Computer vision
MATLAB · camera calibration · image processing
Built from background subtraction and morphology to a calibrated HSV detector for six pin positions.
Final vision & integration reportA vision-guided bowling robot
The Big Lebowski.
01 / The complete machine
Vision-guided. Feedback-controlled.
A tabletop robot that detects pins, positions a ramp and releases a steel ball.
MATLAB. Simulink. Mechatronics.
Avi Patel · Team 19. Computer vision, encoder feedback and mechanical integration.
Camera → MATLAB → Simulink → Arduino → motion
Scroll to explore3D reconstruction from team drawings & photographs
The complete machine
Vision-guided. Feedback-controlled.
A tabletop robot that detects pins, positions a ramp and releases a steel ball. Avi Patel · Team 19. Computer vision, encoder feedback and mechanical integration.
Camera → MATLAB → Simulink → Arduino → motionPosition & release
The ramp makes the move.
A rack and pinion slides the ramp sideways. Encoder feedback brings it to the selected position. A small servo moves the release dowel. The steel ball rolls down the V-channel and into the lane.
16-tooth gear · 8 mm guide rods · 5/8-inch steel ballCamera & calibration
Six positions. One overhead view.
The camera checks color saturation and brightness at six calibrated pin locations. A fixed camera and ring light help the software distinguish occupied positions from empty ones.
Logitech Brio · controlled lighting · calibrated pin occupancyThe team’s engineering work
Skills you can
see in the build.
Five students connected sensing, software and mechanisms to aim and release a bowling ball. The operator places the ball and pins.
Controls & embedded systems
Simulink · Arduino Mega · encoders · motor control
Mapped image coordinates to rack travel, tuned the position loop, and integrated motor feedback with servo release.
Controls & integration reportMechanical integration
Fusion 360 · 3D printing · rack and pinion
Combined a translating ramp, guide rods, release gate, camera fixture and containment into one working prototype.
Design drawings & assemblyEngineering validation
DFMEA · concept comparison · staged testing
Documented design tradeoffs, compared controller responses, resolved lighting problems and evaluated the integrated build.
Development & risk analysisAvi Patel’s documented contributions
MATLAB/Simulink work, early camera interfacing and technical documentation within Team 19. The complete machine and demonstration are team achievements.
Explore the source evidence04 / The logic
A decision becomes
a physical shot.
Read the pins. Choose a target. Move the ramp. Release the ball. Then look again.
- 01
Sense
Check six calibrated positions.
- 02
Choose
Select a target manually or automatically.
- 03
Position
Move the ramp with encoder feedback.
- 04
Release
Open the servo gate.
- 05
Refresh
Recenter and capture the remaining pins.
CALIBRATED PIN POSITIONSTOP VIEW
BALL DIRECTION
Present Selected target Absent
TRY THE TARGETING LOGIC
Which shot
would it choose?
Toggle the pins to change the lane. The archived app scores a fixed set of expected knockdowns and picks the occupied target with the highest score.
Logic demonstration based on the archived MATLAB app. Expected knockdowns come from a lookup table; they are not a physics prediction. Ties select the first numbered target.
Under the surface
Three disciplines,
one conversation.
The image becomes a coordinate. The coordinate becomes a motor command. The mechanism turns it into motion.
CALIBRATION → OCCUPANCY
Six known locations.
One clear decision.
The operator marks the lane edges and six pin dots. The app inspects a local image patch at each location, converts it to HSV, and checks mean saturation and brightness.
- Patch radius
- 30 pixels
- Saturation threshold
- > 0.70
- Brightness threshold
- > 0.40
These are archived code settings. The final detector uses saturation and brightness at calibrated points; it does not use a trained model.
05 / The build
It began with
a single servo.
Four stages of iteration: from a simple signal path to a complete camera-guided machine.
19 SEP 202501Interface
Start with the signal path.
A potentiometer input commanded servo motion through MATLAB, Simulink and an Arduino. The first stage established communication, sensing and actuation before the robot took shape.
The foundation: connect a physical input to a controlled output.
Read the stage report10 OCT 202502Perception
Make the scene readable.
A fixed camera, PVC frame and controlled lighting supported background subtraction, morphology and color-and-shape measurements. Colored stickers made the sensing pipeline visible and testable.
Lighting and fixture stability mattered as much as the image-processing code.
Read the stage report31 OCT 202503Feedback
Connect vision to motion.
A MATLAB interface placed selectable buttons at detected object centroids. Choosing a target rotated a motor-driven pointer toward it, using encoder feedback and a tuned position controller.
Lane reflections disrupted sensing; the team used a white surface during the demonstration.
Read the stage report07 DEC 202504Integration
Turn the pointer into a bowler.
The final design replaced angular aiming with a laterally translating ramp. A servo gate, rack and pinion, ring light, rails and catch box brought the complete bowling sequence together.
Calibrated pin occupancy replaced the earlier general background-subtraction method.
Read the stage report06 / In motion
The workbench.
The actual machine.
The original final demonstration: a three-pin arrangement is cleared, and the MATLAB interface updates the board.
The reported result
From the team’s final presentation, slide 6.
95%
accurate pin hits
20
consecutive frames
Team-reported results, without raw scored trial logs in the archive. The team reported no troubleshooting intervention during those frames; ball and pin setup remained manual. The video is one recorded sequence. Read the presentation ↗
07 / The record
Follow the work
all the way back.
Original reports, source code, drawings and footage. The engineering behind every part of the story.
01PDF02PDF03ZIP04BLEND
Final engineering report
Design decisions, dimensioned drawings and the integrated system.
Final presentation
Requirements, demonstration and team-reported results.
MATLAB + Simulink
Archived application, motor model and development files.
The Blender assembly
Editable reconstruction, with inferred layout documented separately.
Reports and slides open as PDFs. Original editable files remain in the archive.
53 files. Four stages. The full archive.Prototype boundaries & source notes +
Operation. The operator places the ball and pins; the robot handles aiming, release and return. Detection depends on six calibrated locations and stable camera and lighting conditions.
Timing. The presentation reports a cycle under 10 seconds. The archived app contains 12 seconds of fixed waits before image refresh. The version and timing boundary remain unresolved.
3D reconstruction. This model uses original drawings where available and inferred layout elsewhere. The report’s illustrative 6-inch lane differs from the app’s 3.85-inch calibration setting. Read the modeling notes ↗ Download the GLB ↗
Reproducibility. These source files preserve the development record. The Simulink model references an encoder support component that is not separately included; the hardware has not been rerun here.
Clemson University / Team 19
Built together.
Corey Golec · John Phillips · Jacob Criswell
Avi Patel · Josue Morales-Sanchez
A five-person ECE 4950 project, Fall 2025. Avi’s documented contributions include MATLAB/Simulink, early camera interfacing and technical documentation.
The project builds on course-provided vision code and Simulink models. Original photography, documents and footage belong to Team 19.
Original team site