## Learning generative models of invariant features (2004)

Venue: | in Proceedings of the IEEE/RSJ Conference on Intelligent Robots and Systems (IROS |

Citations: | 5 - 0 self |

### BibTeX

@INPROCEEDINGS{Sim04learninggenerative,

author = {Robert Sim},

title = {Learning generative models of invariant features},

booktitle = {in Proceedings of the IEEE/RSJ Conference on Intelligent Robots and Systems (IROS},

year = {2004},

pages = {3481--3488}

}

### OpenURL

### Abstract

Abstract — We present a method for learning a set of models of visual features which are invariant to scale and translation in the image domain. The models are constructed by first applying the Scale-Invariant Feature Transform (SIFT) to a set of training images, and matching the extracted features across the images, followed by learning the pose-dependent behavior of the features. The modeling process avoids assumptions with respect to scene and imaging geometry, but rather learns the direct mapping from camera pose to feature observation. Such models are useful for applications to robotic tasks, such as localization, as well as visualization tasks. We present the model learning framework, and experimental results illustrating the success of the method for learning models that are useful for robot localization. I.

### Citations

3719 |
Stochastic Relaxation, Gibbs Distributions and the Bayesian Restoration of Images
- Geman, Geman
- 1984
(Show Context)
Citation Context ...ture model, and as such, the lower edge of the image is left unshaded. It may be that these regions can be shaded by extrapolating from the nearby texture using Markovian reconstruction methods [20], =-=[21]-=-. B. Localization The goal of the feature learning framework is to achieve accurate localization performance with a mobile robot. Given a set of feature models, the task of robot localization can be p... |

3158 |
A.: “Multiple View Geometry in Computer Vision
- Hartley, Zisserman
- 2004
(Show Context)
Citation Context ...metrically using a stereo camera [8], this work does not compute explicit geometric models of the extracted features. Similarly, this approach differentiates our work from structure from motion (SFM) =-=[9]-=-, which imposes assumptions on both the features and the imaging geometry. The remainder of this paper will address related work, and introduce the feature modeling approach, including implementation ... |

1585 | Object recognition from local scale-invariant features
- Lowe
- 1999
(Show Context)
Citation Context ...t they are difficult to track reliably and subject to outlier matches, resulting in degradations in the inferred models.sIn this paper, the Scale-Invariant Feature Transform (SIFT), developed by Lowe =-=[7]-=- is employed (Figure 1 a). SIFT features provide enhanced stability against variations due to pose and illumination, as well as viewpoint-invariant descriptors for matching in the presence of changes ... |

766 | Texture synthesis by non-parametric sampling
- Efros, Leung
- 1999
(Show Context)
Citation Context ...le feature model, and as such, the lower edge of the image is left unshaded. It may be that these regions can be shaded by extrapolating from the nearby texture using Markovian reconstruction methods =-=[20]-=-, [21]. B. Localization The goal of the feature learning framework is to achieve accurate localization performance with a mobile robot. Given a set of feature models, the task of robot localization ca... |

403 | A probabilistic approach to concurrent mapping and localization for mobile robots
- Thrun, Fox, et al.
- 1998
(Show Context)
Citation Context ...exploratory stage (such as assistance from a second observing robot). This assumption can be relaxed with more sophisticated mapbuilding approaches, such as the utilization of expectationmaximization =-=[15]-=-. IV. IMPLEMENTATION The feature learning framework is divided into four stages, extraction, tracking, modeling and evaluation. The following sections will describe the details of each stage. A. Featu... |

366 |
Neural networks
- Haykin
- 1994
(Show Context)
Citation Context .... The width, or influence, of the RBF is defined by σ. The computation of the weight vectors wi is well understood in the context of regularization and interpolation theory and is described elsewhere =-=[17]-=-. In brief, the optimal weights W = [wij] are the solution to the linear least squares problem i (G + λI)W = Z (7) where the elements Gij of the design matrix G correspond to Equation 6 evaluated at o... |

335 |
Condensation-Conditional Density Propagation for Visual Tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context .... It should be noted that conventional feature tracking methods, such as conditional density propagation are dependent on time-series inputs and are not well suited to the requirements of our problem =-=[16]-=-. SIFT features are matched by comparing their invariant descriptors. The invariant descriptor it for feature fi in the database is defined by the descriptor for the observation that is closest in pos... |

297 | Real-time simultaneous localisation and mapping with a single camera
- Davison
- 2003
(Show Context)
Citation Context ...hes to active localization using the Kalman Filter and Markov chains [11], [12]. These principles have also been applied using 3D point features using both stereo, and pinhole camera models [8], [9], =-=[13]-=-. Finally, a variety of linear analysis techniques have resulted in features that are extracted implicitly [1], [2], and localization techniques using linear combinations of views [14]. Our work is si... |

273 |
Durrant-Whyte, “Mobile robot localization by tracking geometric beacons,” Robotics and Automation
- Leonard, F
- 1991
(Show Context)
Citation Context ...thods for localizing a robot in the plane using 2D point landmarks [10], and culminated in the development of probabilistic approaches to active localization using the Kalman Filter and Markov chains =-=[11]-=-, [12]. These principles have also been applied using 3D point features using both stereo, and pinhole camera models [8], [9], [13]. Finally, a variety of linear analysis techniques have resulted in f... |

204 | Mobile robot localization and mapping with uncertainty using scale-invariant visual landmarks
- Se, Lowe, et al.
(Show Context)
Citation Context ...real-world processes and hence captured implicitly by the Gaussian noise model. Note that, unlike the approach taken by Se, et al., where SIFT features are modeled geometrically using a stereo camera =-=[8]-=-, this work does not compute explicit geometric models of the extracted features. Similarly, this approach differentiates our work from structure from motion (SFM) [9], which imposes assumptions on bo... |

86 | Multi-robot collaboration for robust exploration
- Rekleitis, Dudek, et al.
- 2001
(Show Context)
Citation Context ... a laser tracking system was deployed to estimate the ground-truth position of the exploring robot to an accuracy of approximately 4cm. The implementation of the laser tracking system is described in =-=[19]-=-. The observer employed a laser range-finder to accurately determine the camera position from the range and orientation of a threeplane target mounted on the exploring robot. For the purposes of this ... |

80 | Active markov localization for mobile robots
- Fox, Burgard
- 1998
(Show Context)
Citation Context ...for localizing a robot in the plane using 2D point landmarks [10], and culminated in the development of probabilistic approaches to active localization using the Kalman Filter and Markov chains [11], =-=[12]-=-. These principles have also been applied using 3D point features using both stereo, and pinhole camera models [8], [9], [13]. Finally, a variety of linear analysis techniques have resulted in feature... |

61 | Global localization using distinctive visual features
- Se, Lowe, et al.
- 2002
(Show Context)
Citation Context ... probability that they are visible. When the prior is uniform, that is, no information is available, one can employ the efficient indexing methods described by Se et al. to efficiently match features =-=[22]-=-. Such a mechanism is beyond the scope of this paper. Once matched, each feature observation zi contributes a probability density function p(zi|x), which is defined as the product of the distribution ... |

50 |
Some location problems for robot navigation using a single camera
- Sugihara
- 1988
(Show Context)
Citation Context ...LATED WORK The problem of feature-based localization has been the subject of extensive research. Early work examined triangulation methods for localizing a robot in the plane using 2D point landmarks =-=[10]-=-, and culminated in the development of probabilistic approaches to active localization using the Kalman Filter and Markov chains [11], [12]. These principles have also been applied using 3D point feat... |

48 | View-based object recognition using saliency maps
- Shokoufandeh, Marsic, et al.
- 1999
(Show Context)
Citation Context ...o globally derived features, such as principal components analysis [1], [2] can provide robustness to a variety of factors, such as illumination variation, dynamic environments, and sensor noise [3], =-=[4]-=-. Other work has provided psychophysical evidence for featurebased recognition, as well as support for improvements in computational efficiency [5]. Our work is motivated by Gregory Dudek Department o... |

31 | A structured probabilistic model for recognition
- Schmid
- 1999
(Show Context)
Citation Context ...sed to globally derived features, such as principal components analysis [1], [2] can provide robustness to a variety of factors, such as illumination variation, dynamic environments, and sensor noise =-=[3]-=-, [4]. Other work has provided psychophysical evidence for featurebased recognition, as well as support for improvements in computational efficiency [5]. Our work is motivated by Gregory Dudek Departm... |

29 |
Tracking Visual Appearance
- Nayar, Murase, et al.
- 1994
(Show Context)
Citation Context ... such as object recognition and robot localization, has demonstrated that inferences based on local image features, as opposed to globally derived features, such as principal components analysis [1], =-=[2]-=- can provide robustness to a variety of factors, such as illumination variation, dynamic environments, and sensor noise [3], [4]. Other work has provided psychophysical evidence for featurebased recog... |

26 |
Cross-validation of regression models
- Picard, Cook
- 1984
(Show Context)
Citation Context ...g examples. D. Model Evaluation Once features have been tracked and modeled, it is desirable to evaluate the quality of the models. Each feature model is evaluated using leave-one-out crossvalidation =-=[18]-=-, which operates by constructing the model with one data point z excluded, predicting that data point z ∗ using the construction and measuring the difference ze = ||z−z ∗ || between the actual point a... |

24 | Continuity properties of the appearance manifold for mobile robot position estimation
- Pourraz, Crowley
- 1998
(Show Context)
Citation Context ...lems, such as object recognition and robot localization, has demonstrated that inferences based on local image features, as opposed to globally derived features, such as principal components analysis =-=[1]-=-, [2] can provide robustness to a variety of factors, such as illumination variation, dynamic environments, and sensor noise [3], [4]. Other work has provided psychophysical evidence for featurebased ... |

14 | Learning generative models of scene features
- Sim, Dudek
- 2001
(Show Context)
Citation Context ... from a novel viewpoint. these ideas and specifically aims for robust behavior in dynamic environments. The main contribution of this paper is the application of the Visual Map framework developed in =-=[6]-=- to the problem of learning models of scale and rotationally invariant features. The modeled features demonstrate improved robustness for unambiguous recognition and tracking, and greater versatility ... |

2 |
Tsotsos, The complexity of perceptual search tasks
- K
- 1989
(Show Context)
Citation Context ...riation, dynamic environments, and sensor noise [3], [4]. Other work has provided psychophysical evidence for featurebased recognition, as well as support for improvements in computational efficiency =-=[5]-=-. Our work is motivated by Gregory Dudek Department of Computer Science McGill University dudek@cim.mcgill.ca Fig. 1. a) Detected SIFT features in an image. Each arrow corresponds to a detected featur... |

1 | Localization using combinations of model views
- Basri, Rivlin
- 1993
(Show Context)
Citation Context ...odels [8], [9], [13]. Finally, a variety of linear analysis techniques have resulted in features that are extracted implicitly [1], [2], and localization techniques using linear combinations of views =-=[14]-=-. Our work is similar to the earlier localization techniques in that it applies probabilistic methods to localization from feature observations. However, it is more similar to the latter techniques in... |

1 |
The ‘summation hack’ as an outlier model,” http://www.stat.cmu.edu/˜minka/papers/minka-summation.pdf
- Minka
- 2003
(Show Context)
Citation Context ...le image and, furthermore, outliers can lead to catastrophic cancellation of the joint distribution. Instead, we employ a mixture model defined by n� p(z|x) ≈ p(zi|x) (13) i=1sThe reader may refer to =-=[23]-=- for a discussion of why this approach is reasonable. Fig. 7. A posteriori pose distributions for a selection of the Scene I validation images. Each image represents an overhead view of the 2m by 2m p... |