WHERE A CAMERA HAS TO DECIDE, NOT JUST RECORD
SpectralSense builds advanced computer vision and inference software that runs on discrete silicon, at the edge, with our own sensor and optical designs behind it. Does this person or object belong here, at this time, and what should happen next? Those calls used to need a trained person watching a screen, and people get tired, look away and make mistakes. We train our systems to understand the scene in front of them and make that call themselves, using rules you set.
We do this to put the decision right where it is needed. No data center, no backhaul, no network. The device sees what is in front of it, runs its logic locally, builds context, and acts in milliseconds on the parameters you give it. It does this on its own, with no human in the loop. The intelligence lives on the device, where the decision actually happens.
With that understanding of the situation, we have built optical systems that are used at workplace gates, closets, and even a bedroom with a mosquito in it. Here is what that looks like.
From a security gate to a closet, our systems build context of their situation based upon your criteria. For a door or gate, that means who is expected there, what they should be wearing and whether they belong. We account for time, situation and house rules to determine why.
We apply that logic to a closet too. Is it the look you are going for, does it fit your goals and is it proper. For a room, we look at what should be moving and what should not. We look at speed, size and range and determine if the object belongs.
It is all the same core engine, pointed at different problems, and it all runs on the device.
A real frame from a WorkCheck gate. The shirt reads hi-vis but is not a vest, so the system did not guess. It asked a person.
WorkCheck is a checkpoint for any workplace with a rule about what people wear. It checks each person against the site's rule in about a second, says CLEAR, WAIT or STOP, and writes a record that cannot be quietly edited later. Two vests that look identical to a person, or to a webcam, are not identical to this system. No face recognition, no images and no personal information are stored, just a numerical sequence. Running at live gates now.
LookSense is the same engine pointed at your own closet. It reads what you own and tells you how to get closer to the look you want, based on your preferences. It learns what you have and the style you like, then recommends items that move you toward your appearance goals. It is available on the App Store.
Tripwire is the same engine pointed at small, fast things. It started by tracking mosquitoes in a bedroom. The system learns what objects you want to track and tracks them based on those rules. Male and female mosquitoes may look the same to a human, but not to us. Females carry dengue, malaria and other deadly diseases, so we single them out with markers. Our house rules target the females today. tripwirehome.com
Inside each unit there are two kinds of thinking. A small, fast model watches every frame on a dedicated vision processor and handles the ordinary cases on its own. When it sees something it cannot explain, a much larger model on a second accelerator wakes up, looks at that one moment, and works out what is going on. What it learns goes back into the small model, so the unit gets sharper the longer it runs. No hallucinations.
Until recently the second model needed a data center. Now it fits in a box the size of a hardback book, draws less power than a light bulb, and never sends a picture anywhere. That is the change we are built on.
SpectralSense is in Holladay, Utah. The units are designed and assembled in the USA. More about the company.