Meghna Singh

Human Activity Recognition Based on Silhouette Directionality

Recent advances in computer vision and pattern recognition have fueled numerous initiatives that aim to intelligently recognize human activities. In this paper, we propose an algorithm for nonintrusive human activity recognition. We use an adaptive background-foreground separation technique to extract motion information and generate silhouettes (foreground) from the input videos. We then derive directionality-based feature vectors (directional vectors) from the silhouette contours and use the distinct data distribution of directional vectors in a vector space for clustering and recognition. We also exploit the dynamic characteristic of human motion in order to smooth decisions over time and reduce errors in activity recognition. Our approach is monocular, tolerant to moderate view changes, and can be applied to both frontal and lateral views of most activities. Experiments with short and long video sequences show robust recognition under conditions of varying view angles, zoom depths, backgrounds, and frame rates.

Gaussian and Laplacian of Gaussian weighting functions for robust feature based tracking

Object tracking algorithms found extensively in the computer vision literature either are inhibited by various assumptions such as simplicity of motion and shape characteristics of objects or are overly sensitive to noise. We propose and successfully test two new weighting functions for a feature-based object-tracking algorithm to achieve superior performance in tracking motion of non-rigid objects under noisy conditions. We present the implications of using the weighting functions in real and synthetic image sequences to overcome the noise produced at acquisition source (charge coupled device-CCD), or in the background environment. We also present a mechanism for determining the optimal weighting function based on image parameters, more specifically the edge characteristics of objects in the image.

Visual gesture recognition for ground air traffic control using the Radon transform

Human gesture recognition is an active topic of vision research which has applications in diverse fields such as collaborative virtual environments and robot teleoperation. We propose a novel method for the recognition of hand gestures, used by air marshals for steering aircraft on the runway, using the Radon transform. Various aspects of the algorithm, including acquisition, segmentation, labeling and recognition using the parametric Radon transform are addressed in this paper. A binary skeleton representation of the human subject is computed. The Radon transform is used to generate maxima corresponding to specific orientations of the skeletal representation. Feature vectors are extracted from the transform space by computing the normalized cumulative projections of the Radon transform on the angle axis. K-means clustering is then applied to recognize static gestures from the extracted features. This technique has the potential to provide information about the exact orientation of gesture segments and can find use in ground control of unmanned air vehicles. Experiments with image data corresponding to the various ground air traffic control gestures used in directing aircrafts, highlight the potential application of this approach.

Pose recognition using the Radon transform

Human pose recognition is an active topic of vision research that has applications in diverse fields such as collaborative virtual environments and robot teleoperation. We propose a novel method for the recognition of human pose using the Radon transform. A binary skeleton representation of the human subject is computed and information concerning the pose is extracted from the parametric Radon transform. The Radon transform generates maxima corresponding to specific orientations of the skeletal representation. A spatial maxima mapping technique is developed to recognize pose from the Radon transform maxima. This technique has the potential to provide information about the exact orientation of pose segments. Experiments with real image data highlight the advantages of this approach.

Event Dynamics Based Temporal Registration

Temporal registration is the establishment of correspondence between two (or more) temporal frames of video sequences, or 3-D volume data. In this paper, we propose to use event dynamics, a property that is inherent to an event and is thus common to all acquisitions of the event, for both global and local temporal registration of video sequences in order to generate high temporal resolution video. We compare our approach to a widely used linear interpolation based temporal registration algorithm and demonstrate that in the case of low temporal acquisition rate, a global event dynamics based approach, such as ours, has smaller temporal registration error. We also present a unique application of our work in solving 3-D (2D + time) high temporal resolution medical data visualization problem.

Optimization of Symmetric Transfer Error for Sub-frame Video Synchronization

In this work we present a method to synchronize video sequences of events that are acquired via uncalibrated cameras at unknown and dynamically varying temporal offsets. Unlike existing methods that synchronize videos of similar events (i.e., videos related to each other through the motion in the scene) up to an integer alignment, we establish sub-frame video synchronization. While contemporary synchronization algorithms implement a unidirectional alignment which biases the results towards a single reference sequence, we adopt a bi-directional or symmetrical alignment approach that results in a more optimal synchronization. To this end, we propose a novel symmetric transfer error which is dynamically minimized, and reduces the propagation of error from feature extraction and spatial mapping into temporal synchronization. The advantages of our approach are validated by tests conducted on (publicly available) real and synthetic sequences. We present qualitative and quantitative comparisons with another state-of-the-art algorithm. A unique application of this work in generating high-resolution 4D MRI data from multiple low-resolution MRI scans is described.

Image Based Temporal Registration of MRI Data for Medical Visualization

The capability of creating video data from MRI has many advantages in visualization for medical practitioners, including (i) not subjecting patients to harmful radiations, (ii) being able to monitor patients at short inter-exam time intervals, and (iii) being able to capture 3D volume data. The quality and speed with which MRI data can be acquired, however, poses a challenge towards supporting good quality visualization. In this work we present results from our preliminary attempts at enhancing the temporal resolution of video captured via MRI. Our initial focus is on visualization of swallowing and associated problems that are broadly categorized as Dysphagia. We present a method to register data from multiple swallows to generate high temporal resolution MRI videos.

Robust KLT tracking with Gaussian and Laplacian of Gaussian weighting functions

Object tracking algorithms extensively found in literature are either constrained with assumptions or are overly sensitive to noise. We propose and successfully test two new weighting functions for a feature-based object tracker to achieve superior tracking performance and noise immunity. The paper also presents a mechanism for image based optimal weighting function determination.

Efficient video sequences alignment using unbiased bidirectional dynamic time warping

In this paper, we propose an efficient technique to synchronize video sequences of events that are acquired via uncalibrated cameras at unknown and dynamically varying temporal offsets. Unlike other existing techniques that just take unidirectional alignment into consideration, the proposed technique considers symmetric alignments and compute the optimal alignment. We also establish sub-frame accuracy video alignment. The advantages of our approach are validated by tests conducted on several pairs of real and synthetic sequences. We present qualitative and quantitative comparisons with other existing techniques. A unique application of this work in generating high-resolution 4D MRI data from multiple low-resolution MRI scans is described.

Choice of low resolution sample sets for efficient super-resolution signal reconstruction

In applications such as super-resolution imaging and mosaicking, multiple video sequences are registered to reconstruct video with enhanced resolution. However, not all computed registration is reliable. In addition, not all sequences contribute useful information towards reconstruction from multiple non-uniformly distributed sample sets. In this paper we present two algorithms that can help determine which low resolution sample sets should be combined in order to maximize reconstruction accuracy while minimizing the number of sample sets. The first algorithm computes a confidence measure which is derived as a combination of two objective functions. The second algorithm is an iterative ranked-based method for reconstruction which uses confidence measures to assign priority to sample sets that maximize information gain while minimizing reconstruction error. Experimental results with real and synthetic sequences validate the effectiveness of the proposed algorithms. Application of our work in medical visualization and super-resolution reconstruction of MRI data are also presented.