Ground Surface Pattern Recognition with Hidden Markov Models for Low Cost Positioning Improvement
We evaluate the feasibility of a low-cost solution to enhance the absolute positioning estimation of a vehicle using the example of a Pedelec (Pedal Electric Cycle). For this purpose, the acceleration sensor and the magnetometer of a smartphone's inertial measurement unit (IMU) are used to record feature sets of ground surface profiles. The feature sets are acquired by riding over certain spots that exhibit non-perfect road condition, e.g. a pothole, a utility hole or a lowered curb. Non-equal recording conditions are countered with the use of statistical methods, namely hidden Markov models (HMMs). An HMM is constructed for each spot described by the recorded ground surface profiles with the purpose of determining their geolocation, showing promising results with error rates in reasonable bounds. Consequently, new possible supporting points for recalibration of inertial navigation systems (INS) are obtained.
Martin Schweigler, Marco Grochowski, Sujan Tamrakar, Stefan Kowalewski