An Intelligent System to Analyze Sketched Solutions to Open-Ended Truss Problems
summary
A web app recognizes hand-drawn 2D truss sketches using shortstraw segmentation plus a substroke graph, computes logical node coordinates via breadth-first search and rotation matrices, and solves beam forces with LU decomposition and superposition to check length and stress constraints. The system analyzes all student submissions correctly, is evaluated on 3000 sketches with f-score 0.997, and is tested in a class study (58 students, 16 in focus groups), with most recognition failures traced to user drawing errors and students preferring the immediate feedback.
AI-written summary · about
chapters (10)
- 0:00Introduction to the project
- 1:10Limitations of existing systems
- 2:54Truss recognition process
- 4:17Logical coordinates and constraints
- 5:44Static determinacy and equilibrium
- 6:46Solving system of equations
- 7:51Student evaluation and feedback
- 9:18Recognition issues and improvements
- 9:58Algorithm accuracy results
- 10:20Conclusion and future work
related talks
22mDreamSketch: Early Stage 3D Design Explorations with Sketching and Generative Design
21mSketchingWithHands: 3D Sketching Handheld Products with First-Person Hand Posture
17mSweepCanvas: Sketch-based 3D Prototyping on an RGB-D Image
14mAutocomplete Animated Sculpting
20mInteractive Room Capture on 3D-Aware Mobile Devices
16mEye-Trace: Segmentation of Volumetric Microscopy Images with Eyegaze
20mMoSculp: Interactive Visualization of Shape and Time
40mBuilding a Handwriting Recognition System