Positioning in Real-Time Public Transport Navigation
Transcription
Positioning in Real-Time Public Transport Navigation
POSITIONING IN REAL-TIME PUBLIC TRANSPORT NAVIGATION DRESDEN Comparison Of Vehicle-based And Smart-phone Generated Acceleration Data To Determine Motion States Of Passengers Dipl.-Ing. Ina Partzsch (Fraunhofer IVI) Dipl.-Ing. Gunther Dürrschmidt (TU Dresden) Prof. Dr.-Ing. Oliver Michler (TU Dresden) Dr.-Ing. Georg Förster (Fraunhofer IVI) © Fraunhofer Overview Motivation: Enhance positioning in difficult urban mobility scenarios Theory: From raw data to signal features of motion states Application: Recognizing vehicle data in smartphone data Outlook: Recognizing all motion states of passengers © Fraunhofer Motivation EU-Project SMART-WAY: Problem Airport Dresden ?? ?? ? Fraunhofer IVI, Zeunerstrasse 38 © Fraunhofer Motivation EU-Project SMART-WAY: Solution Project Goals Proper navigation in public transport like a car navigation considering current events and delays Continous guidance to a final destination even with breaks by the user Application of Satellite Navigation Systems as a basis technology plus ITCS and current smartphone-localisation standards © Fraunhofer Motivation Urban Canyons, Tunnels and Shielded Vehicles Tunnel, shielded vehicle source: http://www.badische-seiten.de/freiburg/loretto-tunnel.php Urban Canyons source: http://autoelectronics.com/telematics/navigation_systems/GPS_2.jpg Metro source: http://kyynel.biz/photography/ © Fraunhofer Motivation Smartphone Sensors Source: http://www.prlog.org/10792126/1 © Fraunhofer Overview Motivation: Enhance positioning in difficult urban mobility scenarios Theory: From raw data to signal features of motion states Application: Recognizing vehicle data in smartphone data Outlook: Recognizing all motion states of passengers © Fraunhofer From Rawdata to Signal Features of Motion States General Data collection Data preproccessing Feature generation Feature extraction Signal classification © Fraunhofer From Rawdata to Signal Features of Motion States Data Preprocessing Determine sample rates of signals to be compared for a defined period Interpolate smartphone signal to retrieve a time-equidistant data set Re-sample the smartphone signal at the sample rate of the vehicle measurement system Use a low-pass filter in order to avoid frequencies introduced by the re-sampling process © Fraunhofer From Rawdata to Signal Features of Motion States Feature Generation Time-Domain statistical metrics correlation metrics signal characteristics Frequency-Domain key coefficents energy entropy Cepstral-Domain © Fraunhofer From Rawdata to Signal Features of Motion States Signal classification statistical approaches probabilistic approaches geometric approaches decision trees hidden Markov models artificial neural networks © Fraunhofer Overview Motivation: Enhance positioning in difficult urban mobility scenarios Theory: From raw data to signal features of motion states Application: Recognizing vehicle data in smartphone data Outlook: Recognizing all motion states of passengers © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Ground truth: Vehicle Signals 0-6 Hz Primary ride: wheel-surface contact, motion relative to the surface 6-30 Hz Secondary ride: motion of sub elements of the vehicle, thereof: • 0-15 Hz: vibration of longitudinal acceleration • 15-20 Hz: vibration of lateral acceleration 30-200 Hz Higher modes of vehicle sub elements In-Vehicle Frequency Ranges for Road vehicles, adapted from [Harrison 2004] © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Measurement Tram Dresden (TU Dresden) x 4 photo: Klaus Habermann © Fraunhofer x 3 x 2 x 1 Recognizing Vehicle Signals in Smartphone Data Using Smartphones as Vehicle Acceleration Sensors Motorola DEFY HTC Wildfire © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Time Patterns in Tram and Smartphone Data (I): Stops 12 10 x MT 8 y MT 6 z MT 4 2 0 -2 -4 0 5000 4000 3000 2000 1000 measurement index (200 Hz) time © Fraunhofer uncalibrated SP accelerations in m/s 2 SP orientation does not equal MT orientation uncalibrated MT accelerations in m/s2 12 10 8 6 4 2 x SP 0 y SP -2 z SP -4 0 5000 4000 3000 2000 1000 measurement index (interpolated, 200 Hz) time Recognizing Vehicle Signals in Smartphone Data Time Patterns in Tram and Smartphone Data (II): Bends Time: Smartphone 1 Time: Measurement tram Time: Smartphone 2 © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Time Patterns in Tram and Smartphone Data (III): Switches Time: Smartphone 1 Time: Measurement tram Time: Smartphone 2 © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Frequency Analysis in Tram and Smartphone Data (FFT) stops © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Frequency Analysis in Tram and Smartphone Data (FFT) stops © Fraunhofer Recognizing Vehicle Signals in Smartphone Data Frequency Analysis in Tram and Smartphone Data (STFT) drive stop drive stop Time (s) stop drive stop Time (s) drive drive stop Frequency (Hz) for x, y, z-axis Measurement tram data © Fraunhofer Frequency (Hz) for x-axis Smartphone 1, x-axis Overview Motivation: Enhance positioning in difficult urban mobility scenarios Theory: From raw data to signal features of motion states Application: Recognizing vehicle data in smartphone data Outlook: Recognizing all motion states of passengers © Fraunhofer Summary and Outlook Smartphone sensors may deliver valuable data on vehicle movements Sensor data may be analyzed and classified in various ways in time/frequency domain Next steps: Define signal features Reduce signal features to important ones Test classifiers Human movements on top of vehicle movements © Fraunhofer POSITIONING IN REAL-TIME PUBLIC TRANSPORT NAVIGATION DRESDEN Comparison Of Vehicle-based And Smart-phone Generated Acceleration Data To Determine Motion States Of Passengers Dipl.-Ing. Ina Partzsch (Fraunhofer IVI) Dipl.-Ing. Gunther Dürrschmidt (TU Dresden) Prof. Dr.-Ing. Oliver Michler (TU Dresden) Dr.-Ing. Georg Förster (Fraunhofer IVI) © Fraunhofer