browsing MPEG-1 encoded videos. MPEG-1 encoded videos are ubiquitous in several multi-media applications, and the direct browsing capability paves the way for a very desirable random access of the video stream. Understanding drive the behavior is of interest not only to government agencies such as NHTSA and the Army, but to many automobile manufacturers and their suppliers as well. In the commercial industry, drive of pose (and fatigue) to seize is seen as a potential trigger for activating all-board driver warning/assistance systems. The designed pose determination approach will be valuable in the design of such trigger mechanisms as well. Keyword-Facial videos, Neural networks, Direct browsing, DCT features, Facial movement, MPEG, Motion estimation,Eigen faces.

Phase II

Contract Number: ----------
Start Date: ----    Completed: ----
Phase II year
----
Phase II Amount
----