Détection Radiale de Visages sur Images Omnidirectionnelles

Détection Radiale de Visages sur Images Omnidirectionnelles

Yohan Dupuis Xavier Savatier  Jean-Yves Ertaud  Pascal Vasseur 

Cerema DTerNC Chemin de la Poudrière F-76120 Le Grand Quevilly

ESIGELEC-IRSEEM Avenue Galilée F-76800 Saint Etienne du Rouvray

LITIS-Université de Rouen Avenue de l’Université F-76800 Saint Etienne du Rouvray

Page: 
143-173
|
DOI: 
https://doi.org/10.3166/TS.31.143-173
Received: 
4 October 2013
| |
Accepted: 
14 April 2014
| | Citation

OPEN ACCESS

Abstract: 

Omnidirectional vision sensors are mainly used for geometrical interpretation of scenes. However, few researchers have investigated how to perform object detection with such systems. The existing approaches require a geometrical transformation prior to the interpretation of the omnidirectional images. The face detection algorithm trained on perspective images is then applied on the unwrapped image. In this paper, we focus on how to process the omnidirectional images as provided by the sensor. While adapting algorithms developed for perspective images to omnidirectional images, our results suggest that the choice of descriptors is a critical step .

Extended Abstract

Face detection algorithms are now widely used in many real-time applications. This has been enabled by the breakthrough achieved by Viola and Jones in the early 2000’s. Their method is based on a tradeoff between high face detection performance and highly computationally optimized operations. However, their work has been developed for perspective images.

Perspectivecamerasareinterestingastheirprovideimagesclosetotherepresentation given by human eyes. However, perspective cameras have a limited field of view (FOV), which is a drawback when applications such as video surveillance are considered for instance. Omnidirectional vision systems provide an elegant alternative to perspective vision systems. Their large FOV is really attractive property, which explains that they are being widely used in mobile robotics. Omndirectional vision systems are used to tackle challenges involving robot navigation, movement estimation or 3D reconstruction. In these applications, omnidirectional images can be processed directly as they involve point matching. 

Most of the face detection algorithms, based on Viola and Jones framework, use region-based feature descriptors. As a results, the distortions induced by the omnidirectional image formation process do not allow the application of face detectors trained on perspective images. As a consequence, approaches found in the state-ofthe art use an intermediate image representation that aims at recreating the geometry found in perspective images. 

In this paper, we propose to investigate how to detect faces on omnidirectional images without requiring an intermediate representation. Our contributions can be divided into two parts. 

Firstofall,wehighlighttheconsequencesofomnidirectionalimageprocessingas compared to perspective image processing. We show that processing omnidirectional images directly has a lot of conceptual and practical benefits. Still, the distortions that exist on omnidirectional images introduce new challenges especially caused by an increased dispersion of the face class with respected to the feature space. 

Secondly, we propose to train a face detector from synthesized omnidirectional-like image face patches.The face patches are obtained from face detected on perspective images. The detector evaluation is performed on natural omnidirectional images. We investigated the influence of the feature descriptor used as well as the strong classifier. Our results suggest that a particular effort should be focused on the design of feature descriptors that take into account omnidirectional imaged istortions. Moreover, the performance achieved in our work indicates that objects detector for omnidirectional images may be trained from synthesized image patches taken from perspective images.

RÉSUMÉ

Les capteurs de vision omnidirectionnelle sont aujourd’hui couramment utilisés pour l’interprétation géométrique de scènes. Cependant, peu de travaux portent sur la détection d’objets à partir de ces capteurs. Les travaux existants passent par un dépliement des images omnidirectionnelles pour obtenir des images pseudo-perspectives. L’algorithmededétectionde visages sur images perspectives est ensuite appliqué directement sur les images dépliées. Dans ces travaux, nous investiguons la manière dont les images omnidirectionnelles doivent être traitées pour être interprétées telles que fournies par le capteur de vision omnidirectionnelle. Nos résultats montrent qu’une attention particulière doit être portée sur le choix des descripteurs lorsdel’adaptationd’algorithmes développés pour la vision perspective à la vision omnidirectionnelle.

Keywords: 

face detection, omnidirectional vision.

MOTS-CLÉS

boosting, détection de visages, vision omnidirectionnelle. 

1. Introduction
2. Théorie de la Vision Omnidirectionnelle
3. Problématiques Liées à la Détection de Visages sur Images Omnidirectionnelles
4. Méthodologie
5. Expérimentations
6. Conclusion
  References

Arican Z., Frossard P. (2010). OmniSIFT: Scale Invariant Features in Omnidirectional Images. In IEEE International Conference on Image Processing, p. 3505-3508. 

Baker S., Nayar S. (1999). A theory of single-viewpoint catadioptric image formation. International Journal of Computer Vision, vol. 35, no 2, p. 175–196. 

Barczak A. L. C., Johnson M. J., Messom C. H. (2005). Real-time Computation of Haar-like featuresatgenericanglesfordetectionalgorithms. Res.Lett.Inf.Math.Sci.,p.1175-2777. 

Barczak A. L. C., Okamoto J. J., Grassi V. J. (2009). Face Tracking Using a Hyperbolic Catadioptric Omnidirectional System. Res. Lett. Inf. Math. Sci., vol. 13, p. 55-67. 

Barreto J. (2003). General Central Projection Systems : Modeling, Calibration and Visual Servoing. Thèse de doctorat non publiée, University of Coimbra. BarretoJ.(2006).Aunifyinggeometricrepresentationforcentralprojectionsystems.Computer Vision and Image Understanding, vol. 103, no 3, p. 208–217. 

Bazin J., Demonceaux C., Vasseur P., Kweon I. (2009). Motion Estimation by Decoupling Rotation and Translation in Catadioptric Vision. Computer Vision and Image Understanding, vol. 114, p. 254-273. 

Bradski G. R., Kaehler A. (2008). Learning OpenCV: Computer Vision with the OpenCV Library. O’Reilly.

Bunschoten R. (2003). Mapping and Localization from a Panoramic Vision Sensor. Thèse de doctorat non publiée, University of Amsterdam.

Demonceaux C., Vasseur P., Fougerolle Y. D. (2011). Central catadioptric image processing with geodesic metric. Image Vision Computing, vol. 29, no 12, p. 840-849.

Douxchamps D., Campbell N. (2007). Robust Real Time Face Tracking for the Analysis of Human Behavior. In International Conference on Machine Learning for Multimodal Interaction, p. 1–10.

Du S., Zheng N., You Q., Wu Y., Yuan M., Wu J. (2006). Rotated haar-like features for face detection with in-plane rotation. Interactive Technologies and Sociotechnical Systems, p. 128–137. 

Dupuis Y., Savatier X., Ertaud J.-Y., Hoblos G. (2010). A Framework for Face Detection on Central Catadioptric Systems. In IEEE International Workshop on Robotic and Sensors Environments. 

Freund Y., Schapire. R. E. (1997). A decision-theoretic generalization of on-line learning and anapplicationtoboosting. JournalofComputerandSystemSciences,,vol.55,p.119-139. 

Geyer C. (2003). Catadioptric Projective Geometry: Theory and Applications. Thèse de doctorat non publiée, University of Pennsylvannia. 

Geyer C., Daniilidis K. (2001). Catadioptric projective geometry. IEEE International Journal of Computer Vision, vol. 45, no 3, p. 223–243. 

Jain V., Learned-Miller E. (2010). FDDB: A Benchmark for Face Detection in Unconstrained Settings. Rapport technique no UM-CS-2010-009. University of Massachusetts, Amherst. 

Lhuillier M. (2008). Toward automatic 3D modeling of scenes using a generic camera model. In IEEE International Conference on Computer Vision and Pattern Recogntion, p. 1–8. 

Lienhart R., Maydt J. (2002). An Extended Set of Haar-like Features for Rapid Object Detection. In IEEE International Conference on Image Processing, vol. 1, p. 900–903. 

Lin S., Gu J., Yamazaki S., Shum H. (2004). Radiometric calibration from a single image. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2, p. II–938. 

Lourenço M., Barreto J. P., Malti A. (2010). Feature detection and matching in images with radial distortion. In IEEE International Conference on Robotics and Automation, p. 10281034. 

Mei C., Rives P. (2007). Single View Point Camera Calibration from Planar Grids. In IEEE International Conference on Robotics and Automation, p. 3945–3950. 

Mei C., Sommerlade E., Sibley G., Newman P., Reid I. (2011, May). Hidden View Synthesis using Real-Time Visual SLAM for Simplifying Video Surveillance Analysis. In IEEE International Conference on Robotics and Automation, p. 4240–4245. 

Pham M., Gao Y., Hoang V., Cham T. (2010). Fast polygonal integration and its application in extendinghaar-likefeaturestoimproveobjectdetection. InIEEE Conference on Computer Vision and Pattern Recognition, p. 942–949. 

Rowley H. A., Baluja S., Kanade T. (1996). Neural Network-Based Face Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, p. 23–38. 

Scaramuzza D., Martinelli A., Siegwart R. (2006). A flexible technique for accurate omnidirectional camera calibration and structure from motion. In ICVS, p. 45–45. 

Schapire R. E., Singer Y. (1999). Improved Boosting Algorithms Using Confidence-rated Predictions. In Machine Learning, p. 80–91. 

Schneiderman H., Kanade T. (2000). A statistical method for 3D object detection applied to faces and cars. In IEEE International Conference on Computer Vision.

Strauss O., Comby F. (2007). Variable structuring element based fuzzy morphological operations for single viewpoint omnidirectional images. Pattern Recognition, vol. 40, p. 3578 3596. 

Sturm P., Ramalingam S., Tardif J., Gasparini S., Barreto J. (2011). Camera Models and FundamentalConceptsUsedinGeometricComputerVision. Foundations and Trends R

inComputer Graphics and Vision , vol. 6, no 1–2, p. 1–183. 

SvobodaT.(1999).Centralpanoramiccamerasdesign,geometry,egomotion.Thèsededoctorat non publiée, Center of Machine Perception, Czech Technical University in Prague. 

Viola M., Jones M. J. (2003). Fast Multi-view Face Detection. In IEEE International Conference on Computer Vision and Pattern Recognition. 

Viola M., Jones M. J. (2004). Robust Real-Time Face Detection. International Journal of Computer Vision, vol. 57, no 2, p. 137–154. 

Wickramaratna J., Holden S., Buxton B. (2001). Performance Degradation in Boosting. In J. Kittler, F. Roli (Eds.), Multiple Classifier Systems, vol. 2096, p. 11-21. Springer Berlin / Heidelberg.