Building a High-Speed Camera for the Masses

By Norman Chan

A high-end "microscope for time", finally within relative reach.

High-speed video cameras aren't cheap. That's a common understanding in video production today, which is what gates the use of extreme slow-motion footage to productions that can afford or have access to professional high-speed camera equipment. In the consumer market, it's true that we're seeing more cameras dip their toes into high-speed recording--notably the iPhone 5S with 120fps recording at 720p and the GoPro Hero 3 with up to 240fps at 480p. And yes, the natural motion of everyday subjects (read: cats, dogs, and babies) look delightful when slowed down by four or eight times their normal speed. But if you want to catch the popping of a balloon or measure the ballistics of a bullet, that's the realm of $100,000+ cameras made by companies like Vision Research and Photron.

But why are these professional high-speed cameras so expensive? And is there a way to make that tech more affordable? Those were the questions posed by Mike Matter and his collaborators in the founding of the bay area startup Edgertronic. Through a (just funded) Kickstarter project, Edgertronic is attempting to democratize professional-grade high-speed videography with its eponymous camera, priced at a reasonable $5,000. A high-end "microscope for time", finally within relative reach.

I visited Mike over the weekend to check out the latest build of the Edgertronic camera and get a demo of it in action. You'll find sample videos below. But more importantly, I wanted to learn the answers to the same why's that Mike and Edgertronic are posing to this exclusive high-end market. The first step: understanding how a high-speed video camera actually works.

At its core, a high-speed video camera is a lot of any typical video camera, which in today's terms is not much different than a digital still camera. That means it captures light passed through a optical lens onto a digital sensor, passing that light data through circuitry to an image processor, and then storing the resulting digital image information on some kind of memory. Sensor, processor, and memory are the essential electronic components to any digital camera, and the type of quality of each factor into what kind of photo or video you're able to record. Standard video recording, for example, requires a sensor with photodiodes that are able to read light data at a rate of over 30 times a second, and a processor that is fast enough to render that data into a video and store it to memory. Fortunately, for today's technology, 30Hz recording of high-resolution video is simple and cheap. But when a camera sensor is required to record light at five or ten thousand times a second, and pass that data to a processor at a rate of several hundred megapixels a second, commodity hardware isn't going to cut it.

Except, off-the-shelf technology is exactly what makes Edgertronic's high-speed camera work. Instead of spending millions to develop their own sensor and processor, Mike and his team looked to components they could readily buy--the trick was finding the right balance of image sensor and processor to suit the needs of high-speed video recording. They ended up with three options for sensors, and ended up choosing a 18x14mm CMOS sensor--roughly the size of APS-C--that accounts for almost half the cost of the camera. It wasn't the fastest sensor available, but had two characteristics that made it ideal for Edgertronic's needs: big pixels and a global shutter.

Prototype of the Edgertronic camera, with a chassis of CNCed aluminum.

You can think of a digital camera sensor as a grid of pixels, each a photodiode receptor that's able to capture light data. A general rule of thumb is that the larger the physical size of these pixels, the more light can be captured, which leads to images with lower noise and better dynamic range. Two primary factors determine the pixel size: the overall size of the sensor, and the number of pixels on it. Manufacturers can design sensors to vary these factors to optimize for pixel size, which is how Apple made a big deal about the iPhone 5S's 8MP sensor having 1.5 micron-wide pixels. Edgertronic's camera sensor, the size of sensors found in entry-level DSLRs, only has a resolution of 1280x1024--paltry by still image standards, but enough for 720p video. But that also means its pixels are a massive 14 microns wide. That kind of light sensitivity is crucial to high-speed video recording because of the other critical video recording factor: shutter speed.

When we talk about a video's frame rate, that's referring to the capability of the sensor to store and pass along light data and the image processor's ability to record those frames to a video file. But also essential to video recording is how much light is allowed to fall onto the sensor for each frame, a factor measured in fractions of second. What gates this duration is the camera's shutter. On a still camera, like a DSLR, a physical mechanical shutter flips at a very high speed to allow a very specific duration of light to hit the sensor. (On high-end DSLRs, very fast shutters are achieved using a combination of mechanical and electronic shutters). Video camera sensors have electronic shutters--extra circuitry on the sensor that relieves the charge of each pixel at extremely quick intervals to "turn off" the light reception capabilities of that pixel at a predetermined rate. The challenge for CMOS sensors is that most use "rolling" shutters, meaning that the pixels are discharged linearly, row by row. Consequently, rolling shutters are susceptible to image artifacts like skewed or wobbly objects for objects in fast motion, which makes them unsuitable for high-speed recording.

A global electronic shutter, on the other hand, discharges all the pixels at once, which gives each pixel on the sensor a consistent amount of light exposure per frame. The CMOS sensor that Edgertronic is using is that type of sensor, rated at a maximum speed of 1/200,000 seconds.

Video is stored on an SD card, and commands are sent through Ethernet.

Of course, there is other important technology that goes into making a high-speed camera. Edgertronic credits its competitive advantage to its engineers' ability to efficiently pair the sensor and processor hardware with good electronics design and FPGA code on its custom board. The data rates that come from a high-speed camera sensor are extremely fast, and according to Mike, the challenge is optimizing for that data load with as little overhead or noise as possible. This engineering--which goes way over my head--is the real secret sauce that allows Edgertronic's camera to work.

And work it does. Mike took me to his backyard, where he set up a demo of the camera. I got to examine the latest Edgertronic design, which is an impressively barebones design. If the idea was to pare the camera to its essentials, Edgertronic has done a good job. The aluminum chassis that houses two electronics boards is lightweight, with ports on the back and a Photodiox lens adapter on the front. Edgertronic is compatible with Nikon F-mount lenses (approximately 2x crop factor), a design decision informed by the wide availability of Nikkor lenses with manual aperture rings. There are two quarter-inch tripod mounts on the sides of the camera, and the body is designed to act as a heatsink with a small fan for active cooling. The camera is almost button-less, with only a tiny button to initiate a hard reset. Controls and the camera's live-view are operated through the Chrome web browser on a tethered computer, in this case a MacBook. There's no HDMI or SDI video out--additions deemed unnecessary to the scope of this camera, and would have only added additional costs and overhead.

From the web browser window (where the live view plays with a low latency delay and at around 7-10fps), you're able to adjust the recording settings of the camera. The settings are listed in order of desired importance, and will automatically fill in the maximum allowed values for some entries depending on what you've chosen for others. Because of the data-processing limits of the camera system, configuring high-speed video recording is a matter of trade-offs. You have to find a balance of video resolution and frame rate.

At the max resolution of 1280x1024, the highest frame rate the Edgertronic can record at is 494fps. And with the max frame rate of 17,791fps, the highest-resolution you can record is 192x96 pixels. For the HD resolution of 1280x720, 700fps is the cap--already far superior to what the Sony FS700 prosumer camera can do. I think that the sweet spot for the Edgertronic is 640x360 resolution (standard for web video), at over 2,000fps. Interestingly enough, because of the way sensor data is processed, increases in horizontal resolution (how tall a video is) have a bigger impact on performance than increasing vertical resolution (how wide a video is).

Another operational consideration is that finding focus with the Edgertronic may be more difficult that other cameras with built-in LCD displays. Manual focus using the lens ring worked just fine for the demos, but the software doesn't yet have any focus assistance tools like edge detection--you have to play it all by eye.

Like high-end high-speed cameras, Edgertronic allows for pre, post and mid triggers of video capture. 8GB of DRAM buffer records up to 8 seconds of video (also a function of resolution and frame rate), so when you hit the capture button on either the software control or a wired physical trigger, the camera can encode to video an event after the fact.

Encoding and record speed is comparable to that of a camera like the FS700, at a rate of 60 frames per second. So for example, if the camera captured one second of 1000fps video for a total of 1000 frames of footage, that video would take about 17 seconds to encode (1000frames divided by 60 frames per second) and store to the SD card. I was told that the H.264 encoding is currently set for a variable bitrate averaging 20-30 mbits/second and tuned for i-frame interval encoding. These are settings that users will be able to adjust in the future, too.

In terms of video quality, Edgertronic's output looks really good, especially at resolutions of 640x480 and higher. The 700fps video recorded at 720p is crisp and detailed, and the high shutter speed removed any blur so each frame is as sharp as the next--important for academic purposes. Even the 5,000fps video recorded at 288x288 looked usable when stretched out. At 17,000 frames per second, the low res video looked about as good as the 960fps video we shot with the FS700 last year, but the framerate was high enough to let me see exactly how a balloon tears apart from being overblown. Edgertronic can't compete with cameras that can record full 1080p or 720p at over 2000fps, and its sensor maxes out at 400 ISO, but it's mighty impressive for a camera that's less than 1/20th the cost of what's used by Zack Snyder or The Slo Mo guys.

Edgertronic's target market isn't necessarily Hollywood, though. As Mike tells it, his goal is to get these cameras into the hands of hobbyists and independent filmmakers--people who otherwise wouldn't otherwise dream of having access to this tool. The idea is to make high-speed video--real high-speed video--more attainable to the masses. It's the law of the instrument: put a high-speed camera in everyone's hands and we all start seeing the world differently.

Test 1: 700fps, 1280x720, ISO 200, Shutter 1/5000s

Test 2: 5000fps, 288x288, ISO 200, Shutter 1/6000s

Test 3: 17,791fps, 192x96, ISO 200, Shutter 1/22000s