Turning trains into a giant flatbed scanner with an industrial line camera

A photographer adapted the slit-scan technique — the same idea behind 1990s digital scanning camera backs — by inverting it: instead of sweeping a single line of pixels across a stationary subject, the whole camera moves while the world stays put. Pointed out the window of a moving train, the sensor captures one thin vertical line thousands of times per second, and those columns are stitched together into a complete image of whatever passes by, such as a container port. Early proof-of-concept runs with a phone (scanning a sofa, then riding Boston’s MBTA Orange Line) worked in principle but came out squished and smeared, exposing the core problem: image quality depends entirely on knowing the vehicle’s speed precisely, and even small velocity errors visibly distort the result.

The breakthrough was hardware built for factory conveyor belts — a Basler ruL2048-19gm linear camera with a 1x2048 sensor that reads out nearly 19,000 lines per second. Listed new around $700, it was sourced on eBay for a tenth of that. The trade-off is light: with a maximum exposure of 1/100s, it only works in bright daylight, ruling out dim stations and tunnels. Basler made the project easier than expected by offering its SDK freely and still supporting the 2013-era camera, which connects over gigabit Ethernet via APIPA addressing.

The rig combines the line camera with a Vivitar 28mm lens, a 3D-printed tripod mount, an IMU (accelerometer/gyro) for deriving speed, a GPS module, and a SAMD21 microcontroller feeding data to a laptop, all powered from a USB-C battery bank. GPS proved unreliable because Boston’s trains block satellite signals, and raw accelerometer data was too noisy to integrate into usable velocity — at one point implying the train was rolling backwards. The write-up is a candid maker log of the messy engineering reality behind a clever imaging hack, where the hard part isn’t capturing lines but accurately reconstructing motion.