Page 44 - MSDN Magazine, November 2018

ARTIFICIAL INTELLIGENCE
Sensors in Sports: Analyzing Olympic Diving with Sensors and Vision AI
Kevin Ashley, Phil Cheetham, Olga Vigdorovich, Dan Laak, Kevin Kang, Daria Fradkin
Internet of Things (IoT) and Machine Learning (ML) have been used for everything from inventory tracking and manufactur- ing quality control to traffic management and weather forecasting. One area where these two disciplines have come together to herald big changes is the arena of high-stakes sports. In the April issue of MSDN Magazine (msdn.com/magazine/mt846466), we explored how sensors are being used to track and improve performance in Winter Olympic sports events like alpine skiing. Now in this second installment, the focus shifts to the Summer Olympics, where IoT and ML are being leveraged to improve springboard diving.
It’s an area of urgency for the U.S. Olympic Team. There are four diving events—3-meter springboard, 10-meter platform, 3-meter synchronized diving and 10-meter synchronized diving—that account for 24 total available medals in each Olympics. Sensors and ML promise to improve training regimes and maximize athlete performance on the biggest stage.
Today, video is the predominant method of analysis in diving, as it has been for years. Coaches use slow motion and frame- by-frame video to analyze dives and give immediate feedback. However, this qualitative approach does not allow rigorous anal- ysis and effective comparison of performance changes. To truly measure and document changes, quantitative analysis that marries video with data captured from specialized sensors is needed.
In this project, we use video and sensor technology to measure the diving takeoff—the critical first moments that are key to a suc- cessful dive. We designed a sensor to be placed underneath the tip of the springboard. It includes an inertial measurement unit (IMU), an accelerometer and a gyroscope to measure the import- ant characteristics of the springboard takeoff. This includes the approach to the end of the board, the hurdle onto the end of the board, the pressing down and flexing of the springboard, and the lift into the air by the springboard. A great takeoff is critical for a great dive, with maximum height and spin needed for incredibly difficult dives like a forward 4 1/2 somersault.
Sensor data can determine if the board was maximally depressed on the way down, and if the diver rode the board all the way back up and waited for it to throw him or her into the air. Many divers don’t do this well, and tend to “skip off the board” on the way up, losing height and rotation in the process. This project aims to build a system that gives divers and coaches the information they need to rapidly improve the quality of the takeoff, allowing difficult and more difficult dives to be performed successfully.
This article discusses:
• Combining IoT and video for sports movement analysis
• Using the OpenCV computer vision library for video analysis
of athletes
• Analyzing high-performance sports with sensors Technologies discussed:
SensorKit, OpenCV, R, C#
38 msdn magazine