In this paper, we propose a new solution to the sparse image-based rendering (IBR) problem. Given two images taken from different viewpoints, our algorithm can accurately generate images from new viewpoints in between the original two views. This paper contributes to the sparse IBR in the following aspects: (1) Direct range space matching and multiple depth maps rendering. (2) Good rendering can be achieved with incorrect depth values, especially those textureless regions.(3) An efficient coarse-to-fine strategy to accelerate rendering and avoid problems caused by repetitive patterns. (4) A weak scene consistency constraint to simultaneously refine the rendering results and detect the occlusions. The proposed algorithm has been implemented and tested on real images and promising experimental results are demonstrated in the paper. 1

Estimating the number of people in a crowded environment is a central task in civilian surveillance. Most vision-based counting techniques depend on detecting individuals in order to count, an unrealistic proposition in crowded settings. We propose an alternative approach that directly estimates the number of people. In our system, groups of image sensors segment foreground objects from the background, aggregate the resulting silhouettes over a network, and compute a planar projection of the scene's visual hull. We introduce a geometric algorithm that calculates bounds on the number of persons in each region of the projection, after phantom regions have been eliminated. The computational requirements scale well with the number of sensors and the number of people, and only limited amounts of data are transmitted over the network. Because of these properties, our system runs in real-time and can be deployed as an untethered wireless sensor network. We describe the major components of our system, and report preliminary experiments with our first prototype implementation.

We present an O(nlog/sup 1+/spl epsiv// n)-time algorithm for computing the optimal robot motion that maintains line-of-sight visibility between a target moving inside a polygon with n vertices which may contain holes. The motion is optimal for the tracking robot (the observer) in the sense that the target either remains visible for the longest possible time, or it is captured by the observer in the minimum time when feasible. Thus, the algorithm maximizes the minimum time-to-escape. Our algorithm assumes that the target moves along a known path. Thus, it is an off-line algorithm. Our theoretical results for the algorithm's runtime assume that the target is moving along a shortest path from its source to its destination. This assumption, however is not required to prove the optimality of the computed solution, hence the algorithm remains correct for the general case.

This paper describes a new approach to keep a moving target into the field of view of a robot, despite obstacles. Unlike previous work, this approach does not assume that a complete or partial model of the target's behavior is available, nor that a map of the obstacles is given. At approximately 10 IIz, the robot constructs a local map of the environment from range data, localizes the target in this map, and computes its new velocity command. This command is selected to minimize a function that measures the risk that the target may escape the robot's field of view by crossing an occlusion ray created by an obstacle. The risk function defined in this paper combines a reactive term, whose minimization handles immediate threats, and a look-ahead term, whose minimization improves the robot's future ability to react to evading motions of the target. The experimental robotic system has been remarkably successful at tracking a target performing brisk maneuvers and swift turns behind obstacles.

Many applications require continuous monitoring of a moving target by a controllable vision system. Although the goal of tracking objects is not new, traditional techniques usually ignore the presence of obstacles and focus on imaging and target recognition issues. For a target moving among obstacles, the goal of tracking involves a complex motion problem: a controllable observer must anticipate that the target may become occluded by an obstacle and move to prevent such an event from occurring. The paper describes a strategy for computing the motions of a mobile robot operating in a 2-D workspace without prior knowledge of the target's intention or the distribution of obstacles in the scene. The proposed algorithm governs the motion of the observer based on current measurements of the target's position and the location of the local obstacles. The approach is combinatorial in the sense that the algorithm explicitly computes a description of the geometric arrangement between the target and the observer's visibility region produced by the local obstacles. The algorithm computes a continuous control law based on this description. The new tracking strategy has been implemented in a real-time robotic system.

This paper deals with the problem of computing the motions of one or more robot observers in order to maintain visibility of one or several moving targets. The targets are assumed to move unpredictably, and the distribution of obstacles in the workspace is assumed to be known in advance. Our algorithm computes a motion strategy by maximizing the shortest distance to escape $the shortest distance the target needs to move in order to escape the observer's visibility region. Three main points are discussed: 1) the design and implementation of a reactive planner; 2) the integration and testing of such a planner in a robot system which includes perceptual and control capabilities; and 3) the design and simulation of a motion planner for the task of maintaining visibility of two targets using two mobile observers.

This paper descirbes a placement strategy to compute a set of “good” locations where visual sensing will be most effective. Throughout this paper it is assumed that a {\em polygonal 2-D map} of a workspace is given as input. This polygonal map --- also known as a {\em floor plan} of {\em layout} --- is used to compute a set of locations where expensive sensing tasks (such as 3-D image acquisition) could be executed. A map-building robot, for example, can visit these locations in order to build a full 3-D model of the workspace.

We introduce the problem of computing robot motion strategies that maintain visibility of a moving target in a cluttered workspace. Both motion constraints (as considered in standard motion planning) and visibility constraints (as considered in visual tracking) must be satisfied. Additional criteria, such as the total distance traveled, can be optimized. The general problem is divided into two categories, on the basis of whether the target is predictable. For the predictable case, an algorithm that computes optimal, numerical solutions is presented. For the more challenging case of a partially-predictable target, two online algorithms are presented that each attempt to maintain future visibility with limited prediction. One strategy maximizes the probability that the target will remain in view in a subsequent time step, and the other maximizes the minimum time in which the target could escape the visibility region. We additionally discuss issues resulting from our implementation and experiments on a mobile robot system.

The authors address the problem of determining a time-dependent trajectory when a path planner provides a minimal distance path between two points. In contrast with most of the known methods that work in the velocity-time (v-t) space, theirs works directly in the velocity-arc length (v-s) space, which is the data provided by geometrical path planners. In a first instance, the method assumes a linear path to obtain a nonconstrained trajectory in the v-s plane. The dynamical constraints are then projected into this space to limit the original solution. As the algorithm uses discretized data, the authors provide its cost on space and time and the exact evaluation for the approximation error which depends on the desired accuracy.