By: Dylan S.
School: University High
Science Teacher: Valerie Thompson
This project explores the development of a two-stage computer vision and LSTM neural network program that can assist archers by giving them an evaluation on their release. The computer vision function was designed specifically to replace a wrist-worn device developed in a prior project. Computer vision detected the head and hands in videos of archery shots and the LSTM neural network used the positions of these to classify six different types of shots with above 95% accuracy. Using a crop-in algorithm, shot detection was capable of running on 30 frames per second video. The only archery subject was myself and hundreds of videos of me shooting were taken.
Archery is a sport that requires accuracy and consistency. An archer needs consistent form, and one of the most important parts of an archer’s form is their release. Last year, I built a wearable device with an accelerometer that used AI to evaluate my release with 94.9% accuracy in real time. To improve upon this, can computer vision replace the accelerometer and continue to perform at the same level of accuracy?
The computer vision model successfully replaced the accelerometer device and met the design criteria. It performed with above 95% accuracy on average for all six types of shots. Finally, with some modifications, I was able to get the program to run in real-time on my laptop. In the future, I can extend the evaluation to other parts of my form, including my draw and follow-through.