Mobility Models

Pedestrian mobility models are used to quantitatively represent the stochastic nature of pedestrian movement. Research in such movement models has mainly been applied to planning tasks, such as estimating pedestrian flows when planning a train station or airport, or optimization of evacuation procedures and evacuation paths for a large shopping mall or theater. Another application area which is becoming increasingly important is that of dynamic indoor positioning and navigation. The reason that a movement model is needed in these cases lies in the dynamic nature of most pedestrian indoor navigation applications: The user’s position will be estimated continuously so as to allow services such as personalized travel assistance or indoor navigation directions. In addition, it can be shown that a dynamic positioning system is more accurate than a “single-shot” static estimator which essentially provides a position estimate based on positioning measurements of a single time instance. To implement mathematically sound dynamic estimators one needs an accurate and realistic movement model (also known as the a-priori state transition model) of the dynamic system: Here the user’s stochastic movement (position, velocity, attitude, etc). Additionally, a goal oriented movement model based on the diffusion algorithm is combined with the stochastic movement model to take goal oriented movement into account. All models are extended to handle the 3D environment.

In order to handle critical scenarios like large rooms new angular PDFs based on maps are applied in our cascaded Bayesian estimation architecture. The location dependent angular PDFs can serve also for predicting the heading within an motion model. More can be found here ...

Angular PDFs can also be used as a prior map for FootSLAM. More can be found here ...

This page will serve as a central point of access to our work on mobility models.

Demos:

Diffusion Demo - This demo shows the generation of "reasonable" paths in our map-based mobility model

Figures:

Diffusion Model Calcualtions in 3 floors building and a Simulated user walking from Level 3 to Level 1

Handling the 3D environment within the diffusion algorithm


Layout Matrix for areas with different accessibility levels. The walls are given in black, not easy reachable forest area is marked with dark gray, and flowerbed area is given in light gray. The area where people may walk is given in white.	Diffusion results after reaching steady state, where the gas concentration is high in the dark red area and low in the blue area

Videos:


A video showing many simulated walking pedestrians (represented by the black dots) using a simple stochastic motion model. After running the simulation for sometime, we notice towards the end of the video that such models is not suited for situations in which walls or roads have a strong influence on the movement. This model leads to a high probability of getting stuck in a room or having problems in getting through narrow openings and sharp turns. This is because the random movement which the model is following does not react to the presence of a door, a narrow opening or a sharp turn. Additionally, the model does not include the behavior of a pedestrian heading a specific destination. Download video [MPG].		A video showing many simulated walking pedestrians (represented by the black dots) using a Novel Movement Model Suited for Pedestrian Navigation. The model represents an intelligent combination between a Stochastic Behavioral Movement Model (suitable of a non-goal oriented movement) and a Diffusion Movement Model (suitable for a target driven movement). After running the simulation for sometime, we notice towards the end of the video that pedestrians are not stuck in rooms anymore and having no problems getting through narrow openings and sharp turns like in the Stochastic Behavioral Movement Model. Additionally, the model includes the behavior of a pedestrian heading a specific destination. Download video [MPG].

Videos demonstrating angular PDFs:

For comparison please click on both videos at the same time.


Critical Simulation Scenario: A large Room within a building causes problems when particles (upper Particle Filter) enter the large room. When using walls to eliminate particles in a binary way, the particles within the more restricted area may suffer from the wall constraints whereas particles within the large room will not suffer. Therefore, the wrong cloud survives when using no angular PDF.	Same Critical Simulation Scenario: When using angular PDFs for weighting, particles within the more restricted area are rewarded. Therefore, the correct cloud survives when using angular PDFs for weighting in the Particle Filter.

Polar Plots of the location dependent angular PDFs for one loop of the track.

Angular PDFs Slides

IPIN2011 Angular PDF from susannakaiser

More Details can be found in Ref. "A Human Motion Model ...", Ref. "A Maps-Based Angular PDF ..." , and Ref. "A pedestrian navigation system ...".

Angular PDFs used as Prior Map for FootSLAM (Plans2012):

For comparison please click on both videos at the same time.

FootSLAM without the use of a prior map: Assuming that the starting position is not exactly known (a noise of 0.2m for the x,y component of the starting position and 0.2° for the starting angle is assumed) the position of the FootSLAM map varies that leads to positioning inaccuracies. The FootSLAM map converges very late. This film was shown at PLANS2012 conference. See also the Slides and the PLANS2012 Paper ...

FootSLAM with angular PDFs as prior map: When using prior maps the positioning accuracy is enhanced assuming a not exactly known starting position (a noise of 0.2m for the x,y component of the starting position and 0.2° for the starting angle is assumed). In addition, the FootSLAM algorithm converges faster with the prior map. This film was shown at PLANS2012 conference.

Polar Plots of the location dependent angular PDFs for one loop of the track used in the simulations above.

Angular PDFs and FootSLAM Slides

Plans2012 Presentation: Angular PDFs and FootSLAM from susannakaiser

More Details can be found in Ref. "Maps-based angular PDFs ...".

Publications:

More details can be found in the following publications:

Susanna Kaiser, Maria Garcia Puyol, Patrick Robertson,
Maps-based angular PDFs used as prior maps for FootSLAM, [ELIB]
Proceedings of ION PLANS 2012,
Myrtle Beach, South-Carolina, April 2012

Susanna Kaiser, Mohammed Khider, Patrick Robertson,
A pedestrian navigation system using a map-based angular motion model for indoor and outdoor environments, [ELIB]
Journal of Location Based Services, Taylor&Francis, DOI:10.1080/17489725.2012.698110,
June 2012

Susanna Kaiser, Mohammed Khider, Patrick Robertson,
A Maps-Based Angular PDF for Navigation Systems in Indoor and Outdoor Environments, [ELIB]
Proceedings of IPIN 2011,
Guimaraes, Portugal, September 2011

Susanna Kaiser, Mohammed Khider, Patrick Robertson,
A Human Motion Model Based on Maps for Navigation Systems, [ELIB][]
EURASIP Journal on Wireless Communications and Networking,
Springer, August 2011

Mohammed Khider, Susanna Kaiser, Patrick Robertson, Michael Angermann,
Maps And Floor Plans Enhanced 3D Movement Model For Pedestrian Navigation, [ELIB][]
Proceedings of the ION GNSS 2009,
Georgia, USA, September 2009

Mohammed Khider, Susanna Kaiser, Patrick Robertson, Michael Angermann,
A Three Dimensional Movement Model For Pedestrian Navigation, [ELIB][]
Proceedings of the European Navigation Conference - Global Navigation Satellite Systems (ENC-GNSS) 2009,
Napoli, Italy, May 2009

Mohammed Khider, Susanna Kaiser, Patrick Robertson, Michael Angermann,
A Novel Movement Model for Pedestrians Suitable for Personal Navigation, [ELIB][]
Proceedings of the ION NTM 2008,
San Diego, USA, Jan 2008

Mohammed Khider, Susanna Kaiser, Patrick Robertson, Michael Angermann,
The Effect of Maps-Enhanced Novel Movement Models on Pedestrian Navigation Performance, [ELIB][]
European Navigation Conference (ENC-GNSS 2008),
Toulouse, France, April 2008

Kai Wendlandt, Mohammed Khider, Michael Angermann, Patrick Robertson,
Continuous location and direction estimation with multiple sensors using particle filtering, [ELIB][]
Proceedings of the International Conference in Multisensor Fusion and Integration for Intelligent Systems, Library of Congress: 2006930785, IEEE Verlag, MFI 2006, Heidelberg, Germany, 2006-09-04

Mohammed Khider
Implementation of a Simulator/Demonstrator for the SoftLocation Concept using Bayesian Filters, [ELIB][]
Masterarbeit, S. 166, Universität Ulm (Ingenieurwissenschaften und Informatik) ,
Ulm, Germany, 2005

Michael Angermann, Jens Kammann, Bruno Lami,
A New Mobility Model Based on Maps, [ELIB][]
Proceedings of the 58th IEEE Semiannual Vehicular Technology Conference (VTC Fall 2003),
Orlanda, USA, October 6-9, 2003

In case of any questions please contact:
Michael Angermann
Mohammed Khider
Susanna Kaiser