Abstract

Depth Images are viable representations that can be computed from the real world using cameras and/or other scanning devices. The depth map provides 2-kD structure of the scene. A set of Depth Images can provide hole-free rendering of the scene. Multiple views need to blended to provide smooth hole-free rendering, however. Such a representation of the scene is bulky and needs good algorithms for real-time rendering and efficient representation. In this paper, we present a discussion on the Depth Image representation and provide a GPU-based algorithm that can render large models represented using DIs in real time. We then present a proxy-based compression scheme for Depth Images and provide results for the same. Results are shown on synthetic scenes under different conditions and on some scenes generated from images. Lastly, we initiate discussion on varying quality levels in IBR and show a way to create representations using DIs with different trade-offs between model size and rendering quality. This enables the use of this representation for a variety of rendering situations.

Abstract

We present a methodology for optimal target detection in a multi sensor surveillance system. The system consists of mobile sensors that guard a rectangular surveillance zone crisscrossed by moving targets. Targets penetrate the surveillance zone with poisson rates at uniform velocities. Under these conditions we present a motion strategy computation for each sensor such that it maximizes target detection for the next T time-steps. A coordination mechanism among sensors ensures that overlapping and overlooked regions of observation among sensors are minimized. This coordination mechanism is interleaved with the motion strategy computation to reduce detections of the same target by more than one sensor for the same time-step. To avoid an exhaustive search in the joint space of all the sensors the coordination mechanism constrains the search by assigning priorities to the sensors and thereby arbitrating among sensory tasks. A comparison of this methodology with other multi target tracking schemes verifies its efficacy in maximizing detections. "Sample" and "time-step" are used equivalently and interchangeably in this paper.

Abstract

In navigation that involves several moving agents or robots that are not in possession of each other's plans, a scheme for resolution of collision conflicts between them becomes mandatory. A resolution scheme is proposed in this paper specifically for the case where it is not feasible to have a priori the plans and locations of all other robots, robots can broadcast information between one another only within a specified communication distance, and a robot is restricted in its ability to react to collision conflicts that occur outside of a specified time interval called the reaction time interval. Collision conflicts are resolved through velocity control by a search operation in the robot's velocity space. The existence of a cooperative phase in conflict resolution is indicated by a failure of the search operation to find velocities in the individual velocity space of the respective robots involved in the conflict. A scheme for cooperative resolution of conflicts is modeled as a search in the joint velocity space of the robots involved in conflict when the search in the individual space yields a failure. The scheme for cooperative resolution may further involve modifying the states of robots not involved in any conflict. This phenomenon is characterized as the propagation phase where cooperation spreads to robots not directly involved in the conflict. Apart from presenting the methodology for the resolution of conflicts at various levels (individual, cooperative, and propagation), the paper also formally establishes the existence of the cooperative phase during real‐time navigation of multiple mobile robots. The effect of varying robot parameters on the cooperative phase is presented and the increase in requirement for cooperation with the scaling up of the number of robots in a system is also illustrated. Simulation results involving several mobile robots are presented to indicate the efficacy of the proposed strategy. © 2005 Wiley Periodicals, Inc.

Abstract

A methodology for real-time control of unmanned air vehicles (UAV) in the absence of a-priori knowledge of hostile territory is presented. The control methodology generates a sequence of waypoints to be pursued by the UV until the goal is reached. The controller computes the waypoints every time new information is obtained regarding the presence or absence of a hostile agent at a particular grid of the territory. The waypoints are the result of an A* search. The cost function of the search is a weighted combination of dangers arising due to probability of being hit by the attacks of hostile terrestrial agents on the ground and the danger of UAV grounded as a result of its prolonged stay over the hostile territory. The multi-agency in the system is due to the broadcast of newly observed information by an UAV to its remaining counterparts. The sequence of waypoints defines the path of the UAV to its goal. By varying the corresponding weights the paths can be altered to obtain a particular performance criterion. Simulation results are presented for various parametric combinations to validate the methodology.

Abstract

In this paper we expand on our previous efforts to evaluate strategies for improving tracking performance of a multi-sensor surveillance system. The quality of tracking performance is based on two measures: (i) the mean track quality (MTQ) and (ii) total expected number of misses (TEM). Five strategies of resource allocation were considered based on (a) local versus coordinated, (b) dedicated versus distracted and (c) benevolent. The simulation results suggest that a coordinated and distracted strategy of resource allocation yields the best result throughout while such a strategy coupled with benevolence further improves tracking performance marginally in most cases

Abstract

This paper presents a methodology for optimal target detection in a multi sensor surveillance system. The system consists of mobile sensors that guard a rectangular surveillance zone crisscrossed by moving targets. Targets percolate the surveillance zone in a poisson fashion with uniform velocities. Under these statistics this paper computes a motion strategy for a sensor that maximizes target detections for the next T time steps. A coordination mechanism between sensors ensures that overlapping areas between sensors is reduced. This coordination mechanism is interleaved with the motion strategy computation to reduce detections of the same target by more than one sensor. To avoid an exhaustive search in the joint space of all sensors the coordination mechanism constraints the search by assigning priorities to the sensors. A comparison of this methodology with other multi target tracking schemes verifies its efficacy in maximizing detections. A tabulation of these comparisons is reported in results section of the paper

Abstract

We present a framework for modeling and analysis for a surveillance network consisting of multiple sensors. Sensors monitor targets that crisscross a rectangular surveillance zone. When a sensor pursuits a target it leaves areas unguarded through which other targets can get past undetected. A methodology that computes the tracking time for a sensor such that a fraction of the targets expected to cross its home area is detected to an arbitrary probabilistic guarantee is presented based on the framework. Targets enter the surveillance zone according to Poisson statistics. The time spent by a target within a sensor's home area follows uniform random statistics. The home area of the sensor is the area guarded by it when it is stationed at its home position, its default position when it is not in pursuit of a target. The framework is further extended to situations where multiple sensors monitor the same home area. Simulation results presented corroborate with the probabilistic framework developed and verify its correctness for single as well as multi-sensor cases.

Abstract

We present an extension of the FCM over the loss functions used in the M-estimators of robust statistics akin to the generalization of the fuzzy-C-means algorithms over the norm distances [1]. The effect of these estimators in reducing the bias of the outliers while estimating the cluster prototypes are studied and compared. The comparisons have been done over synthetic data as well as simulated data consisting of range sensor readings representative of the objects in the neighborhood of a navigating mobile robot. For the synthetic data set the comparison is attempted over the popular FCM algorithm. For the sensory data set the Adaptive Fuzzy Clustering (AFC) algorithm [2] has been employed and extended over the loss functions of robust statistics. The AFC is utilized considering the shape of the objects encountered by the robot in a typical indoor environment.

Abstract

We present a document understanding system in which the arrangement of lines of text and block separators within a document are modeled by stochastic context free gram-mars. A grammar corresponds to a document genre; our system may be adapted to a new genre simply by replacing the input grammar. The system incorporates an optical character recognition system that outputs characters,their positions and font sizes. These features are combined to form a document representation of lines of text and separators. Lines of text are labeled as tokens using regular expression matching. The maximum likelihood parse of this stream of tokens and separators yields a functional labeling of the document lines. We describe business card and business letter applications.

Abstract

Multilingual speech recognition system is required for tasksthat use several languages in one speech recognition application. In thispaper, we propose an approach for multilingual speech recognition byspotting consonant-vowel (CV) units. The important features of spottingapproach are that there is no need for automatic segmentation of speechand it is not necessary to use models for higher level units to recognisethe CV units. The main issues in spotting multilingual CV units are thelocation of anchor points and labeling the regions around these anchorpoints using suitable classifiers. The vowel onset points (VOPs) have beenused as anchor points. The distribution capturing ability of autoassocia-tive neural network (AANN) models is explored for detection of VOPsin continuous speech. We explore classification models such as supportvector machines (SVMs) which are capable of discriminating confusableclasses of CV units and generalisation from limited amount of trainingdata. The data for similar CV units across languages are shared to trainthe classifiers for recognition of CV units of speech in multiple languages.We study the spotting approach for recognition of a large number of CVunits in the broadcast news corpus of three Indian languages