Protocol for simulation of falls in watersports activities
Falls are a frequent cause of unintentional injuries. The development of a water sports fall detection algorithm usually requires a dataset collection of fall events that are difficult to replicate in a laboratory environment. To address that problem, this article proposes a simulated boat falls protocol that can be used to record a dataset of falls for various water sports. A dataset of 296 samples comprised in 129 falls and in 167 non-fall events was gathered using the protocol. The data collection was made with 3 different smartphones and with 1 external IMU. A fall detection algorithm was trained with the dataset with machine learning techniques and tested over the same dataset and real sailing data. The algorithm achieved 99.9% of accuracy, 99% of specificity and 100% of sensitivity when tested in the dataset and detected the only fall that occurred in one hour and a half of a real sailing activitiy.