Transforming the Railway into a Giant Flatbed Scanner

By Philo | Published August 17th, 2026

For several months, I have been experimenting with an industrial linear scanning camera to capture ultra-wide panoramic photographs from the vantage point of ferries and trains. While the technical hurdles were significant, the visual output justifies the effort.

Example: A grayscale image captured on the San Francisco to Oakland ferry (Dimensions: 56,894 x 2,048 pixels). Scroll to zoom and drag to navigate.

I recently detailed this project during a presentation at EMFcamp 2026. You can view the recording or continue reading for a deep dive into the process.

---

🛠️ How It Works: The Concept

"What am I even looking at?"

To understand the imagery, you have to understand the capture method. Instead of a traditional rectangular sensor, the camera captures a single vertical slice of the world at a time.

By capturing these thin grayscale strips rapidly and stitching them side-by-side, a cohesive, massive image is formed. While the theory is simple, achieving a high-quality result is easy incredibly difficult.

📚 Background and Inspiration

In the 1990s, digital sensors couldn't compete with the resolution of large-format film. To bridge this gap, engineers created digital scanning backs. These devices didn't use a massive pixel grid; instead, they moved a small line of pixels (one for grayscale, three for RGB) across the film plane.

Comparison of Scanning Approaches

I had long considered building a scanning back for my own large-format camera, but the mounting logistics were too daunting. Furthermore, vintage units on eBay are prohibitively expensive and require ancient legacy computing environments to operate.

While watching a build by Gigawipf, a realization hit me: Why move the sensor across the subject when I can move the entire camera past the subject?

Loading film in Vermont, because I'm allergic averse to conventional photography.

I researched similar attempts—including the Scannoramic project, experiments by John Hikerbiker, Daniel Lawrence Lu's reversed camera, and Martin Liebscher's film work—but I felt there was room for improvement, specifically regarding motion speed compensation.

🛋️ Phase 1: "Slit Scanning My Sofa"

My first attempt was a late-night experiment involving my living room furniture. I placed my phone on an office chair and pushed it slowly across the room while recording video.

I wrote some "slapdash" code to extract the leftmost column of every frame:

# Hypothetical 'slapdash' extraction logic
for frame in video_stream:
    slit = frame[:, 0] # Grab the leftmost column
    final_image.append(slit)

Note: My code comments were filled with lyrics from The Beths' "Future Me Hates Me", which proved prophetic when I later had to write a post-processor for the actual camera.

The Result: The image was a distorted, squished version of my sofa. Even after doubling the columns to fix the aspect ratio, the image remained a mess because my pushing speed was inconsistent.

🚆 Phase 2: The Train Trials

I realized that speed measurement was the critical variable. I attempted to use a phone's accelerometer to track velocity.

• Tape phone to seat for accelerometer data.
• Hold second phone to window at 60 fps.
• Achieve clean velocity data.

The data was a disaster. If we assume the train's movement is along the $y$-axis, the velocity $v$ should be the integral of acceleration $a$: $\text{Position} = \int \left( \int a \, dt \right) dt$

However, the noise was so extreme that the data suggested the train was moving backward at the end of the trip.

During EMFcamp, I met Tim Jacobs (mitxela), who was also presenting on slit-scan cameras. While we started from the same place, he took a more artistic route, creating trippy animations by shifting the slit position within existing videos.

📷 The Hardware: Industrial Linear Cameras

To get the necessary resolution and speed, I upgraded to a Basler ruL2048-19gm. This camera is designed for industrial conveyor belts.

Technical Specifications:

• Sensor Size: $1 \times 2048$ pixels.
• Readout Speed: Nearly $19,000$ lines per second.
• Retail Price: \approx \700$ USD.
• Acquisition Cost: $1/10\text{th}$ of retail (via eBay).

Because the exposure time is so short (minimum $1/100\text{s}$), the system requires immense amounts of light. This limits my shooting to daylight hours; dark tunnels and dim stations are effectively off-limits.