Event Data Recorder and Analyzer for Motorcycles using Low-Cost Inertial Measurements

Based on autarkic data from a low-cost inertial measurement unit - fixed onto- and power-supplied by the motorcycles battery - we reconstruct forward velocity and driving behavior of a motorcycle.

Steckbrief

  • Departement Technik und Informatik
  • Forschungsschwerpunkt Energie, Verkehr, Mobilität
  • Forschungsfeld Institut für Optimierung und Datenanalyse, IODA
  • Laufzeit 01.01.2010 - 01.04.2013
  • Projektverantwortung Kurt Hug
  • Projektleitung Kurt Hug
  • Projektmitarbeitende Roger Filliger
    Nathan Munzinger
    Simon Bayes
  • Mitwirkende Projektpartner Wirtschaft AXA Winterthur AG
  • Mitwirkende Projektpartner Forschungsinstitutionen inkl. BFH Topo - Geodetic Engineering Laboratory - EPFL
  • Schlüsselwörter Accident reconstruction, Sensor fusion, Motorcycle dynamics

Starting Point

Gaining objective insight into the actual driving behavior of vehicular traffic system users is a crucial issue for many private and governmental economic partners. One of the most important partners, the Automotive Industry, started such activities more than 40 years ago using Event Data Recorders (EDRs). Experimental EDR technologies for cars were developed with limited capacities as early as 1974. For powered two wheelers, no open-access EDR technology is available. The main reason for this absence is the much more involved dynamics of single track vehicles compared to car-dynamics.

SRF - Einstein Beitrag

Goals

Our aim is to develop low-cost EDR-technology for motorists which, for privacy reasons should be GNSS-free and MEMS-IMU based. Given these constraints, the focus is on the reconstruction of motorcycle forward speed – the most significant state variable for accident reconstruction.

Approach

Based on autarkic data from a low-cost, 6-axes inertial measurement unit (IMU), which is fixed onto- and power-supplied by the motorcycles battery, we reconstruct forward velocity and elementary driving behavior of a motorcycle using a Hidden Markov Model (HMM) combined with a navigation filter (Kalman filter) based on an Inertial Navigation System (INS). The later integrates accelerometer and gyroscope observations for retrieving vehicle orientation, velocity (and possibly position).

Solution

The stochastic errors affecting inertial sensors cause notorious drift problems which are mastered using two ingredients. First, the voltage ripples of the motorcycle’s battery is used as a stabilizing external signal to estimate speed. These measurements feed the navigation filter and finally bound the drifts. Despite the structural simplicity of this algorithm and the relatively low performance of the EDR IMU, the proposed off-line estimator is, after a short learning phase, accurate for a large class of motorcycles.  Second, a novel stochastic modeling approach enabling the estimation of more complex error models, has been developed, validated and applied to the EDR inertial sensors. This new framework enables to do correct stochastic assumptions during the navigation filter design and to significantly reduce drift.