Obsidian’s thermal imaging camera offers high resolution at low costs thanks to a new production technique
Thermal imaging has been a crucial technology in the war in Ukraine, allowing hot targets such as vehicles and soldiers to be spotted on the darkest nights. Military-grade thermal imaging cameras used on large Baba Yaga night bombers are far too expensive for drone makers who assemble $400 FPV kamikaze drones and have to rely on cheaper devices. But a new technology developed by US company Obsidian Sensors Inc could transform the thermal imaging market with affordable high-resolution sensors.
Thermal imaging for drones
While digital cameras developed rapidly after their introduction, thermal imaging is in the doldrums. This is because without a mass market there is no incentive for manufacturers to invest in the next generation. The resulting difference in development has been dramatic.
Apple’s 1994 Quicktake camera offered a resolution of 640 x 480 pixels for $1000. In 1999, Nikon’s Coolpix 800 offered six times the resolution and cost less. Now even a $10 kids camera has a much higher resolution.
Low-resolution grainy thermal image from a Seek Thermal sensor
On the other hand, the Seek Thermal image maker for consumers was launched in 2015 with 206 x 156 pixel images for around $200. The same product is the market leader today at the same price. More powerful thermal imaging cameras – such as this $8,000 military system with 1024 x 768 pixels – are sold in small numbers.
This makes it difficult to make a cheap drone for night operations. A drone maker passing by FPVDronesUA on Twitter/X and builds FPVs for Ukraine and told me about the trade-off between price and performance in the imagers they’ve worked with, invariably from Chinese companies.
The minimum feasible sensor is an image sensor of 256 x 192 pixels. This costs about $200, but produces a very grainy image.
“256 gives you quite poor image quality, compared to better thermals, and is very dependent on weather factors such as humidity, cloud cover and how cold it is outside,” FPVDronesUA told me. “256 is really hard to fly.”
The next level is 388 x 254 pixels, which is significantly better but three times as expensive: about $600.
“They are less dependent on the weather because they capture more pixels. Their Spectral Band and NETD are about the same as 256,” says FPVDronesUA. “These are much more comfortable to use on a drone.”
But its sensor of choice is 640 x 480 (VGA resolution), starting at $800 if purchased in batches of 50 or more, but typically well over $1000 if purchased individually.
“640 thermals are great, but very expensive,” says FPVDronesUA. “These are very comfortable thermals, very similar to the ones we see in Mavic 3T [an industrial drone for thermal mapping]. Higher resolution thermals also have a larger field of view.” This means that the larger cameras offer a much wider view; it’s less like looking through a straw and steering is easier. “256 Thermal with a 9mm lens would be quite narrow, while the 640 Thermal with a 9mm lens gives you more than twice the field of view.”
Ideally, the camera would have a higher resolution to detect targets (and obstacles) at longer distances and make piloting less challenging. A $15 FPV daylight camera has a resolution of 1280 x 960, four times that of VGA.
Adding an $800 imager turns a $400 drone into a $1,200 drone, reducing the number of drones he can afford to build by a factor of three. That’s extremely important in a conflict where drone manufacturers like FPVDronesUA are still funded by charitable donations, and the Russians, with more funding, initially launched more FPVs with thermal imaging cameras. And there is no question of including them as standard in all FPVs, thermal imaging cameras are only intended for drones for night attacks.
New thermal imaging technology
Older digital cameras were based on CCDs (charge coupled devices), the current generation uses cheaper CMOS image sensors that produce an electrical charge in response to light. The vast majority of thermal imaging cameras use a different technology: an array of microbolometers, miniature devices whose pixels absorb infrared energy and measure the resulting change in resistance. The conventional design neatly integrates the microbolometers and the circuitry that reads them on the same silicon chip.
Scaling up this technology would be possible, but would require a multi-billion dollar “fab” (chip manufacturing plant), as happened with digital cameras. There is no incentive to invest that kind of money,
John Hong, CEO of San Diego-based Obsidian Sensors, believes he has a better approach that can scale to high resolution at low cost and, crucially, high volume, at established foundries. The new design does not integrate everything into one unit, but separates the bolometer array from the readout circuits. This is more complex, but allows the use of a different production technique.
The readout circuitry is still on silicon, but the sensor array is produced on a glass plate, using technology perfected for flat-screen TVs and mobile phone displays. Large glass plates are much cheaper to process than small silicon plates and bolometers on glass cost about a hundred times less than on silicon.
Hong says the process can easily produce multi-megapixel arrays. Obsidian is already producing test batches of VGA sensors and plans to move to 1280×1024 this year and 1920×1080 in 2025.
Obsidian has been quietly developing its technology for six years and is now able to produce units for evaluation at a price three to four times lower than comparable models. Further evolution of the production process will bring prices even lower.
That could bring a 640×480 VGA sensor imager to well under $200.
Cars and drones
The thermal image from the Obsidian sensor (left) shows pedestrians invisible in the glare … [+]
Obsidian’s first target group is car safety and self-driving cars. Currently, these typically use LIDAR (laser-based radar) sensors to stay on the road and avoid accidents. But LIDAR cannot easily distinguish pedestrians from the background, which thermal imaging cameras can.
Last month, the US National Highway Traffic Administration issued a new safety standard for automatic emergency braking for all new passenger cars from September 2029. It must automatically detect and brake pedestrians, and the requirement to operate in the dark will heavily favor thermal imaging systems. Millions of new cars sold every year represent a huge potential market and this week Obsidian announced a partnership with Quanta Computer Inc to produce thermal imaging cameras for cars
Delivery drones are another potential market. These are currently in the small-scale testing phase, but large-scale operations are likely to require high-resolution thermal imaging cameras for safety reasons, and again demand could reach millions.
Meanwhile, Ukraine plans to build more than two million drones this year and import a million more, and their drone makers have already expressed interest in the technology. Other countries are likely to follow suit, creating even more demand for small, cheap thermal imaging cameras.
The future of thermal imaging
Hong says they plan to sell a thousand VGA cameras in pilot production this year, and are currently working on a series B to reach much higher volumes in 2025 and beyond. That should be just right to ride the demand wave in the coming years.
The emerging market will attract new suppliers. Many will likely be based in China and will simply scale up existing production techniques. But the cost and scale benefits, and the desire for greater supply chain security, could drive the new approach to building sensors on glass in the US.