Olympic show jumping consists of horse and rider pairs competing over a course of 12-17 jumps that are between 1.40 and 1.65 meters tall. Faults occur when the horse knocks down the obstacle, refuses to jump the obstacle, or steps a foot in a water obstacle, as well as if the pair exceeds the allotted time or the rider falls. Eliminations occur after a second refusal, fall of the horse, or the second time the rider falls. Using Fédération Équestre Internationale (FEI) data, several studies have classified when faults and falls are most likely to occur based on the type of jump, whether the jump is in a turn, how late in the course a jump occurs, etc. However, the FEI data only lists total faults, not when the faults occurred, and the studies have relied on hand-coding from videos.
This project will first train a model to detect knock downs from event footage, then hopefully use the tool to analyze the faults at an FEI event. While live replays are available on the FEI YouTube channel, there may be a terms of service issue around using these videos.

Video overview of the sport https://www.youtube.com/watch?v=9N5nnon1Qbw
Paper exploring where faults are most likely to occur, with diagrams of different jump types https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8926200/


If you have an idea for another sport, and a data set is publicly available, that is another option for the project.