Hewitt Consulting

Computer vision solutions for businesses

Phone: 858-472-2989

 

Example Projects

  • Improved the accuracy and speed of a portable, commercial product that uses a camera to locate and read car license plates

  • Designed and implemented the computer vision portion of a proof-of-concept projector-camera system for a large computer hardware manufacturer (name is under disclosure restriction). This proof-of-concept demonstrated feasibility and the product is now being developed for commercial introduction

  • Developed 3D extensions to SIFT that improved visual object recognition for client Evolution Robotics. For details, please see our BMVC2007 paper

  • Invented a fast, accurate, self-calibrating method for tracking faces in videos

  • Performed metric reconstruction from surveillance video for the Innocence Project to produce a legally defensible estimate, with error bars, of the height of the suspect in a crime scene

 

Licensable IP for Object Recognition Available

SIFT has become the de facto standard for object recognition in the research community. Use of this patented technology in business applications within the U.S. requires a license from the University of British Columbia. Unfortunately, some uses may not be available for licensing due to certain exclusivity rights having been sold previously.

This situation poses a problem for many businesses that would like to use object recognition. To address this problem, I've invented an object recognition method that can replace SIFT with comparable or improved accuracy. This patent-pending technology is owned by Hewitt Consulting and is available for licensing. More about this here.

 

High Quality Classifier Training Available

To improve computer vision solutions that make use of classification, I utilize a variety of tools for training high-performance classifiers. Developed and refined over a period of years, these in-house tools significantly outperform classifier training available from OpenCV.

As an example of the benefits, an eye detector trained with my custom implementation of Gentle Boost with active learning was trained overnight, improved with active learning, then retrained for a total turnaround time of two days. In contrast, OpenCV's classifier training typically takes a week or longer. A further consideration for businesses is that OpenCV's Haar classifier uses the patented Viola-Jones method. The eye detector trained with my tools avoided the patented Haar cascade while improving accuracy by 65% when compared to the best result from the various OpenCV classifier cascades for eye detection.

 

Other Skills and Knowledge

  • Proficient in C/C++, Matlab/Octave, OpenCV, and Java

  • Familiar with patent issues in computer vision and how these affect business decisions

 

Free Initial Consultation

If you're considering using computer vision in your business, I'd encourage you to contact me for a free initial consultation. I can help you determine if a computer vision solution is your best option and, if so, how you can pursue that in a cost-effective manner.

 

Phone: 858-472-2989

 

Home