the problem
Pipe measurements are still often collected manually. The work is slow, difficult to reproduce and disconnected from the digital models that depend on it.
01 · AI · LiDAR · iOS · prototype
teaching an iphone
to understand pipes
master’s thesis2026Swift · Python
PipeDetect began as my master’s thesis and turned into an experiment in making the invisible measurable.
the idea
What if the device already in someone’s pocket could capture enough spatial information to make part of that process visible? PipeDetect uses an iPhone as both sensor and interface.
where ai helps
Computer vision locates the pipe in the camera image. A YOLO model provides the first useful clue: not the final answer, but a region in which geometry can begin working.
where geometry takes over
LiDAR depth, camera calibration and geometric modelling turn a two-dimensional detection into measurements and a reconstruction in space.
building the app
The iOS application brings capture, feedback and results into one flow using Swift, SwiftUI, ARKit and CoreML, with Python supporting the model experiments.
what worked
The project proved that AI and spatial sensing become more useful when neither is asked to solve the entire problem alone.
what didn’t
Real-world surfaces, partial visibility and noisy depth data made clean laboratory assumptions disappear quickly, which became one of the most valuable parts of the work.
what i would do next
Test across more pipe types and environments, improve uncertainty feedback, and make the reconstruction easier to inspect outside the capture moment.