We instrumented everything. Every detection event writes a structured log line: timestamp, peak amplitude, estimated interval since the last dribble, and a confidence score from the detector. Every BLE connect and disconnect gets logged. Every filter state transition gets logged at debug level.
This wasn't just developer hygiene. It became a product feature. Coaches can export a session log as CSV and see not just *how many* dribbles a player did, but the cadence — whether they sped up, slowed down, or got sloppy as the drill wore on. One coach told us he could spot fatigue patterns in the interval data before he saw them in the player's body language.
We also used those logs heavily during tuning. When a player reported a missed count, we pulled the raw IMU trace from that window and compared it against the detector's output. In nine out of ten cases, the miss was a genuine edge case — a very soft dribble, or a crossover where the hand motion changed direction mid-bounce. We tuned the detector against real data instead of guessing.
Himanshu Dada
Discussions
Become a Hackaday.io Member
Create an account to leave a comment. Already have an account? Log In.