
Data driven bike fitting
- Author
- Jarich Braeckevelt
- Organization
- Abstract
- During bike fitting sessions, the optimal cyclist’ position is determined. Finding this ‘optimal’ cycling position is often a time-consuming and labor-intensive process, i.e., a standard bike fitting procedure takes at least two hours when done by an expert bike fitter. The best position is a combination of comfort, performance and injury prevention [1,2,3,4]. To date, however, bike fitting suffers from expert ‘subjectivity’ as there is no consensus among bike fitters on which parameters to focus on. In order to examine this hypothesis about expert subjectivity, a reproducible test methodology is currently under development to perform and evaluate a bike fit by a group of independent experts in Flanders. Results of these tests will be discussed during the presentation. To solve the expert subjectivity problem, and to improve the overall fitting process, we started a research project to develop a methodology to perform automatic bike fitting based on novel data-driven decision-making processes. The data is provided by the Bioracer Motion mo-cap system (shown in Figure 1), which consists of 2 arrays of high-speed IR cameras which capture positional data of the active infrared markers which are placed on the cyclists’ body. Up till now, mainly feature engineering techniques were studied and evaluated on the Bioracer Motion datasets. Preliminary experiments focusing on saddle height optimization already show the feasibility of the proposed methodology. Saddle height is a determining factor in knee injuries [5,6,7] and the outputted power [8]. However, it is important to mention that saddle height optimization is only a small step in the bike fitting process [9]. In these experiments our methodology was to compare three different bike configurations (i.e., saddle too high, too low and the 'optimal' position) for different pairs of markers. An example of these spatio-temporal comparisons is shown in Figure 2. This graph shows the relation between the crank angle and the right knee Z speed over time. A good feature to track would be the occurrence of the minimum with regard to the crank angle. If the saddle is in a position that is too high, for example, the minimum occurs at a particularly lesser crank angle. Several of these kind of features are evaluated on the Bioracer Motion dataset. The lesser false positives, the higher the weight of this feature. In the end, a series of 8 features (focusing on the left/right foot and knee movement in X/Y direction) are fed into a weighted feature sum, based on which the saddle height correction is suggested. This methodology results in a 100% correct saddle height up to an accuracy of 5mm for a test set of 40 fits. Future work will include other optimizations (i.e. saddle setback, handlebar drop, ... ) and evaluate machine learning techniques to determine new features. The final goal of our project is to have a fully autonomous bike fitting system, which can fit a cyclist with sufficient accuracy in a short period of time. This system will have a significant impact on the cycling world, as less knowledge will be required to successfully fit cyclists.
- Keywords
- Bike fitting, Data-analytics, Feature learning
Downloads
-
(...).pdf
- full text
- |
- UGent only
- |
- |
- 667.19 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-8591301
- MLA
- Braeckevelt, Jarich. “Data Driven Bike Fitting.” Science and Engineering Conference on Sports Innovation, Proceedings, 2018, pp. 47–49.
- APA
- Braeckevelt, J. (2018). Data driven bike fitting. Science and Engineering Conference on Sports Innovation, Proceedings, 47–49.
- Chicago author-date
- Braeckevelt, Jarich. 2018. “Data Driven Bike Fitting.” In Science and Engineering Conference on Sports Innovation, Proceedings, 47–49.
- Chicago author-date (all authors)
- Braeckevelt, Jarich. 2018. “Data Driven Bike Fitting.” In Science and Engineering Conference on Sports Innovation, Proceedings, 47–49.
- Vancouver
- 1.Braeckevelt J. Data driven bike fitting. In: Science and Engineering Conference on Sports Innovation, Proceedings. 2018. p. 47–9.
- IEEE
- [1]J. Braeckevelt, “Data driven bike fitting,” in Science and Engineering Conference on Sports Innovation, Proceedings, Groningen, 2018, pp. 47–49.
@inproceedings{8591301, abstract = {{During bike fitting sessions, the optimal cyclist’ position is determined. Finding this ‘optimal’ cycling position is often a time-consuming and labor-intensive process, i.e., a standard bike fitting procedure takes at least two hours when done by an expert bike fitter. The best position is a combination of comfort, performance and injury prevention [1,2,3,4]. To date, however, bike fitting suffers from expert ‘subjectivity’ as there is no consensus among bike fitters on which parameters to focus on. In order to examine this hypothesis about expert subjectivity, a reproducible test methodology is currently under development to perform and evaluate a bike fit by a group of independent experts in Flanders. Results of these tests will be discussed during the presentation. To solve the expert subjectivity problem, and to improve the overall fitting process, we started a research project to develop a methodology to perform automatic bike fitting based on novel data-driven decision-making processes. The data is provided by the Bioracer Motion mo-cap system (shown in Figure 1), which consists of 2 arrays of high-speed IR cameras which capture positional data of the active infrared markers which are placed on the cyclists’ body. Up till now, mainly feature engineering techniques were studied and evaluated on the Bioracer Motion datasets. Preliminary experiments focusing on saddle height optimization already show the feasibility of the proposed methodology. Saddle height is a determining factor in knee injuries [5,6,7] and the outputted power [8]. However, it is important to mention that saddle height optimization is only a small step in the bike fitting process [9]. In these experiments our methodology was to compare three different bike configurations (i.e., saddle too high, too low and the 'optimal' position) for different pairs of markers. An example of these spatio-temporal comparisons is shown in Figure 2. This graph shows the relation between the crank angle and the right knee Z speed over time. A good feature to track would be the occurrence of the minimum with regard to the crank angle. If the saddle is in a position that is too high, for example, the minimum occurs at a particularly lesser crank angle. Several of these kind of features are evaluated on the Bioracer Motion dataset. The lesser false positives, the higher the weight of this feature. In the end, a series of 8 features (focusing on the left/right foot and knee movement in X/Y direction) are fed into a weighted feature sum, based on which the saddle height correction is suggested. This methodology results in a 100% correct saddle height up to an accuracy of 5mm for a test set of 40 fits. Future work will include other optimizations (i.e. saddle setback, handlebar drop, ... ) and evaluate machine learning techniques to determine new features. The final goal of our project is to have a fully autonomous bike fitting system, which can fit a cyclist with sufficient accuracy in a short period of time. This system will have a significant impact on the cycling world, as less knowledge will be required to successfully fit cyclists.}}, author = {{Braeckevelt, Jarich}}, booktitle = {{Science and Engineering Conference on Sports Innovation, Proceedings}}, keywords = {{Bike fitting,Data-analytics,Feature learning}}, language = {{eng}}, location = {{Groningen}}, pages = {{47--49}}, title = {{Data driven bike fitting}}, url = {{http://ssig.nl/wp-content/uploads/2018/04/Abstract-book.pdf}}, year = {{2018}}, }