Automatic Recognition of Safety and Performance Related Activities in Motocross

Master thesis (2017)

Authors

B. Breider Mechanical Engineering

Contributors

J.C. van Gemert Pattern Recognition and Bioinformatics - EEMCS (mentor)
F.C.T. van der Helm Biomechatronics & Human-Machine Control - Mechanical, Maritime and Materials Engineering (mentor)
J.C.F. de Winter Human-Robot Interaction - Mechanical, Maritime and Materials Engineering (graduation committee member)
Faculty
Mechanical Engineering, Mechanical Engineering
Copyright: © 2017 Bas Breider
More Info
expand_more

Published Date

28-08-2017

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Mechanical Engineering

Abstract

Motocross is a popular, but dangerous sport: improvements in performance and safety should be made to make it more attractive and less dangerous. By automatically recognizing activities of the rider on the track, riders can be informed about dangerous situations, and fans can be provided with insights into the performance of the riders. The goal of this study is to develop and validate an automatic activity recognition methodology that can determine safety and performance related activities in motocross. A 3D accelerometer and gyroscope were used to collect movement data of the rider and motorcycle.
Time and frequency domain features were extracted and used to evaluate several machine-learning classifiers: decision tree, knearest neighbor model, support vector machine, and multilayer perceptron neural network. These classifiers were evaluated based on accuracy, precision, recall, and speed to show overall
classifier performance in real time, and to identify classification patterns for individual activities. The results were validated for multiple riders at different types of motocross tracks to test generalizability of the approach. Overall accuracy showed no large differences between the individual classifiers (74%-78% ± 6.8%). Similar results were found when the approach was validated with new riders and tracks (73%-79% and 68%-72%). The neural network classifier showed the highest precision for the safety related activities: stopping and falling (82%-95%). However, low precision was found for the performance related activities: jumping, turning and driving straight (20%-78%). To conclude, the neural network approach can be used for the detection of safety related activities, but more data of different riders is needed to confirm the proposed approach.