Author Identifier
Benjamin Chalk: https://orcid.org/0000-0002-4943-2045
Date of Award
2026
Keywords
Roll, cycling, bicycle, Inertial Measurement Unit, IMU, sensorisation, wavelet, bicycle roll dynamics, signal superposition, field-based measurement, kinematic analysis, in-saddle, out-of-saddle, seated, standing, sway, lateral lean, manoeuvring, path
Document Type
Thesis - ECU Access Only
Publisher
Edith Cowan University
Degree Name
Doctor of Philosophy
School
School of Engineering
First Supervisor
Nando Guzzomi
Second Supervisor
Chris Abbiss
Abstract
Whilst data analysis continues to play an ever-increasing role in athlete development and assessment, the kinematics of elite cyclists in real-world conditions remain poorly quantified due, at least in part, to the absence of a non-intrusive method for continuous, field-based assessment of the bicycle-cyclist system. This research develops and validates a minimal sensorisation framework that infers bicycle-cyclist state from a single bicycle-mounted IMU without requiring athlete-mounted instrumentation. This thesis builds upon sensorisation lessons and practices in road cycling, where the management of data collection (e.g., power output) is largely considered successful as mechanics and other support staff are mainly responsible for sensor management and data collection.
However, single-sensor configurations are inherently vulnerable to signal mixing, where multiple kinematic processes imprint upon the same measured axis. This research addresses this issue by conceptualising bicycle roll as the superposition of kinematic components including, path (e.g., cornering or straight), manoeuvring (e.g., reactive control), cadence-induced oscillation modulated by stance, and noise. For practical inference, roll is decomposed into lateral lean, a low-frequency component capturing path and manoeuvring, as well as lateral sway, a cadence-locked oscillation where amplitude is modulated by cyclist stance (e.g., in-saddle, transition and out-of-saddle). In this thesis, these terms are used to describe properties of the decomposed signal rather than direct labels for underlying mechanics.
The theoretical framework proposed in this research has been evaluated across three settings that are progressively more ecologically valid, ranging from: controlled indoor testing, through to an outdoor circuit, and lastly, to professional criterium racing. Across the three settings, the application of frequency- and time-frequency analysis of bicycle roll reveals: a three frequency-band structure, with path-dominant energy below 0.20 Hz, manoeuvring between approximately 0.20 and 1.0 Hz; and a cadence-dominated lateral sway above 1. This structure supports the separation of gross kinematic components using data from a single frame-mounted sensor.
In this context, this research leverages the coordinated motion of the bicycle arising from feedforward and feedback control interactions of the bicycle-cyclist system to infer the cyclist’s gross stance, without requiring athlete-mounted instrumentation. Change-point analysis was used to operationalise the categorisation of three separate gross stances positions, including in-saddle (< 2.13°), transition period (> 2.13° to < 4.25°) and out-of-saddle ( > 4.25°). In professional racing, linear thresholding of lateral sway, using these thresholds, achieved sensitivity and specificity greater than 0.99 against broadcast video. At the same time, the results revealed that cyclists spend a substantial proportion of time in a distinct transition period, which possesses its own characteristics.
Collectively, these findings provide a minimal-sufficiency validation of the proposed theoretical framework, and demonstrate that bicycle-derived measures can infer gross cycling kinematics, including gross cyclist stance, without requiring athlete mounted instrumentation. This creates new possibilities for in-race kinematic analysis, presenting options that do not interfere with the cyclist, whilst also filing gaps in cycling monitoring, which currently focus on capturing energetic and physiological signals, rather than the kinematic context in which that signal was produced.
Recommended Citation
Chalk, B. A. (2026). From bicycle roll to cyclist state: Method development and validation of a single bicycle-mounted IMU for inferring road-cycling kinematics. Edith Cowan University. https://doi.org/10.25958/44mp-6s19