@inproceedings {pub3151,
	title = {An Approach for Automatic Riding Skill Identification - Methodology and first Results},
	author = {Nils Magiera AND Herbert Jan{\ss}en AND Hermann Winner},
	year = {2016},
	abstract = {Due to the static and dynamic instabilities of a motorcycle the rider performs a highly demanding control task during e.g. cornering, braking or a combination of both. Thus safety and comfort of the ride strongly depend on the personal abilities and skills of the rider. As a result existing Advanced-Rider-Assistance-Systems for powered two-wheelers try to improve both safety and comfort by taking rider characteristics into account and warn or intervene before the occurrence of critical situ-ations. E.g. Biral [1] determines rider characteristics from the distribution of achieved longitudinal and lateral acceleration during braking or cornering. However such characteristics don{\textquoteright}t take into account other qualities of the analyzed maneuver, e.g. smoothness and absence of corrective actions. 
In this paper we present methods to estimate the quality of a cornering maneuver and derive a per-sonalized riding skill model. The approach utilizes machine learning methods together with heuristic assumptions about the vehicle dynamics to detect and evaluate the quality of various cornering maneuvers from the onboard recordings of a motorcycle equipped with an Inertia Measurement Unit.  
Motivated by self-conducted test rides and different to [2] we propose to split the measurement data into segments that represent the control behavior of the rider rather than just cornering: roll-into-corner, stable-lean and roll-out-of-corner. First we use roll angle and roll rate, derived by an extended Kalman-Filter to identify stationary and non-stationary segments in the cornering maneuvers. To solve this objective several models including explicit rule based models and Hidden Markov Models (HMM) are introduced. Additionally, experienced riders labeled video recordings, which were used to evaluate the performance of the models. We show that a segmental HMM approach achieves the best results and outperforms explicit rule based models according to several evaluation criteria. 
Based on the results of the segmentation we discuss multiple methods to calculate scores for the quality of cornering maneuvers and an overall personal rider skill score. We verify the usefulness of our scoring by comparing three riders with different mileage and experience: a novice, a casual and a professional rider. 
We think our work can be employed for applications such as rider-self-training and as support for rider education.

[1] F. Biral et al., {\textquotedblleft}An intelligent curve warning system for powered two wheel vehicles,{\textquotedblright} European Transport Research Review,  vol 2, no. 3, pp. 147-156, Aug. 2010.
[2] Yoneda, Keisuke et al., {\textquotedblleft}Rider Characteristics Assessment Device and Straddle-Ridden Vehicle Provided Therewith,{\textquotedblright} EP 2517952A1, Oct., 31, 2012.
},
	publisher = {Institut fuer Zweiradsicherheit},
	booktitle = {Ifz Konferenz 2016}
}