Products 

Documentation
 
FAQs 

Applications 

Technology 

How Does SIGHTech Do It?

A Technology Statement   

"Any sufficiently advanced technology is indistinguishable from magic."   — Arthur C. Clarke's Third Law  
 

Many machine vision end-users, integrators, and OEM's who have seen SIGHTech’s products in action can identify with Arthur C. Clarke’s statement above.

This paper is intended to remove some of the mystery behind SIGHTech’s patented Neuro-RAM^™ technology.

Eye Catching  

Perhaps the most remarkable aspect of SIGHTech’s products is their ability to learn objects simply by looking at them.  Learning continues unhindered even if the object is moving or rotating.  What’s most impressive is that SIGHTech’s products can learn rotating objects in just a few minutes without any programming.

Another dramatic feature is how quickly SIGHTech’s products make decisions.  For example, Eyebot can make up to 60 decisions in a single second.  Eyebot can recognize objects it has learned in cluttered backgrounds, or identify defects quickly without strobe lighting.

Inspired by Yellow Jackets   

The yellow jacket bee inspired Art Gaffin, SIGHTech’s CEO and CTO, by presenting him with a difficult puzzle: How can yellow jackets see when they have a tiny brain, little processing power, minimal memory, and crude visual sensors?

Mr. Gaffin’s conclusion was simple: it’s not that machine vision does not have sufficiently powerful hardware (because yellow jackets clearly do not), it’s just that we do not know Nature’s algorithm.

Therefore, over 15 years ago Mr. Gaffin started an ambitious quest: understand and decode Nature’s algorithm for vision and learning.  Nine years ago, Brad Smallridge teamed with Mr. Gaffin to produce disarmingly easy-to-use and powerful machine vision systems.  Thus, Neuro-RAM was born in Mr. Gaffin's garage in Silicon Valley. 
 
Neural Net/Fuzzy Logic Attributes   

SIGHTech's technology, therefore, is based on how yellow jackets (and other organisms) see, learn, and navigate.  This results in a self-learning system that does not require any programming.  Instead, the system "programs" itself when it sees new objects.  The more you teach it, the better it gets. This is the concept behind neural networks: repetition reinforces learning.

Learns Objects/Features, not Pixels   

SIGHTech's Neuro-RAM algorithm extracts, learns, and inspects extremely small visual features.  In fact, it learns up to 12 million features a second.  Moreover, it learns these features in relationship to each other.  The processing going inside the black box is simply astounding.

What is a feature?  Generally, features are very small shapes, such as a segment of line, circle, or squiggle.  A simple way to visualize a feature is to look at a human face and zoom into a part of the face, say, the ear.  You will notice that there are many little lines and shapes that make up the concept of an ear.

Eyebot doesn't store the picture of the ear, but instead compiles a massive database of features in association with other features.  During the RUN process, Eyebot looks at 12 million features a second and compares them with the database of features it has in its brain.  If it sees a new feature, it decreases the Score slightly.  The more drastic the new feature is, the more the Score drops. Similarly, whenever many new features occur, the Score will drop precipitously.

The new feature may be a new shape that Eyebot never learned, or an old shape seen in a new way (e.g., rotated or in a new relationship with another feature).  When two features are near each other, they create a third feature. If this third feature is not part of the "known feature database," then Eyebot will signal it

No Strobe or Trigger Required

Another characteristic that mystifies those who try to understand SIGHTech’s technology is how it learns and inspects full video frames on the fly without a frame grabber and with so little memory (Eyebot only has half a megabyte of memory).

The secret lies in Eyebot's pipelined architecture which processes every video field on the fly. It does not grab/store frames or count pixels, but instead learns features and objects. As a result, the algorithm is highly memory efficient.

The moment Eyebot sees a new feature it signals it. Since it looking at every video field, it will catch a defect anywhere in the video field, in real time.  As a result, you don't need a strobe light or trigger sensor telling it when to look. For example, bottles can go by at up to 3600 PPM and not need a trigger.

Software on Hardware   

Much of the speed boost is derived from the fact that SIGHTech has embedded its algorithm directly on an integrated circuit (IC).  By running the algorithm on a Field Programmable Gate Array (FPGA), SIGHTech benefits from a 100 times faster learning and decision making ability than if the algorithm were on a PC’s hard disk.

SIGHTech has encrypted its algorithm onto the IC using some of the toughest encryption techniques available, making it nearly impossible to reverse-engineer.

Industry Standard Based   

Why are SIGHTech’s products so inexpensive?  Eyebot is only $4,995. Mini-Eyebot, SIGHTech’s newest member of its machine vision family, will be priced much lower - especially suited for volume applications.  

SIGHTech can sell at these prices because its technology is based on off-the-shelf industry standard components.  Eyebot works with any NTSC-based camera; indeed, it can work with your camcorder.  Eyebot has two industry standard outputs: RS-232 (serial) port and two optically isolated relays.  As a result, users are not tied to proprietary standards and can easily swap cameras, monitors, PLCs, and PCs to take advantage of SIGHTech’s Neuro-RAM technology.  

SIGHTech’s Foresight