## Gaussian Processes for Signal Strength-Based Location Estimation (2006)

### Cached

### Download Links

Venue: | In Proc. of Robotics Science and Systems |

Citations: | 62 - 8 self |

### BibTeX

@INPROCEEDINGS{Ferris06gaussianprocesses,

author = {Brian Ferris and Dirk Hähnel and Dieter Fox},

title = {Gaussian Processes for Signal Strength-Based Location Estimation},

booktitle = {In Proc. of Robotics Science and Systems},

year = {2006}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract — Estimating the location of a mobile device or a robot from wireless signal strength has become an area of highly active research. The key problem in this context stems from the complexity of how signals propagate through space, especially in the presence of obstacles such as buildings, walls or people. In this paper we show how Gaussian processes can be used to generate a likelihood model for signal strength measurements. We also show how parameters of the model, such as signal noise and spatial correlation between measurements, can be learned from data via hyperparameter estimation. Experiments using WiFi indoor data and GSM cellphone connectivity demonstrate the superior performance of our approach. I.

### Citations

1315 | Radar: An inbuilding rf-based user location and tracking system
- Bahl, Padmanabhan
- 2000
(Show Context)
Citation Context ...about the locations of access points and then model the propagation of signals through space to determine the expected signal strength at any location based on the distance from an access point [15], =-=[1]-=-, [8]. Unfortunately, these parametric models have only limited accuracy, even when taking information about the locations of walls and furniture into account. The second class of techniques compute m... |

788 |
Probabilistic Robotics
- Thrun, Burgard, et al.
- 2005
(Show Context)
Citation Context ...son from signal strength measurements. A key component of a Bayes filter is the observation model, which describes the likelihood of making an observation at the different locations in an environment =-=[16]-=-. Before we discuss the specifics of our approach to localization, we will show how Gaussian processes can be used to generate an observation model for signal strength measurements from calibration da... |

289 | Gaussian processes for machine learning
- Rasmussen
- 2006
(Show Context)
Citation Context ...ng [10], [5], [8], existing techniques have important limitations with respect to considering all available information in a statistically sound way. In this paper we show how Gaussian processes (GP) =-=[13]-=- can be used to overcome these limitations. GPs are nonparametric models that estimate Gaussian distributions over functions based on training data. GP regression has been used with great success in a... |

266 |
Spoken Language Processing: A Guide to Theory
- Huang, Aero, et al.
(Show Context)
Citation Context ...o account by “smoothing” the likelihood model: � n� �γ p(zt[1:n]|xt) = p(zt[i] | xt) (14) i=1 Here, n is the number of detected access points and γ ∈ [0 : 1] plays the role of a smoothing coefficient =-=[6]-=-. In our experiments we set γ to 1/n, resulting in the geometric mean of the individual likelihoods. A. Mixed Graph / Free Space Representation Our representation of a person’s locations is motivated ... |

185 | Robotics-based Location Sensing Using Wireless Ethernet
- Ladd, Bekris, et al.
- 2002
(Show Context)
Citation Context ...nd furniture into account. The second class of techniques compute measurement likelihoods using location-specific statistics extracted from calibration data. These local statistics include histograms =-=[7]-=-, Gaussians [5], [4], [10], or even raw measurements [1]. Local approaches typically result in higher localization accuracy if enough calibration data is available. A main problem for these models, ho... |

134 | The Gaussian Process Latent Variable Model
- Lawrence
(Show Context)
Citation Context ...t estimate Gaussian distributions over functions based on training data. GP regression has been used with great success in a variety of applications, including sensor networks [3], data visualization =-=[9]-=-, and computer animation [12]. GPs have several properties that make them ideally suited for modeling signal strength measurements: Continuous locations: GPs do not require a discretized representatio... |

115 |
914 mhz path loss prediction models for indoor wireless communications in multifloored buildings. Antennas and Propagation
- Seidel, Rappaport
- 1992
(Show Context)
Citation Context ...ledge about the locations of access points and then model the propagation of signals through space to determine the expected signal strength at any location based on the distance from an access point =-=[15]-=-, [1], [8]. Unfortunately, these parametric models have only limited accuracy, even when taking information about the locations of walls and furniture into account. The second class of techniques comp... |

113 | Practical robust localization over large-scale 802.11 wireless networks
- Haeberlen, Flannery, et al.
- 2004
(Show Context)
Citation Context ...count. The second class of techniques compute measurement likelihoods using location-specific statistics extracted from calibration data. These local statistics include histograms [7], Gaussians [5], =-=[4]-=-, [10], or even raw measurements [1]. Local approaches typically result in higher localization accuracy if enough calibration data is available. A main problem for these models, however, is to generat... |

101 | Learning physics-based motion style with nonlinear inverse optimization
- Liu, Hertzmann, et al.
- 2005
(Show Context)
Citation Context ...tions over functions based on training data. GP regression has been used with great success in a variety of applications, including sensor networks [3], data visualization [9], and computer animation =-=[12]-=-. GPs have several properties that make them ideally suited for modeling signal strength measurements: Continuous locations: GPs do not require a discretized representation of an environment, or the c... |

46 | Large-scale localization from wireless signal strength
- Letchner, Fox, et al.
- 2005
(Show Context)
Citation Context .... The second class of techniques compute measurement likelihoods using location-specific statistics extracted from calibration data. These local statistics include histograms [7], Gaussians [5], [4], =-=[10]-=-, or even raw measurements [1]. Local approaches typically result in higher localization accuracy if enough calibration data is available. A main problem for these models, however, is to generate like... |

40 | Voronoi tracking: Location estimation using sparse and noisy sensor data
- Liao, Hightower, et al.
- 2003
(Show Context)
Citation Context ...c mean of the individual likelihoods. A. Mixed Graph / Free Space Representation Our representation of a person’s locations is motivated by the Voronoi motion graphs introduced by Liao and colleagues =-=[11]-=-. The key idea of their approach is to represent indoor environments by graphs whose edges correspond to the Voronoi graph of an environment. Liao et al. showed that by constraining a person’s locatio... |

29 | GPPS: A Gaussian Process Positioning System for Cellular Networks
- Schwaighofer, Grigoras, et al.
- 2004
(Show Context)
Citation Context ...arameters include the spatial correlation between measurements and the measurement noise. The use of GPs for signal strength based location estimation has been proposed by Schwaighofer and colleagues =-=[14]-=-. In this paper we extend their work in several directions. More specifically, we introduce a Bayesian filter for location estimation that builds on a mixed graph / free space representation of indoor... |

25 | An experimental study of localization using wireless ethernet
- Howard, Siddiqi, et al.
- 2003
(Show Context)
Citation Context ...to account. The second class of techniques compute measurement likelihoods using location-specific statistics extracted from calibration data. These local statistics include histograms [7], Gaussians =-=[5]-=-, [4], [10], or even raw measurements [1]. Local approaches typically result in higher localization accuracy if enough calibration data is available. A main problem for these models, however, is to ge... |

22 | Selfmapping in 802.11 location systems
- LaMarca, Hightower, et al.
- 2005
(Show Context)
Citation Context ... the locations of access points and then model the propagation of signals through space to determine the expected signal strength at any location based on the distance from an access point [15], [1], =-=[8]-=-. Unfortunately, these parametric models have only limited accuracy, even when taking information about the locations of walls and furniture into account. The second class of techniques compute measur... |

1 |
Near-optimal sensor placements using Gaussian processes
- Guestrin, Krause, et al.
- 2005
(Show Context)
Citation Context ...nonparametric models that estimate Gaussian distributions over functions based on training data. GP regression has been used with great success in a variety of applications, including sensor networks =-=[3]-=-, data visualization [9], and computer animation [12]. GPs have several properties that make them ideally suited for modeling signal strength measurements: Continuous locations: GPs do not require a d... |