A. Maki, P. Nordlund, and J.-O. Eklundh, "Attentional Scene Segmentation: Integrating Depth and Motion from Phase," Computer Vision and Image Understanding, vol. 78, pp. 351-373, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of a scene. Cognitive scientists believe that fixations are selected both bottom up, in response to the contents of the current scene, and top down, reflecting current goals [18, 24] There are a number of computational systems that attempt to model aspects of selective attention, including [13, 14, 24, 23, 19, 8, 22, 16]. Most of these systems were intended as cognitive models of human attention, and have therefore been evaluated by cognitive science measures, such as the pop out effect [6] their speed in relation to human performance [6] and fidelity to human eye tracking experiments [20] More and more, ....

....until the desired number of fixations have been selected. For more information, see [11, 10, 8, 6, 20] It is important to note that there are other computational models of selective attention. Maki et al. present a system that exploits motion and stereo depth perception for selection attention [13]. Tsotsos et al. present a more general neural network model that uses selective tuning [23] Park et al. recently introduced a system similar to NVT, except that they use independent component analysis to impliment a feature competition scheme [19] The mentioned here are just a few in a quickly ....

A. Maki, P. Nordlund, and J.-O. Eklundh. Attentional scene segmentation: Integrating depth and motion from phase. Computer Vision and Image Understanding, 2000.

....part of the scene, the space is determined by a fast object based segmentation of the scene, or by grouping effects. Examples of the successful incorporation of object based approaches into computer models of visual attention have been demonstrated at various levels by Fellenz [20] Maki et al. [21], and Dickinson et al. 22] We aim to contribute to these achievements with a special focus on dynamic aspects of controlling attention. C. Limitations of conventional models By applying conventional models (see section II A) to dynamic scenes we identify three major problems: Inhibition of ....

A. Maki, P. Nordlund, and J.-O. Eklundh, "Attentional scene segmentation: Integrating depth and motion," Computer Vision and Image Understanding, vol. 78, pp. 351--373, 2000.

....many others, includes no depth feature, although in robotic applications depth is often employed for object detection tasks. Objects usually have range discontinuities at their borders which can help to detect them. Models comprising depth as a feature typically use stereo vision to compute it [10, 2]. But stereo vision is computationally expensive, and only a fraction of the image pixels contribute to the computed 3D point clouds. As an alternative, 3D laser scanners are a class of sensors suitable for the fast acquisition of precise and dense depth or range information. The multi modal 3D ....

....Figure 1. Left: The custom 3D range finder mounted on top of the mobile robot KURT 2. Right: An office scene imaged with the 3D scanner in remission value mode, medium resolution, 256 360 pixels. model by Tsotsos et al. 14] Attentional systems using depth information can be found in [10] and [2] where stereo vision is applied to retrieve depth information. In robotics, attentional mechanisms are often used to direct the gaze (i.e. a camera) to interesting points in the environment. In [4] a robot shall look at people or toys and in [3] it uses attention to play at dominoes. ....

A. Maki, P. Nordlund, and J.-O. Eklundh. Attentional scene segmentation: Integrating depth and motion. CVIU, 78(3):351--373, 2000.

....attentional spotlight to other image locations. The WTA computation is applied to the di erence of the master map of attention and the inhibition map. Other models of visual attention have modi ed and extended this model, but the problem of depth and motion is seldom touched. Recently one model [8] has addressed the problem of integrating depth and motion information into attentional control by incorporating depth and motion as features for the computation of saliency. In contrast to this we will focus on the mechanism of attentional control itself, while the inclusion of such features is ....

A. Maki, P. Nordlund, , J.-O. Eklundh. Attentional scene segmentation: Integrating depth and motion. Computer Vision and Image Understanding, 78:351373, 2000.

No context found.

A. Maki, P. Nordlund, and J.-O. Eklundh, "Attentional Scene Segmentation: Integrating Depth and Motion from Phase," Computer Vision and Image Understanding, vol. 78, pp. 351-373, 2000.

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