Greg Okopal

Exact and approximate moments of a propagating pulse

We have recently developed a phase space approach for studying dispersive wave propagation. Using this approach, we have derived a simple approximation method. We show that this approximation gives the exact low order moments of the wave for all time. In particular, the mean motion and spread of a pulse are exact. The approximation also gives the exact moments of the spatial spectrum of the wave, for all orders. We also consider local moments, and show that the low-order local mean and spread are exact. We argue that the reason why the approximation works well for all time is precisely because it preserves important moments of the wave. We compare these results with the moments of the stationary phase approximation, which are accurate only for large time.

Robust Human Tracking Based on DPM Constrained Multiple-Kernel from a Moving Camera

In this paper, we attempt to solve the challenging task of precise and robust human tracking from a moving camera. We propose an innovative human tracking approach, which efficiently integrates the deformable part model (DPM) into multiple-kernel tracking from a moving camera. The proposed approach consists of a two-stage tracking procedure. For each frame, we first iteratively mean-shift several spatially weighted color histograms, called kernels, from the current frame to the next frame. Each kernel corresponds to a part model of a DPM-detected human. In the second step, conditioned on the tracking results of these kernels on the later frame, we then iteratively mean-shift the part models on that frame. The part models are represented by histogram of gradient (HOG) features, and the deformation cost of each part model provided by the trained DPM detector is used to constrain the movement of each detected body part from the first step. The proposed approach takes advantage of not only low computation owing to the kernel-based tracking, but also robustness of the DPM detector without the need of laborious human detection for each frame. Experimental results have shown that the proposed approach makes it possible to successfully track humans robustly with high accuracy under different scenarios from a moving camera.

Voice activity detection using subband noncircularity

Many voice activity detection (VAD) systems use the magnitude of complex-valued spectral representations. However, using only the magnitude often does not fully characterize the statistical behavior of the complex values. We present two novel methods for performing VAD on single- and dual-channel audio that do completely account for the second-order statistical behavior of complex data. Our methods exploit the second-order noncircularity (also known as impropriety) of complex subbands of speech and noise. Since speech tends to be more improper than noise, higher impropriety suggests speech activity. Our single-channel method is blind in the sense that it is unsupervised and, unlike many VAD systems, does not rely on non-speech periods for noise parameter estimation. Our methods achieve improved performance over other state-of-the-art magnitude-based VADs on the QUT-NOISE-TIMIT corpus, which indicates that impropriety is a compelling new feature for voice activity detection.

Deformable multiple-kernel based human tracking using a moving camera

In this paper, we propose an innovative human tracking algorithm, which efficiently integrates the deformable part model (DPM) into the multiple-kernel based tracking using a moving camera. By representing each part model of a DPM detected human as a kernel, the proposed algorithm iteratively mean-shift the kernels (i.e., part models) based on color appearance and histogram of gradient (HOG) features. More specifically, the color appearance features, in terms of kernel histogram, are used for tracking each body part from one frame to the next, the deformation cost provided by DPM detector is further used to constrain the movement of each body kernel based on the HOG features. The proposed deformable multiple-kernel (DMK) tracking algorithm takes advantage of not only low computation owing to the kernel-based tracking, but also robustness of the DPM detector. Experimental results have shown the favorable performance of the proposed algorithm, which can successfully track human using a moving camera more accurately under different scenarios.

Estimating the noncircularity of latent components within complex-valued subband mixtures with applications to speech processing

This paper describes an approach that estimates the circularity coefficients of multiple underlying components within complex subbands of an additive mixture of voiced speech and noise via the strong uncorrelating transform (SUT). For the SUT to be effective, the latent source signals must have unique nonzero circularity coefficients; this requirement is satisfied by using narrow filters to impose a degree of noncircularity upon what would typically be circular noise. The circularity coefficient estimates are then used for voice activity detection, pitch tracking, and enhancement.

Speech analysis with the strong uncorrelating transform

The strong uncorrelating transform (SUT) provides estimates of independent components from linear mixtures using only second-order information, provided that the components have unique circularity coefficients. We propose a processing framework for generating complex-valued subbands from real-valued mixtures of speech and noise where the objective is to control the likely values of the sample circularity coefficients of the underlying speech and noise components in each subband. We show how several processing parameters affect the noncircularity of speech-like and noise components in the subband, ultimately informing parameter choices that allow for estimation of each of the components in a subband using the SUT. Additionally, because the speech and noise components will have unique sample circularity coefficients, this statistic can be used to identify time-frequency regions that contain voiced speech. We give an example of the recovery of the circularity coefficients of a real speech signal from a two-channel noisy mixture at -25 dB SNR, which demonstrates how the estimates of noncircularity can reveal the time-frequency structure of a speech signal in very high levels of noise. Finally, we present the results of a voice activity detection (VAD) experiment showing that two new circularity-based statistics, one of which is derived from the SUT processing, can achieve improved performance over state-of-the-art VADs in real-world recordings of noise.

Ground-Moving-Platform-Based Human Tracking Using Visual SLAM and Constrained Multiple Kernels

This paper proposes a robust ground-moving-platform-based human tracking system, which effectively integrates visual simultaneous localization and mapping (V-SLAM), human detection, ground plane estimation, and kernel-based tracking techniques. The proposed system systematically detects humans from recorded video frames of a moving camera and tracks the humans in the V-SLAM-inferred 3-D space via a tracking-by-detection scheme. To efficiently associate the detected human frame by frame, we propose a novel human tracking framework, combining the constrained-multiple-kernel tracking and the estimated 3-D information (depth), to globally optimize the data association between consecutive frames. By taking advantage of the appearance model and 3-D information, the proposed system not only achieves high effectiveness but also well handles occlusion in the tracking. Experimental results show the favorable performance of the proposed system, which efficiently tracks humans in a camera equipped on a ground-moving platform such as a dash camera and an unmanned ground vehicle.

Propagation-invariant classification of sounds in channels with dispersion and absorption

In a previous paper [G. Okopal et al., J. Acoust. Soc. Am. 123, 832-841 (2008)], a method to obtain features of a wave that are unaffected by dispersion, per mode, was developed for improving classification of underwater sounds (e.g., sonar backscatter). The current paper builds on this work and presents additional contributions. First, it is shown that the dispersion-invariant moments developed previously are not invariant to frequency-dependent attenuation (absorption); consequently, their classification performance degrades in such channels. Second, a feature extraction method is developed to obtain features that are invariant to dispersion, and to two forms of absorption (known a priori): namely, absorption that yields spectral magnitude attenuation (in dB) that is linear with frequency, and linear with log-frequency. Third, the relationship of these absorption- and dispersion-invariant moment (ADIM) features to the cepstrum of the wave is examined, and it is shown that cepstral moments are also invariant to dispersion, and to the first form of absorption for odd-order moments. Finally, simulations are conducted to illustrate the performance of the ADIMs and cepstral moments on classifying the backscatter from steel shells in a dispersive channel with absorption. Receiver operator characteristic curves quantify the superior discriminability of the ADIMs and cepstral moments compared to ordinary moments.

Tracking drifting surface objects with aerial infrared and electro-optical sensors

Recently, researchers at the Applied Physics Laboratory at the University of Washington collected a unique dataset by suspending two cameras, one infrared and one electro-optical, from a balloon. This apparatus was then used to image objects drifting on the surface of Lake Washington. The authors took that data and built a processing stream to track the movements of those drifting surface objects.

Environmentally invariant features for classification of active sonar signals.

Dispersion and damping (frequency‐dependent spreading and attenuation) can be significant in shallow water sound propagation. These propagation‐induced effects can be detrimental to classification of active sonar returns because the observed backscatter depends not only on the target, but also on the propagation environment and how far the wave has traveled, resulting in increased variability in the received signals. We address this problem by extracting propagation‐invariant features from the wave that can be used in an automatic classifier. In this talk, we review various moment‐like features we have developed that are invariant per mode to dispersion and damping. Simulations of the backscatter from different steel shells propagating in a channel with dispersion and damping are presented to demonstrate the classification utility of the various invariant features. [Work supported by ONR Grant N00014‐06‐1‐0009.]