Raghavan Sudhakar

On improving eye feature extraction using deformable templates

Abstract An improved method of extracting eye features from facial images using eye templates is described. It retains all advantages of the deformable template method originally proposed by A. L. Yuille, P. W. Hallinan and D. S. Cohen ( Int. J. Comput. Vision 99–111 (1989)) and rectifies some of its weaknesses. This is achieved by the following modifications. First, the original eye template and the overall energy function to represent the most salient features of the eye are modified. Secondly, in order to simplify the issue of selecting weights for the energy terms, the value of each energy term is normalized in the range 0–1 and only two different weights are assigned. This weighting schedule does not require expert knowledge therefore it is more user friendly. Thirdly, all parameters of the template are changed simultaneously during the minimization process rather than using a sequential procedure. This scheme prevents some parameters of the eye template from being overly changed, helps the algorithm to converge to the global minimum, and reduces the processing time. The selection of initial parameters of the eye template is based on an eye window obtained in preprocessing. Experimental results are presented to demonstrate the efficacy of the algorithm. A comparison study of various processing schemes is also given.

Simultaneous robot/world and tool/flange calibration by solving homogeneous transformation equations of the form AX=YB

The paper presents a linear solution that allows a simultaneous computation of the transformations from robot world to robot base and from robot tool to robot flange coordinate frames. The flange frame is defined on the mounting surface of the end-effector. It is assumed that the robot geometry, i.e., the transformation from the robot base frame to the robot flange frame, is known with sufficient accuracy, and that robot end-effector poses are measured. The solution has applications to accurately locating a robot with respect to a reference frame, and a robot sensor with respect to a robot end-effector. The identification problem is cast as solving a system of homogeneous transformation equations of the form A/sub i/X=YB/sub i/,i=1, 2, ..., m. Quaternion algebra is applied to derive explicit linear solutions for X and Y provided that three robot pose measurements are available. Necessary and sufficient conditions for the uniqueness of the solution are stated. Computationally, the resulting solution algorithm is noniterative, fast and robust. >

On-line learning control of autonomous underwater vehicles using feedforward neural networks

A neural-network-based learning control scheme for the motion control of autonomous underwater vehicles (AUV) is described. The scheme has a number of advantages over the classical control schemes and conventional adaptive control techniques. The dynamics of the controlled vehicle need not be fully known. The controller with the aid of a gain layer learns the dynamics and adapts fast to give the correct control action. The dynamic response and tracking performance could be accurately controlled by adjusting the network learning rate. A modified direct control scheme using multilayered neural network architecture is used in the studies with backpropagation as the learning algorithm. Results of simulation studies using nonlinear AUV dynamics are described in detail. The robustness of the control system to sudden and slow varying disturbances in the dynamics is studied and the results are presented. >

Corner detection by a cost minimization approach

Abstract Corner detection, which is a valuable tool in biological and machine vision systems, is cast as a problem of cost optimization. The cost function is suitably devised to capture different desirable characteristics of corners such as edginess, curvature and region dissimilarity. The geometrical structure of the corner as well as the gray level variation of the image are accounted for in cost factors to evaluate the quality of corner configurations. The cost function is minimized using a simulated annealing algorithm. This approach also provides corner orientations and angles in addition to corner locations. The efficacy of the approach is demonstrated by experimental results.

Performance analysis of a DS/CDMA system with noncoherent M-ary orthogonal modulation in Nakagami fading

The probability of error performance of a direct sequence code-division multiple-access (DS/CDMA) system employing noncoherent M-ary orthogonal signaling in a Nakagami multipath fading channel is analyzed. A RAKE receiver structure with square-law demodulation is used at the receiver. The multiple-access interference are modeled as Gaussian and expressions derived for the exact probability of error. The performance is also evaluated in terms of the number of users that can be supported by the system at a given probability of error. The effect of correlated fading on system performance is also investigated by considering two correlation models, which can be characterized by a single correlation coefficient /spl rho/. In the first model, the correlation coefficient between any two diversity branches is constant. In the second model, it is assumed that the correlation coefficient between any two diversity branches decreases exponentially as the separation between them increases. For both models, it is found that the presence of correlation deteriorates system performance. The use of larger signal alphabets than binary modulation in conjunction with diversity reception provides a considerable performance improvement even in the presence of correlated fading.

A cascaded scheme for eye tracking and head movement compensation

The authors previously (1995) proposed an efficient method for tracking the eye movements. The proposed algorithm did not address the issue of compensating for head movements. Head movements are normally much slower than eye movements and can be compensated for using another tracking scheme for head position. In this paper, a hybrid method employing the two tracking schemes is developed. To this end, first a measurement model for the compensation of the head movement is formulated and then the overall tracking scheme is implemented by cascading two Kalman filters. The tracking of the iris movement is followed by the compensation of the head movement for each image frame. Experimental results are presented to demonstrate the accuracy aspects and the real-time applicability of the proposed approach.

A recurrent neural network controller and learning algorithm for the on-line learning control of autonomous underwater vehicles

A new on-line direct control scheme for the Autonomous Underwater Vehicles (AUV), using recurrent neural networks, is investigated. In the proposed scheme, the controller consists of a three-layer network architecture having feedforward input and output layers, and a totally recurrent hidden layer. All the interconnection strengths are synchronously updated using a computationally inexpensive learning algorithm called Alopex. The updating is based on the output error of the system directly, rather than using a transformed version of the error employed in the other neural network based direct control schemes. In the present implementation, the network starts from random initial conditions without needing any prior training, and learns the dynamics of the AUV to provide the correct control signal. Based on the simulation experiments using the nonlinear dynamics of an AUV, we demonstrate that the proposed learning algorithm and the network architecture provide stable and accurate tracking performance. We have also addressed the issue of robustness of the controller to system parameter variations as well as to measurement disturbances.

An improved scheme for direct adaptive control of dynamical systems using backpropagation neural networks

This paper presents an improved direct control architecture for the on-line learning control of dynamical systems using backpropagation neural networks. The proposed architecture is compared with the other direct control schemes. In this scheme the neural network interconnection strengths are updated based on the output error of the dynamical system directly, rather than using a transformed version of the error employed in other schemes. The ill effects of the controlled dynamics on the on-line updating of the network weights are moderated by including a compensating gain layer. An error feedback is introduced to improve the dynamic response of the control system. Simulation studies are performed using the nonlinear dynamics of an underwater vehicle and the promising results support the effectiveness of the proposed scheme.

Motion estimation from a sequence of stereo images: a direct method

This paper presents an approach for estimating motion from a stereo image sequence. First a stereo motion estimation model is derived using the direct dynamic motion estimation technique. The problem is then solved by applying a discrete Kalman filter that facilitates the use of a long stereo image sequence. Major issues in a motion estimation method are stereo matching, temporal matching, and noise sensitivity. In the proposed approach, owing to the use of temporal derivatives in the motion estimation model, temporal matching is not needed. The effort for stereo matching is kept to a minimum by the use of a parallel binocular configuration. Noise smoothing is achieved by the use of a sufficiently large number of measurement points and a long sequence of stereo images. Both simulation and experimental studies have been conducted to assess the effectiveness of the proposed approach. >

Depth estimation from a sequence of monocular images with known camera motion

Abstract This paper reports an approach of computing depth maps from a monocular image sequence, under the assumption that the camera motion is known. The direct depth estimation method is combined with the optical flow based method to improve estimation accuracy. The optical flow on and near moving edges are computed using a correlation technique. The optical flow information is then fused with gradient information to estimate depth not only on the moving edges but also in the internal regions. The depth estimation problem is formulated as a Kalman filter problem and is solved in three stages. In the prediction stage, the depth map estimated for the current frame, together with knowledge of the camera motion, is used to predict the depth and depth variance at each pixel in the next frame. In the estimation stage, a Kalman filter is employed to refine the predicted depth map. The resulting estimation algorithm takes into account the information from the neighboring pixels. In the smoothing stage, based on the error covariance information, morphological filtering is applied to reduce the effect of measurement noise and to fill in untrustable areas. Simulation results are provided to demonstrate the effectiveness of the proposed method.