Playing around with Mario Kart analytics.
- Given images taken during races, emit a JSON object describing the current state (list of players, each of which includes at least position, and ideally also whether the player has finished, the current lap number, character, and weapons). Also ideally include the current map.
- Given this data, bucket these into discrete races, identifying at least each player's finishing position.
- Given this data, plus metadata describing the starting conditions and changes in player <-> human mappings, compute per-human stats.
For data ingestion:
- Consider starting with only a description of the end state of each race, that includes date, race number, track name, player <-> character, and final positions.
- Add JPEG images taken at regular intervals. (Will it be necessary to also include a picture of the end state to identify the actual winner and to identify track changes?)
- Or: use video instead, which would allow arbitrarily fine-grained stats, plus the ability to measure actual times per player, allowing for all kinds of stats about times taken on each track.
- Main interface: web-based
- Restify for web backend, bunyan for logging
For processing images, libjpeg is probably the best bet (or libjpeg-turbo). There's already a node-jpeg, but it can only write files, not read them.