Computer Science Graduate Seminar

Wednesday, May 05, 2021, 11:00am

Ground Surface Pattern Recognition for Enhanced Navigation



With the continuous increase in sales of electrical assisted bicycles over the last decade, the number of bicycle accidents across Europe has simultaneously grown significantly. At the same time the technology lacks on active safety systems, even though the electrification of the so-called Pedelecs would allow their development. This dissertation can be seen as the first step in the process of developing position and situation dependent active safety systems by improving the position determination accuracy of bicycle navigation systems. 

In the core of this work a position estimation system is developed, which uses road sections with significant surface conditions to improve the positioning accuracy of a conventional GNSS/INS. Based on the vertical accelerations acting on the moving Pedelec, the system recognizes individual spots in the road surface, e.g. manholes or potholes. To be more precise, the individual acceleration profiles that occur when passing different spots, are recorded with a smartphone and statistically modeled offline with the help of continuous hidden Markov models during the training phase. In online mode, the trained models are then used to recognize the spots by the acceleration profiles of the revisited road sections. The absolute positions of the Pedelec, relative to the global coordinates of the recognized spots, are subsequently determined by an inertial calculation of the distances traveled in the time between their detection and classification. The system thus uses statistical models to estimate the absolute position of the Pedelec and is consequently called Statistical Absolute Position Estimator, or SAPE.

In the second part of this work, SAPE is used to develop a navigation system, which shows the potential of the ground surface pattern recognition. For this purpose the SAPE and GNSS position determinations are fused with an inertial navigation system using an extended Kalman filter. Since the inertial sensors provided by the chosen smartphone are not accurate enough to realize a stand-alone INS, an odometry is developed and implemented to support the navigation solution. The resulting GNSS, SAPE and odometry supported INS is finally evaluated using an RTK GNSS and its accuracy is compared to that of a conventional odometry supported GNSS/INS created with the same low-cost hardware.


The computer science lecturers invite interested people to join.