Gregory Hager and Kentaro Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Technical report, Department of Computer Science, Yale University, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... Design Due to the fact that speed of standard computer processor continues to increase, prices continue to drop, and multi processor architectures are becoming more widely available and accessible, some interesting problems such as visual tracking can be carried out on standard workstations [5]. The Visual Tracking System used by this thesis work as an application program is a vivid example in multimedia computing which needs to interact with end user dynamically. In the remainder of this chapter, we will discuss some design issues for the Visual Tracking System and related networking ....

....for teaching and research. The model of computation is one of independent tracking agents that communicate information to one another in a hierarchical network. The system is object oriented. As a preview of the system structure, the figure 4. 1 shows the current set of classes defined in system [5]. There are two essential ideas for the design of Visual Tracking System. They are window based image processing and state based programming of hierarchy networks of objects being tracked [5] A brief summary of these classes is given below. 31 Framework Extension CompFeature CFV class ....

[Article contains additional citation context not shown here]

Gregory Hager and Kentaro Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Technical report, Department of Computer Science, Yale University, 1994.

....visual feature tracking optimized to be simple to configure at the user level. Its performance relies on using only small sub windows for tracking and representing features in canonical configuration which allow faster computation of the actual feature position. More information can be found in [11, 6, 21]. It is free available for non commercial use. Human Interface Technology Lab. www.hitl.washington.edu) ARToolkit: provides a portable computer vision tracking library. It is a software library that can be used to calculate camera position and orientation relative to physical markers in real ....

Gregory D.Hager and Kentaro Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Technical report, Yale University Computer Science Dept., 1994.

....for evaluation of geometric constraints on the feature. The current implementation of Argus only has tracking agents for straight edge segments. Other tracker types are under development, including a color blob tracker, a corner tracker, and a parametric curve tracker. The Yale tracking system[5] already has tracking algorithms for various types of features, so adding them to Argus should be straightforward. Figure 1 shows a snapshot of an edge segment tracker following the top edge of a white flash card with a diamond shape printed on it. The tracker is seeing a 20 by 51 pixel window ....

Gregory D. Hager and Kentaro Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Draft documentation, August 1994.

....Our monolithic vision routines do not provide the AEexibility needed to exploit up to the second context, certainly not in novel ways. What we need is the composability that is available in most vision software packages, but not at the price required by these packages. For example, XVision [5] allows new operators to be derived very AEexibly from others, but requires new C classes and recompilation to use them. Khoros Cantata is an interpreter that can combine operators into new routines at run time. Unfortunately Cantata s design is not suited for on board real time processing, ....

Gregory D. Hager and Kentaro Toyama. A framework for real-time window-based tracking using ooe-the-shelf hardware. Technical Report 0.95 Alpha, Yale University Computer Science Dept., 1994.

....when people tilt their heads to the side and cause a rotation of the feature. So tracking of the facial feature fails and there is no way to recover from this error. Implementations of gesture interfaces based on image correlation have been previously reported [ Azarbayejani et al. 1993; Hager and Toyama, 1996; 1995 ] The work of [ Azarbayejani et al. 1993 ] reported on the manipulation of a virtual reality clone. This system automatically acquired the tracking templates at runtime using heuristic filters. With this approach it is not possible to determine the initial state of features e.g. eyes are open ....

....reality clone. This system automatically acquired the tracking templates at runtime using heuristic filters. With this approach it is not possible to determine the initial state of features e.g. eyes are open and the position of the face can only be measured relative to the initial image. The [ Hager and Toyama, 1996; 1995 ] implementation uses deformable template matching performed in software to track a face. This system does not exploit the geometry of the facial features to assist in tracking or error recovery so only small rotations of the head are allowed. Related to the template based approach is the method ....

G.D. Hager and K. Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Technical report, Yale University, Department of Computer Science, 1995.

....the system to scale up on a multi cpu host. Off the shelf hardware: The CPU price performance and even the raw speed of Intel and Power PC based machines appears increasingly attractive compared to that of RISC based UNIX machines, especially when special purpose frame grabbing hardware is added [7]. Our target was to produce a system that could run on off theshelf PC hardware at least half as quickly as our current RISC Datacube system, and that would surpass it within a year or so as performance continues to improve. General purpose: The system should apply to more than real time robot ....

....real time gesture recognition of Perseus, and off line image database indexing, and be useful for teaching computer vision. The last especially implies that it should be easy to use and to extend. We considered several alternative image processing systems (Vista, Susan, MIL, Teleos AVP, XVision [7]) and found that some of them would be well suited for onboard image processing and robot vision using off the shelf PC hardware, but that none of them offered both the runtime flexibility, performance, and remote controllability we wanted. This paper describes the major architectural features of ....

Gregory D. Hager and Kentaro Toyama. A framework for real-time window-based tracking using offthe -shelf hardware. Technical Report 0.95 Alpha, Yale University Computer Science Dept., 1994.

....For example, Khoros Cantata is an interpreter that can combine operators into new routines at run time. Unfortunately Cantata s design is not suited for real time processing, mainly because it uses expensive inter process communication to pass copies of images from one module to another. Xvision (Hager Toyama 1994) allows new operators to be derived very flexibly from others, but requires new C classes and recompilation to use them. Although no vision software package provides the performance and run time flexibility we seek, we don t want to write an entirely new system if there is one that provides an ....

Hager, G. D., and Toyama, K. 1994. A framework for real-time window-based tracking using off-the-shelf hardware. Technical Report 0.95 Alpha, Yale University Computer Science Dept.

....fairs poorly; when shears and rotations are expected, computing time increases drastically; and if the target image is almost constant or if it contains repetitive elements, the matches are susceptible to minute effects of noise. At the other end of the spectrum lie simple edge detection methods [2, 4, 6]. These methods are temporally independent in that they make decisions about a feature s state by observing only the current window. We define s = s x ; s y ; s ) T where s x and s y denote the center of a finite line segment, and denotes the direction of the line in image coordinates. S ....

....We now describe how Coarse SSD is employed in two real tracking applications. These applications are separated into two levels, where the lower level deals with local tracking and feature matching, and the upper level directs window placement and computes the state of the entire object. See [6] for more information on layered feature tracking. 5.1 Low Level Tracking The algorithm chosen above is incorporated into a 1 dimensional edge detector. Each window is simply a single line of pixels which monitors a part of the contour. The position and orientation of the windows (the w s) are ....

[Article contains additional citation context not shown here]

G. Hager, S. Puri, and K. Toyama. A framework for real-time window-based tracking using off-the-shelf hardware. Yale Tech Report: Yale-TR-988, 1993.

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