Search

The problem of the local scale parameter selection for recognition techniques based on Guassian derivatives is addressed. The study shows promising results of recognition under variable scale and the results demonstrate that the approach holds promise for real-world applications.

An augmented reality tool for collaborative work called the MagicBoard, based on an ordinary white board, which is enhanced by a video-projector and steerable camera is described. Findings show that the white board scanner eliminates global luminosity differences by adaptive thresholding and the result can serve to digitize the content of the board.

A technique for visual recognition in which the appearance of objects is represented by families of surfaces in a local appearance space is presented. The results show that in many common situations, a single window is sufficient to obtain the correct recognition and the confidence of a recognition is easily estimated from the distance form the point to the surface, and the uniqueness determined by the number of surfaces which lie near the point.

A study was conducted to develop mobile robot position estimation based on a principal component evaluation of laser range information. The relative position and rotation at which scans best overlap were determined by scan correlation while a synthetic scan generator was used to obtain a sample of possible range data measurements. Results indicated that the approach is computationally efficient and robust to occlusion errors.

An analogy between an object recognition process and the transmission of information through a communication channel is presented based on the statistical representation of 3D objects by a collection of multidimensional receptive field histograms. Through the analogy, the capacity, redundancy, transinformation and efficiency of the object recognition process can either be defined or calculated. Thus, an active object recognition algorithm that is able to resolve the ambiguities of a single-view recognition algorithm has been developed.

Computer vision is essential in the interaction between man and machine. Computer vision and video projection allow a human to use convenient objects as digital input devices. Computer vision techniques for watching hands and recognizing gestures as well as detecting, tracking and recognizing faces were described. These techniques, when integrated with real time image processing and active control of camera parameters, can reduce the communications bandwidth needed for videophone and videoconference applications.

A study was conducted of the supervised learning for perception to action mapping for a mobile robot. The 'grow and learn' (GAL) network was chosen as the neural approach for supervised learning of a function for its incrementality and flexibility. The results indicate that this type of learning is possible using a GAL network because the GAL network allows the adaptation of the complexity of the system to the complexity of the task.

A reactive vision process is developed to integrate and control the operating visual processes of robots. The visual processes are represented by transformations which map signals from virtual sensors into commands. The method follows the behavioral approach to manipulation and mobility. Signals from the supervisory controller and those obtained from perception regulate the transition between the visual processes. Visual systems with various visual abilities can be constructed in a learnable and scalable way.

The concept of active control of perceptions emerged from the Second International Symposium on Intelligent Robotics Systems held in Grenoble, France from July 11 to 15, 1994. Two major issues were discussed, the problems of describing the transformation from perception to action and position estimation regarding navigation system. The idea of multiple representations to form a composite model from various sensors emerged during the symposium.

Autonomous mobile robots are navigated in partially-known environments by a navigation planning algorithm applied to discrete workspaces. 'Efficient path' format is applied to the first search algorithm to solve multiple constraints imposed by environmental specifications. The path computation depends on state representation that solves the constraints. The algorithm complexity properties are similar to those of the A* algorithm.

Four position estimation techniques are compared using occupancy grids to evaluate the grid efficiency in the estimation process. A global modeling technique is developed using local and global grids for extracting line segments from the occupancy grid which uses the Kalman filter equation for position estimation. Experimental results indicate the superiority of grid segment matching of local and global grids over direct grid matching.

Locomotion reflexes during reactive navigation of mobile robot systems are regulated by an artificial neural system. The robot is administered with a steering angle by the neural system that consists of the Fuzzy ART neural self-organizing classifier for space partitioning, neural associative memory and recording of different behaviors during motion. The learning system starting from zero is recorded for efficiency and few initial fluctuations.

The design and operating characteristics of a general purpose controller for robot vehicles are discussed. The controller's operation is regulated by a control protocol which is organized in three layers having different functions. Seven commands, three of which refer to the movement of the vehicle, are directly translated into calls to a set of interface procedures. These commands are then transmitted to the controller which in turn regulates the forward translation and rotation of the robot vehicle.

The Intelligent Robotics System (IRS) '93 symposium featured 'behavioral' approaches and techniques for neural and fuzzy control based on autonomous driving. The program schedule consisted of talks and visual shows with a demonstration of automatic driving system by the Carnegie Mellon University in Pittsburgh, PA. The symposium brought out the unified works of Eastern and Western Europe and North American scientists.