Engineer Creates High-Speed Camera That Captures 400 Frames Per Second

If you've ever watched an episode of Mythbusters, chances are you've seen them use a high-speed camera. This is a camera with the capability to capture moving images with exposures of less than 1/1,000 second or frame rates in excess of 250 frames per second. It's often used to record fast-moving objects that can be played back in slow motion later. But even the most advanced high-speed cameras have speed limitations. Engineer and tinkerer Shane Colton recently pushed these limitations to build a high-speed camera that captures 12,000 frames of continuously recorded 4k 400fps video, a feat similar devices can't match.

Colton was able to reach such speeds using modern SSDs, which are quite fast. Typically, high-speed cameras employ RAM to buffer raw frame data in real-time. NVMe flash is able to write speeds into the GB/s range. Not only does this make it possible to capture long clips, but it also has a simplified interface: push to record, push to stop. He even used wavelet compression to reduce the data rate by a ratio of roughly 5:1.

In terms of hardware, Colton chose the lower-end Zynq UltraScale+ SoCs, 134 of the 144 LVDS-capable I/O, all four PCIe Gen3-capable transceivers, as well as created a new carrier that includes an HDMI transmitter and an isolated GPIO connector.

To ensure the files transferred and recorded correctly, Colton had to deal with how frames are buffered in RAM and how they are written to files on an SSD. He also had to address the codestream RAM address and size, which were built into the frame header, making it easier to validate and analyze the output.

"The codestream size per frame shows how bandwidth is being distributed to the 16 codestreams, with more bits per frame going to the low-frequency subbands," says Colton. "Compression ratios relative to raw 10-bit data are shown on the codestream size axis. Spikes can be seen during the portions of the clip where the steel wool burns brightest. There's also a gradual increase in codestream size over the 30 seconds, especially in the high-frequency subbands, which I believe is due to image sensor noise increasing with temperature. Feeding the codestream size back into the quantizer settings to maintain a roughly constant bit rate will be a problem for another day."

With all that sorted out, the 12,000 frames were able to be recorded onto the SSD in 30 seconds. Though the end results were a success, Colton says there's still a lot more work to do. He'll continue to make improvements on the device, such as adding the HDMI output, USB mass-storage device access to the SSD, and figuring out an alternative programmable login configuration for the CMV12000s subsampled read-out mode. Still, the project has come a long way from when he first took on challenge back in June 2019.

Read Colton's full process here.