Silicon-based CMOS camera sensors are cheap and plentiful, but they’re rarely used to their full potential: they can detect a greater range of wavelengths in the infrared spectrum than they can in the visible spectrum, but in most cameras this is blocked by an IR-cut filter. [Project 326]’s infrared camera system reverses this: it records infrared images in color while blocking out visible light.
The system uses three USB webcams, each with its IR cut filter removed and replaced with a different dichroic IR band-pass filter. One filter is centered at 750 nm, one at 850 nm, and one at 940 nm. There is no band overlap; in testing, each camera only detected an infrared flashlight tuned to its own filter wavelength. The original cameras didn’t hold the sensors in a consistent position, so [Project 326] designed new housings. Using three lenses, each with distinct aberrations, introduced some difficulties in alignment. [Project 326] originally intended to use a pair of beam-splitting prisms with only one lens, but this proved too difficult to align using 3D-printed frames.
A Raspberry Pi records a separate monochromatic stream from each camera, which can then be processed into a composite color video. The first frames need to be manually aligned, but afterwards a script can apply the alignment to the rest of the video. Finally, the channels are mapped to colors, with the precise mapping being freely changeable. There were some few unexpected issues: each camera has its own, not terribly precise, local oscillator, and they drifted apart by about one or two frames per minute. Parallax error, on the other hand, was less severe than might be expected: at close range it’s noticeable, but by a distance of 35 meters, it represents less than one pixel of distortion.
The resulting images look great, and it’s easy to forget that they’re being captured without the use of any visible light. We’ve seen a similar technique (though extending into the visible range) used to recreate the surreal effect of Aerochrome film.
