Line scan camera image processing for train photography
20 hours ago
- #line_scan_camera
- #image_processing
- #python
- Using an Alkeria Necta N4K2-7C line scan camera with a 4096×2 Bayer sensor.
- Detecting moving objects by computing an energy function to distinguish vertical structures from horizontal background stripes.
- Automated speed estimation by comparing green channels in the Bayer array and fitting a robust spline.
- Resampling images with a Hann window to avoid aliasing and flipping images when necessary.
- Demosaicing with bilinear interpolation to reduce fringing issues from the Bayer array.
- Removing vertical stripes caused by clock jitter and dark objects using linear regression and iteratively-reweighted least squares.
- Patch-based denoising to handle repeated textures in train images, though it is slow.
- Skew correction planned using Hough transform after speed estimation.
- Color calibration done by eye but looks decent.
- Implementation in Python with numpy, handling large data in chunks and mixed results with AI assistance.