This paper introduces a technique for extracting structure and motion using directionally selective matches between linear features. A world-centered coordinate system is used to make these computations without the intermediate calculation of depth. In order to constrain the possible structure and motion configurations, we assume that the threedimensional direction of gravity relative to each image frame is known. The direction of gravity, along with the directionally selective linear feature matches, form a set of quadratic equations which can be used to determine structure and motion. 1 Introduction The extraction of environmental structure and motion from a sequence of two-dimensional images is a common problem in computer vision. Typically solutions to this problem are expressed in camera-centered coordinate systems where environmental geometry is specified by the depth along an image feature's ray of projection. Unfortunately, parameters computed from this cameracen...

this memory overhead could be overwhelming. For example, on a dual SMP system with 256 megabytes of physical memory, the number of ppage objects is well over 60,000. If each ppage object uses a scalable spinlock, the amount of memory overhead reaches 32 megabytes. Using the trivial implementation, each spinlock resides on a cache line, yielding a small memory overhead of less than 2 megabytes. The second benefit concerns latency. The trivial algorithm has lower lock acquire and release latencies because it does not need to perform enqueue and dequeue operations. Since latency is a concern with fine grained locking, the trivial algorithm becomes attractive

Abstract not found

We address the problem of observing an agent. Weadvocate a modeling approach for the visual system and its observer, where a discrete event dynamic system (DEDS) framework is developed and" events" are defined as ranges on parameter subsets. The dynamic recursivecontext for finite state machines (DRFSM) is described with some applications in the inspection and reverse engineering domain. We propose a system for observing a manipulation process, where a robot hand manipulates an object. We recognize the hand/object interaction over time and a stabilizing observer is constructed. Low-level modules are developed for recognizing the events that causes state transitions within the dynamic manipulation system. The work examines closely the possibilities for errors, mistakes and uncertainties in the manipulation system, observer construction process and event identification mechanisms. The DRFSM DEDS systems utilizes different tracking techniques in order to observe and recognize tasks and agents in an active, adaptive and goal-directed manner.