College Basketball Computer Vision

Phase 1 – Proof of Concept

Use existing consumer grade camera
Process a recent game using cloud resources
Determine what is beneficial
Create rules for subjective values (contested shot, good pick...)
Manual effort to correct camera shortcomings: 8 hours for tracks and 4 hours for plays = 12 hours total

Multiple mounted camera infrastructure
Streamlined workflow
Less touches and effort
Practices could be processed
Manual effort improvement estimates: track fixes 1 hour, plays 2 hours = 3 hours total

Scouting – identify opponent tendencies
Portal / Recruiting – focused tracking of individual players. Other products
Pose technology – identify body parts to get detailed measurements for distance and movement
- Shooting technique
- Defensive distance for ball pressure

This is inteded to demonstrate the starting point for computer vision.
The focus is on fundamentals for shooting decisions, rebounding effort, and picks
Good and bad grades were subjectively decided and coaches need to give their definition
Grades can be team or individual, and rolled up to units
The Scorecard uses a simple +/- system
Picks were identified using mathematics and only for on ball screen. There are probably many more picks than in this review.
The consumer grade camera loses players so the data is not perfect
Spacing needs better cameras but shows how the offense flows.

Potential Enhancement