M.J. Black. Explaining optical flow events with parameterized spatiotemporal models CVPR, Fort Collins, pp. 326-- 332, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....IIS 9876145, and ARO grant DAAD19 99 1 0139. We wish to thank Alessandro Chiuso. Recognition of complex motion patterns in images is an active area of research in computer vision. Extensive work has been conducted for the case of human motion and in particular facial expressions, for instance [2, 8, 3, 16, 13]. Some methods are based on optical flow. For each frame the flow can be approximated with a small dimensional vector in a suitable basis, as in [7] and the recognition is done with hidden Markov models (HMMs) or, as in [2] a spatiotemporal representation of the optical flow can be built. ....

....motion and in particular facial expressions, for instance [2, 8, 3, 16, 13] Some methods are based on optical flow. For each frame the flow can be approximated with a small dimensional vector in a suitable basis, as in [7] and the recognition is done with hidden Markov models (HMMs) or, as in [2], a spatiotemporal representation of the optical flow can be built. Others look at different spatio temporal features [12] In this paper we take a different approach: we do not choose local features, nor do we compute optical flow. Instead, we start from the assumption that the sequences of ....

M. G. Black. Explaining optical flow events with parameterized spatio-temporal models. In Proc. of Conference on Computer Vision and Pattern Recognition, volume 1, pages 326--32, 1999.

....into account restrictions for a particular non rigid motion and they are sensitive to occlusion and noise. Other trackers have emerged to solve these problems, such as Active Contours [1] all techniques relaying on template matching in image or Eigenspace [4, 12, 10] the ones using optical flow [2], deformable templates, feature trackers [9] and Active Shape Models [6] Active contours [1] are not well suited for tracking complex objects without contextual information or special signals. All optical flow techniques [2] have reported good results for recognition purposes, but are not valid ....

....matching in image or Eigenspace [4, 12, 10] the ones using optical flow [2] deformable templates, feature trackers [9] and Active Shape Models [6] Active contours [1] are not well suited for tracking complex objects without contextual information or special signals. All optical flow techniques [2] have reported good results for recognition purposes, but are not valid ones for accurate tracking, since the optical flow constraint is violated in situations where changes are due to appearance. All the techniques based on normalized correlation need a template which is not invariant to affine ....

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In CVPR, 1999.

....complex actions, context sensitive [19, 18] and multi agent models [12] have been proposed. Some methods are based on optical flow. For each frame the 1 flow can be approximated with a small dimensional vector in suitable basis, as in [11] and the recognition done with HMMs as before, or, as in [3], a spatio temporal representation of the optical flow can be built. Finally other methods, such as in [5] use a statistical hierarchical approach based on local appearance. 1.2 Contributions In this work we pose the problem of unsupervised recognition of actions and present a general model of ....

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In Proc. of Conference on Computer Vision and Pattern Recognition, volume 1, pages 326--332, 1999.

....2 complex actions, context sensitive [21, 20] and multi agent models [14] have been proposed. Some methods are based on optical flow. For each frame the flow can be approximated with a small dimensional vector in suitable basis, as in [13] and the recognition done with HMMs as before, or, as in [4], a spatio temporal representation of the optical flow can be built. In [12] a biological motion pattern is represented by a linear combination of prototypical image sequences. Other methods, such as in [6] use a statistical hierarchical approach based on local appearance. 1.2 Contributions In ....

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In Proc. of Conference on Computer Vision and Pattern Recognition, volume 1, pages 326--332, 1999.

....actions, context sensitive [19, 18] and multi agent models [12] have been proposed. Some methods are based on optical flow. For each frame the 1 flow can be approximated with a small dimensional vector in a suitable basis, as in [11] and the recognition done with HMMs as before, or, as in [3], a spatio temporal representation of the optical flow can be built. Finally other methods, such as [5] use a statistical hierarchical approach based on local appearance. 1.2 Contributions In this work we pose the problem of unsupervised recognition of actions and present a general model of an ....

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In Proc. of Conference on Computer Vision and Pattern Recognition, volume 1, pages 326--332, 1999.

....a non trivial Riemannian structure. 1.1 Contributions of this work and relation to previous work Recognition of complex motion patterns in images is an active area of research of computer vision. Extensive work has been conducted for the case of human motion and in particular facial expressions [12, 4, 13, 14]. A common approach consists in extracting low level features by local spatio temporal filters and then using hidden markov models to model the sequence of poses of the feature responses over time (see for instance [19, 26, 2, 6, 33, 34, 18, 30] Some methods are based on the optical flow. For ....

....poses of the feature responses over time (see for instance [19, 26, 2, 6, 33, 34, 18, 30] Some methods are based on the optical flow. For each frame the flow can be approximated with a small dimensional vector in suitable basis, as in [17] and the recognition done with HMMs as before, or, as in [4], a spatio temporal representation of the optical flow can be built. In [16] a biological motion pattern is represented by a linear combination of prototypical image sequences. Other methods, such as [7, 3, 5, 36] use a statistical hierarchical approach based on local appearance. Others look at ....

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In Proc. of Conference on Computer Vision and Pattern Recognition, volume 1, pages 326--332, 1999.

....ergodic HMM s to detect the entering and leaving of an office, kitchen, and communal areas with approximately 94 accuracy. Unlike these previous systems which identify discrete events, our system will concentrate on identifying continuous paths through an environment. In computer vision, Black [2] and Blake [3] have used the Condensation algorithm to perform activity recognition. In mobile robot navigation, Thrun et al. 7] also use the Condensation algorithm with the brightness of the ceiling as the observation model. A camera is mounted on top of a robot to look at the ceiling, and the ....

M. Black. Explaining optical flow events with parameterized spatiotemporal models. In CVPR99, 1999.

....in performing on line recognition of activities within the same framework. Activity recognition using PCA has been developed by Yacoob and Black[27] Isard and Blake[15] who proposed condensation for on line switching between different motion models and simultaneous recognition. Recently Black [3] has extended this approach allowing spatio temporal models of optical flow. Figure 1 describes an overview of the proposed system. The system is composed of three modules: Training, Tracking and Activity Recognition. The training module creates a global model of eyes (GME) for detection using ....

....the parameter search space and simultaneous recognition. It extends and unifies previous work [4] allowing flexible motion, local Eigenspaces with different spatial support, switching and temporal constraints within a stochastic framework. It also extends previous work on online recognition [3, 15], using appearance based representation and non linear dynamics for particle filtering. 2 Learning Local Models Since low dimensional parametrization for representing faces have been proposed by Sirovich and Kirby [22] several authors have used this idea to parametrize subspace representations ....

[Article contains additional citation context not shown here]

M. J. Black. Explaining optical flow events with parameterized spatio-temporal models. In CVPR, 1999.

....of 3D curve motion, and provides significant robustness against image noise and camera calibration errors. 1. Introduction There has been considerable interest in recovering the three dimensional shape and motion of an unknown dynamic scene from sequences of images e.g. work on optical flow [1], structure from motion [2] and 3D motion estimation [3] One common characteristic of these approaches is the assumption that all images are acquired by a single camera. Unfortunately, because the scene is viewed from just a single viewpoint at a time, this assumption imposes strong constraints ....

M. J. Black, "Explaining optical flow events with parameterized spatio-temporal models," in Proc. CVPR, pp. 326-332, 1999.

No context found.

M.J. Black. Explaining optical flow events with parameterized spatiotemporal models CVPR, Fort Collins, pp. 326-- 332, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC