Capstone: Depth-Sensing Woodcarving Aid
Project Description
Sept 2025 - Apr 2026
Problem: Woodcarving Dexterity is a capstone project I built with a team of four to lower the barrier to entry for beginner woodcarvers. Learning woodcarving demands strong spatial reasoning and depth judgment that most beginners lack, and the fear of making an unfixable mistake discourages many from starting at all.
Solution: Our system turns woodcarving into a "trace the light" experience: it scans the wood stock with a 3D depth camera, compares the resulting point cloud against a user-uploaded CAD model, and calculates the difference in depth between the two. This difference is projected directly onto the stock as a colour map, showing exactly where and how deep to carve next.
Features: Other features of the project include an iterative closest point algorithm (ICP) used to initialize the coordinate system, adjustable height and origin position, and hand detection to automatically trigger a new scan once the user's hand moves out of the way.
Result: Through iterative design and repeated user testing, the project reached its final stage, where it won the N.S. Robertson Foundation Capstone Design Award in Systems Design Engineering.
Mechanical Iterations
CAD · Iterative Design · CNC · Fabrication
Approach: Multiple iterations were designed, built, and tested to continuously improve the design, with each iteration solving issues identified during testing of the one before it.
Main frame: The main prototype went through three iterations. The first was built quickly as a proof of concept. The second improved stability and added support for the projector's weight. The third introduced indication markings for the user, along with adjustability to accommodate different user preferences.
Camera-projector holder: The top camera-projector holder went through five iterations, not all of which were built. These iterations tested and optimized the tradeoff between keeping the projector close enough to the workspace and far enough to project into cuts the user had already made. The final iteration also minimized the distance between the camera's and projector's optical origins to reduce projection error, and was designed with maintainability in mind, keeping all bolts accessible for easy replacement and repair.
ICP Zeroing
Depth Sensing · Point Cloud Processing · ICP · Python · Troubleshooting
Why ICP: We used an iterative closest point (ICP) algorithm to detect the "L" corner piece and establish the system's origin. A simpler approach using pre-defined retro-reflectors was considered, but it would not have provided the depth information we needed for initialization. Using ICP instead allowed users to make adjustments and recalibrate the system automatically at the push of a button.
Challenges: Getting ICP to run consistently and accurately proved more difficult than expected. We had to segment out background points even when the base was not coplanar with the camera, and segment out the bolts, which were not included in the CAD file ICP used for comparison. Significant effort also went into optimizing the algorithm itself to reliably find the correct origin every time.
UI Development
Qt · Python
Flow: The user interface was designed to guide users through a clear, low-effort action at each stage, minimizing cognitive load. Users begin by uploading a CAD file, are walked through an onboarding sequence explaining cutting stages and recommended tools, and then reach a home screen where they can manually or automatically update the projected colour map. As they carve, the system provides success, warning, or error feedback depending on cut accuracy.
Customization: A key design decision was making the colour map fully customizable. During user interviews, we found no consensus on what an ideal colour map should look like, with preferences varying significantly from person to person. To account for this, the interface lets users adjust the colour map's palette and discretization, ensuring the depth cues remain clear and distinguishable regardless of individual preference, including for colourblind users.
More Projects
- Captured Point Cloud
- Onto Stock Projection
- Top Holder Stress Analysis
- Poster