Title:

Event Detection in Sports-Related Sensor Data with Machine Learning

This research is being conducted in partnership with SPAITR (Sports AI Tracker), to train machine learning models to classify sports motion data without relying on assumptions relating to the capabilities of the users or sensors. SPAITR is a UNH undergraduate startup developing time-series motion-based tracking devices to make athletes’ analytics more accessible. This work utilizes accelerometer and gyroscope data, collected by the NeuroTM Lacrosse plug, with the goal of predicting how many shots occur in a given recording session. The dataset is composed of windows of 70 observations (each comprised of 2.8 seconds of activity) with 145 isolated shots and 145 windows of other lacrosse events: cradling, ground balls, etc. Decision Tree and Support Vector Machine (SVM) models were trained on flattened data to predict if the window overlaps with a shot event. These models performed with 100% accuracy on the test set, with precision and recall scores of 1.0. This research improves upon the previous shot detection approach and will be implemented by SPAITR in the continued development of their products.

Name

Comment*

Comments are viewable only by submitter

Captcha*