DESIGN AND IMPLEMENTATION OF STRIDE CLASSIFICATION AND MACHINE LEARNING FOR RACEWALKING FAULT DETECTION

• Main Author: Daffa Daniswara, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81891
• Institution: Institut Teknologi Bandung

Racewalking, according to the definition of World Athletics, is when an athlete performs a step movement that meets several conditions. The first rule requires that the athlete must always have at least one foot in contact with the ground, so there should be no visible phase where both feet are airborne simultaneously, a violation known as "loss of contact." The second rule mandates that the supporting leg must remain straight from the moment it touches the ground until it passes beneath the body, prohibiting any bending at the knee during this phase, commonly referred to as "knee bend." To address this issue, a solution in the form of a wearable device is proposed, designed to assist referees in monitoring athletes' movements more effectively. This device uses an inertia sensor module to track the linear acceleration and angular velocity of the athlete's calf. By collecting and analyzing this data, the device can provide real-time feedback to the referee about the athlete's movements concerning the racewalking rules using a dashboard device. This thesis document focuses on the development and implementation of the main algorithm and machine learning model to classify racewalking movements based on the movements produced by the athlete's right calf. The main system developed consists of an algorithm on a microcontroller used for step detection using an inertia sensor and the creation of a machine learning model applied on the Amazon SageMaker cloud computing service. The designed and implemented system was tested and verified to meet the product specifications previously established. The step classification algorithm implemented has been verified to classify steps with an error of less than 8%. The designed machine learning model has been tested and verified, resulting in accuracy higher than the established specification of 70%. Based on the results of good system testing and verification, it becomes a tool that can assist referees in accurately classifying athletes' movements during competitions. For further development, it would be beneficial to involve more subjects with real racewalking competition experience, ensure balanced gender representation, and include a wider range of participant heights.