← Back to work
02 · Undergraduate Research, Purdue ME

Basketball Shot Biomechanics

A self-built, dual-camera computer-vision rig that extracts joint-angle kinematics from a basketball shot and separates makes from misses — validated on Purdue Men's Basketball players, including 2023 National Player of the Year Zach Edey.

Role
Co-developed CV pipeline, camera calibration study & trajectory simulator
Timeframe
2022–23, published April 2023
Tools
MATLAB, image/video processing, kinematics modeling
Advisor
Prof. Euiwon Bae, with Chase Martin & Ryan Carter
Photo of a basketball player mid-shot with green tracking markers on shoulder, elbow, and wrist, alongside a schematic diagram of the tracked joint angles and pair distances.
Fig. 1 — Marker tracking output from the custom MATLAB pipeline, mid-release.Subject: Purdue Men's Basketball

The problem

Detailed shot-biomechanics analysis usually requires a marker-based motion capture lab most players will never have access to. We set out to see how far a near-zero-cost setup — two smartphones and colored stickers — could get toward the same insight: what separates the joint angles of a make from a miss.

Approach

Two phones positioned at 90° to each other recorded each shot, with green stickers marking the shoulder, elbow, wrist, hip, knee, and ankle. My contribution was the tracking algorithm itself: a custom MATLAB computer-vision pipeline that isolates the markers frame-by-frame and computes joint angles via dot-product geometry and kinematic equations of motion.

Before collecting real data, I ran a calibration study comparing RGB and HSV color-tracking against camera distance (8 distances, 10–25 ft) to find the setup that minimized error — RGB tracking won decisively, with 11–13.7% average error versus 223–241% for HSV, at lower memory and processing latency. That put the optimal camera distance at 15 ft, holding tracking error under 3% of true height and limb length.

I also built a kinematic trajectory simulator — including backboard-bounce physics — that predicts whether a shot goes in purely from release angle, velocity, and height, so we could sanity-check the tracked data against physical expectation.

Results

We collected full joint-angle and velocity datasets — elbow angle, knee angle, forearm angle, release angle, ball velocity — across six subjects spanning skill levels, including four Purdue Division-I players (Zach Edey, Lance Jones, Chase Martin, Carson Barrett), for both free throws and three-pointers, splitting makes from misses. 3D scatter plots of the resulting data showed visible separation between makes and misses on a per-player basis, and the project was featured by Purdue Athletics and Mechanical Engineering as a research profile.

11–13.7%
Average tracking error, RGB method (vs. 223–241% for HSV)
15 ft
Optimal camera distance — under 3% error in height/limb-length
6
Subjects tracked, including four Purdue D1 players

Why this matters for sports tech

This project is where the sensor-and-biomechanics instincts behind Shot‑Sync actually came from — same joint-angle variables, same collaborators, and the direct predecessor to a hardware product. It's also a small case study in getting real signal out of a cheap sensor setup: something equipment R&D teams care about constantly, whether the sensor is a $30 IMU or two iPhones on tripods.

← Artemis Shot‑Sync DFM/DFA: Injection-Molded Cube →