Embedded Facial Mapping; Tracking Facial Expressions with Limited Hardware

In today’s markets, facial detection technologies are making great strides forward in accuracy and speed, continuously pushing the limitations of computerized vision and offering ever-increasing opportunities for automation and machine learning in what was once a human-dominated concept. But while the accuracy of these systems increases, so do their computational requirements and complexity. Compounding the problem, advancements in hardware have not held the same rapid pace. Researchers now face the very real possibility that their algorithms will be bottlenecked by the very hardware it runs on. While we may be able to run current programs on current hardware, the idea of pairing a full-sized computer with a single camera does not scale well in terms of current or future applications. The goal of this project will be to examine one component of facial recognition software: facial expression detection and evaluation, working to maintain a certain level of accuracy while functioning on hardware which is significantly weaker in computational power relative to the modern desktop workstation. If successful, this could serve as a future step in the ideal of pairing a camera with a cheap, efficient, microprocessor for facial processing.