The Problem
A camera measures pixels; inspection requires physical dimensions. I developed the calibrated relationship between camera measurements, laser geometry, and the physical surface needed to reconstruct repeatable metric profiles.
- Camera + laser
- The baseline defines their relative position.
- Viewing ray + laser plane
- Their intersection locates the illuminated surface point.
- Reconstructed XYZ
- Calibration relates this location to physical geometry.
My Contribution
- Developed a calibrated laser-triangulation 3D sensing pipeline for additive-manufacturing inspection and real-time object tracking.
- Implemented and benchmarked distortion correction and subpixel laser-stripe extraction in Python/OpenCV.
- Evaluated sensor geometry and experimental conditions affecting reconstruction quality.
- Refined alignment and calibration through 20+ benchmark trials to improve measurement consistency.
Sensing Architecture
Reconstruction depends on the physical relationship between the camera, laser plane, and test surface. I refined sensor geometry and alignment alongside calibration so image-space localization could be related consistently to physical surface location.
Experimental photograph slot
Component labels can accompany the final image.
Image Processing Pipeline
I implemented and benchmarked distortion correction and subpixel stripe extraction in Python/OpenCV. Each processing step prepares a more consistent image-space measurement for reconstruction.
Camera pixels capture the illuminated surface.
Correct lens distortion before geometric reconstruction.
Isolate the stripe for consistent image-space measurement.
Estimate the stripe center at finer than integer-pixel resolution.
Subpixel Stripe Localization
Choosing the brightest integer pixel limits localization to the pixel grid. Estimating the stripe center from its intensity distribution provides finer image-space resolution and supports measurement consistency.
Explanatory diagram only · curves and pixel intensities are illustrative, not trial data or a specified extraction algorithm.
Calibration & Reconstruction
Calibration and optical geometry establish the transformation from image-space measurements to physical surface location. Distortion correction precedes triangulation so systematic image distortion does not propagate directly into the metric reconstruction.
Engineering Decisions
Subpixel localization instead of pixel-level detection
I worked on stripe-center estimation to obtain finer image-space localization than selecting an integer pixel, supporting consistent measurements across frames.
Correct distortion before metric reconstruction
I benchmarked distortion correction before reconstruction because lens distortion can introduce systematic spatial error into the camera-to-surface relationship.
Refine geometry and calibration experimentally
I refined camera/laser geometry, physical alignment, calibration, and extraction through repeated testing. Reconstruction depends on the sensing arrangement as well as the software pipeline.
Experimental Development
Early trials
- Observation
- Reconstruction and alignment inconsistencies.
- Engineering change
- Identify geometry, calibration, and extraction factors affecting consistency.
- Learning / evaluation
- Establish what needs refinement before repeated benchmarking.
Geometry / calibration refinement
- Observation
- Reconstruction links physical alignment with image processing.
- Engineering change
- Adjust sensor geometry, alignment, calibration, and stripe extraction.
- Learning / evaluation
- Evaluate the sensing arrangement and processing as one system.
20+ benchmark trials
- Observation
- Repeated reconstruction and tracking behavior.
- Engineering change
- Benchmark the refined pipeline under repeated experimental conditions.
- Learning / evaluation
- Evaluate reconstruction error, repeatability, tracking stability, and consistency.
Validation
20+ trials to evaluate reconstruction error, repeatability, tracking stability, and measurement consistency.
Validation data / error plot
Experimental dataset not supplied · no values plotted
From image data to metric geometry
I developed a calibrated pipeline converting 2D camera measurements into repeatable metric surface profiles. The work combined optical geometry, Python/OpenCV processing, and repeated experimental refinement to support additive-manufacturing inspection and real-time object tracking.