Thermal imaging sensors and cameras keep them hip in the world of computer vision where mobile startups are propelling the field. The Boson is a small thermal network camera and can be used for many applications like:
- Thermal imaging—recognizable to most of us as the “predator vision” demonstrated by the alien in the 80s Schwarzenegger film
- Facial recognition for various security or marketing efforts
- Pedestrian detection (e.g. being able to detect the number of humans in a certain area and their movement or activity level)
There are many more things these cameras can do…even the super-resolution similar to the much internet-maligned CSI “zoom in” is actually possible through software manipulation of these kinds of cameras according to Movidius’s Jack Dashwood.
Regardless of product features, two things interest me about the launch of the Boson (well, besides predator vision because who doesn’t love predator vision). One is the continued trend in miniaturization and its implications on interactive eyewear. The second is the fact that the Boson has a processor integrated directly into it. As a trend, this is called System-on-Sensor.
Miniaturization
The Boson is half the size, one tenth the volume, one seventh the weight and twice as power efficient as the previous model which was called the TAU 2. Movidius’s tiny 12-core, low power processor (which we’ve covered before) allows the Boson camera to get much, much smaller.
This makes integration into interactive eyewear or other smart glasses/helmets much more efficient and thereby allows those wearables to also become smaller and more efficient themselves. This interests me because smaller, more powerful interactive, heads-up displays could sell more units and get more use if they are not giant monstrosities you wear on your head.
System-on-Sensor
This leads to the second trend: System-on-Sensor. Not only are there size benefits by bringing the processor directly into the sensor, there are new abilities in the sensor itself. The 12 cores on the Myriad 2 chip are fully programmable which means the Boson can directly process images, combine thermal pixel information, or process facial recognition algorithms, directly on the Boson itself, in real-time. This camera doesn’t need to offload this processing to another subsystem or to the cloud.
A low power processor like this still has limits to what it can do, for sure, but this is a trend to keep an eye on as it takes the power of IoT (Internet of Things) sensors to a new level where they can network together to complete advanced capabilities on their own.
If there is any truth to what Peter Diamandis predicts will be a trillion-sensor economy in the next 10 years or whether infrastructure like this could lead to intelligent, camera-based neural networks that can “act on information” instead of merely recording it, remains to be seen. But little advances like this could play a part in that possible future.
This article comes from techcrunch edit released