References of "Declercq, Arnaud"
     in
Bookmark and Share    
Full Text
See detailReal-time Simultaneous Modelling and Tracking of Articulated Objects
Declercq, Arnaud ULg

Doctoral thesis (2012)

In terms of capability, there is still a huge gap between the human visual system and existing computer vision algorithms. To achieve results of su cient quality, these algorithms are generally extremely ... [more ▼]

In terms of capability, there is still a huge gap between the human visual system and existing computer vision algorithms. To achieve results of su cient quality, these algorithms are generally extremely specialised in the task they have been designed for. All the knowledge available during their implementation is used to bias the output result and/or facilitate the initialisation of the system. This leads to increased robustness but a lower reusability of the code. In most cases, it also majorly limits the freedom of the user by constraining him to a limited set of possible interactions. In this thesis, we propose to go in the opposite direction by developing a general framework capable of both tracking and learning objects as complex as articulated objects. The robustness will be achieved by using one task to assist the other. The method should be completely unsupervised with no prior knowledge about the appearance or shape of the objects encountered (although, we decided to focus on rigid and articulated objects). With this framework, we hope to provide directions for a more di cult and distant goal: that of completely eliminating the time consuming prior design of object models in computer vision applications. This long term target will allow the reduction of the time and cost of implementing computer vision applications. It will also provide a larger freedom in the range of objects that can be used by the program. Our research focuses on three main aspects of this framework. The rst one is to create an object description e ective on a wide variety of complex objects and able to assist the object tracking while being learnt. The second is to provide both tracking and learning methods that can be executed simultaneously in real-time. This is particularly challenging for tracking when a large number of features are involved. Finally, our most challenging task and the core of this thesis, is to design robust tracking and learning solutions able to assist each other without creating counter-productive bias when one of them fails. [less ▲]

Detailed reference viewed: 21 (6 ULg)
Full Text
Peer Reviewed
See detailAffine Warp Propagation for Fast Simultaneous Modelling and Tracking of Articulated Objects
Declercq, Arnaud ULg

Poster (2010, November)

We propose a new framework that allows simultaneous modelling and tracking of articulated objects in real time. We introduce a non-probabilistic graphical model and a new type of message that propagates ... [more ▼]

We propose a new framework that allows simultaneous modelling and tracking of articulated objects in real time. We introduce a non-probabilistic graphical model and a new type of message that propagates explicit motion information for realignment of feature constellations across frames. These messages are weighted according to the rigidity of the relations between the source and destination features. We also present a method for learning these weights as well as the spatial relations between connected feature points, automatically identifying deformable and rigid object parts. Our method is extremely fast and allows simultaneous learning and tracking of nonrigid models containing hundreds of feature points with negligible computational overhead. [less ▲]

Detailed reference viewed: 8 (1 ULg)
Full Text
Peer Reviewed
See detailAffine Warp Propagation for Fast Simultaneous Modelling and Tracking of Articulated Objects
Declercq, Arnaud ULg; Piater, Justus ULg

in ACCV10 (2010)

We propose a new framework that allows simultaneous modelling and tracking of articulated objects in real time. We introduce a non-probabilistic graphical model and a new type of message that propagates ... [more ▼]

We propose a new framework that allows simultaneous modelling and tracking of articulated objects in real time. We introduce a non-probabilistic graphical model and a new type of message that propagates explicit motion information for realignment of feature constellations across frames. These messages are weighted according to the rigidity of the relations between the source and destination features. We also present a method for learning these weights as well as the spatial relations between connected feature points, automatically identifying deformable and rigid object parts. Our method is extremely fast and allows simultaneous learning and tracking of nonrigid models containing hundreds of feature points with negligible computational overhead. [less ▲]

Detailed reference viewed: 12 (3 ULg)
Peer Reviewed
See detailOnline Learning of Gaussian Mixture Models - a Two-Level Approach
Declercq, Arnaud ULg; Piater, Justus ULg

in VISAPP 2008: Proceedings of the Third International Conference on Computer Vision Theory and Applications - Volume 1 (2008)

Online learning, Gaussian mixture model, Uncertain model. We present a method for incrementally learning mixture models that avoids the necessity to keep all data points around. It contains a single user ... [more ▼]

Online learning, Gaussian mixture model, Uncertain model. We present a method for incrementally learning mixture models that avoids the necessity to keep all data points around. It contains a single user-settable parameter that controls via a novel statistical criterion the trade-off between the number of mixture components and the accuracy of representing the data. A key idea is that each component of the (non-overfitting) mixture is in turn represented by an underlying mixture that represents the data very precisely (without regards to overfitting); this allows the model to be refined without sacrificing accuracy. [less ▲]

Detailed reference viewed: 76 (0 ULg)
Peer Reviewed
See detailOn-line Simultaneous Learning and Tracking of Visual Feature Graphs
Declercq, Arnaud ULg; Piater, Justus ULg

in 2007 IEEE Conference on Computer Vision and Pattern Recognition (CVPR07) (2007)

Model learning and tracking are two important topics in computer vision. While there are many applications where one of them is used to support the other, there are currently only few where both aid each ... [more ▼]

Model learning and tracking are two important topics in computer vision. While there are many applications where one of them is used to support the other, there are currently only few where both aid each other simultaneously. In this work, we seek to incrementally learn a graphical model from tracking and to simultaneously use whatever has been learned to improve the tracking in the next frames. The main problem encountered in this situation is that the current intermediate model may be inconsistent with future observations, creating a bias in the tracking results. We propose an uncertain model that explicitly accounts for such uncertainties by representing relations by an appropriately weighted sum of informative (parametric) and uninformative (uniform) components. The method is completely unsupervised and operates in real time. [less ▲]

Detailed reference viewed: 18 (0 ULg)