Aaron L. Robinson

Modeling target acquisition tasks associated with security and surveillance

Military sensor applications include tasks such as the surveillance of activity and searching for roadside explosives. These tasks involve identifying and tracking specific objects in a cluttered scene. Unfortunately, the probability of accomplishing these tasks is not predicted by the traditional detect, recognize, and identify (DRI) target acquisition models. The reason why many security and surveillance tasks are functionally different from the traditional DRI tasks is described. Experiments using characters and simple shapes illustrate the problem with using the DRI model to predict the probability of identifying individual objects. The current DRI model is extended to predict specific object identification by including the frequency spectrum content of target contrast. The predictions of the new model match experimental data.

LWIR and MWIR fusion algorithm comparison using image metrics

This study determines the effectiveness of a number of image fusion algorithms through the use of the following image metrics: mutual information, fusion quality index, weighted fusion quality index, edge-dependent fusion quality index and Mannos-Sakrison’s filter. The results obtained from this study provide objective comparisons between the algorithms. It is postulated that multi-spectral sensors enhance the probability of target discrimination through the additional information available from the multiple bands. The results indicate that more information is present in the fused image than either single band image. The image quality metrics quantify the benefits of fusion of MWIR and LWIR imagery.

Tilted surfaces in short-wave infrared imagery: speckle simulation and a simple contrast model

An effective simulation of speckle with tilted surfaces illuminated by short-coherence-length lasers is presented. Two new tools for assessing speckle and/or its contrast under these conditions are developed and validated. The first is a simulation of the time-domain tilted-surface effects that provides speckle imagery. The second is a simple intuitive model for contrast derived from speckle reduction due to averaging. Field results of speckle imagery for a laser-illuminated target and a short-wave infrared imager validate the simulation. Simulated speckle is compared visually with the actual speckle. Also, contrasts of the speckle generated by simulation and actually imaged in the field are compared for one tilt angle. Existing analytical models of the contrast also validate the simulated speckle contrast. Contrast-versus-angle characteristics of simulated speckle are compared with the general analytical contrast model for speckle from tilted surfaces.

A NEW APPROACH TO WIRELESS CHANNEL MODELING USING FINITE MIXTURE MODELS

This paper presents a new approach to modeling a wireless channel using finite mixture models (FMM). Instead of the conventional approach of using non mixtures (single) probability distribution functions, FMMs are used here to model the channel impulse response amplitude statistics. To demonstrate this, a FMM based model of Ultrawideband (UWB) channels amplitude statistics is developed. In this research, finite mixture models composed of combinations of constituent PDFs such as Rayleigh, Lognormal, Weibull, Rice and Nakagami are used for modeling the channel amplitude statistics. The use of FMMs is relevant because of their ability to characterize the multimodality in the data. The stochastic expectation maximization (SEM) technique is used to estimate the parameters of the FMMs. The resultant FMMs are then compared to one another and to non-mixture models using model selection techniques such as Akaike’s Information Criteria (AIC). Results indicate that models composed of a mixture of Rayleigh and Lognormal distributions consistently provide good fits for most of the impulses of the UWB channel. Other model selection techniques such as Minimum Description Length (MDL) and Accumulative Predictive Error (APE) also confirmed this finding. This selection of FMM based on Rayleigh and Lognormal distributions is true for both the industrial as well as the university environment channel data

Sparse detector sensor model

This paper details the development of a modularized system level model of a sensor whose detector dimensions may be small with respect to the distance between adjacent detectors. The effects of individual system components and characteristics such as target to background properties, collection optics, detectors, and classifiers will be modeled. These individual effects will then be combined to provide an overall system performance model. The model will facilitate design trade offs for Unattended Ground Sensors. The size and power restrictions of these sensors often preclude these sensors from being effective in high resolution applications such as target identification. Consequently, existing imager performance models are not directly applicable. However, these systems are well suited for applications such as broad scale classifications or differentiations between targets such as humans, animals or small vehicles. Furthermore, these sensors do not have to be spaced closely together to be effective in these applications. Therefore, the demand for these sensors is increasing for both the military and homeland security.

MWIR Persistent Surveillance Performance for Human and Vehicle Backtracking as a Function of Ground Sample Distance and Revisit Rate

Real MWIR Persistent Surveillance (PS) data was taken with a single human walking from a known point to different tents in the PS sensor field of view. The spatial resolution (ground sample distance) and revisit rate was varied from 0.5 to 2 meters and 1/8th to 4 Hz, respectively. A perception experiment was conducted where the observer was tasked to track the human to the terminal (end of route) tent. The probability of track is provided as a function of ground sample distance and revisit rate. These results can help determine PS design requirements for tracking and back-tracking humans on the ground. This paper begins with a summary of two previous simulation experiments: one for human tracking and one for vehicle tracking.

Model selection and Kolmogorov-Smirnov test for ultrawideband channel modeling

Accumulative Predictive Error has been previously used for time series modeling of psychological response time data. In this paper we extend its applicabilty to the identification of tap amplitude statistics of ultrawideband communication channels. We also present channel modeling results from two other model selection techniques: Minimum Description Length and Akaikes Information Criterion. Channel models are also identified by hypothesis testing using Kolmogorov-Smirnov test. The results agree with recent findings that Rayleigh distribution can still be used to model tap amplitude statistics of line of sight ultrawideband communication channels.

Multispectral infrared image classification using filters derived from independent component analysis

Spectral-spatial independent component analysis (ICA) basis functions of visible color images are similar to some processing elements in the human visual systems in that they resemble Gabor filters and show color opponencies. In this research we studied combined spectral-spatial ICA basis functions of multispectral mid wave infrared (MWIR) images. These ICA spectral-spatial basis functions were then used as filters to extract features from multispectral MWIR images for classification. The images were captured in the 3.0–5.0 µm, 3.7–4.2 µm, and 4.0–4.5 µm bands using a multispectral MWIR camera. In the proposed algorithm, phase relationships between the basis functions indicate how the extracted features from the different spectral band images can be combined. We used classification performance to compare features obtained by filtering using multispectral ICA basis functions, multispectral principal component analysis basis functions, and Gabor filters.

An efficient turbulence simulation algorithm

Turbulence mitigation techniques require input data representing a wide variety of turbulent atmospheric and weather conditions in order to produce robust results and wider ranges of applicability. In the past, this has implied the need for numerous data collection equipment items to account for multiple frequency bands and various system configurations. However, recent advancements in turbulence simulation techniques have resulted in viable options to real-time data collection with various levels of available simulation accuracy. This treatment will detail the development and implementation of an extension to the second order statistical turbulence simulation model presented by Repasi1 and others. The Repasi model is extended to include the effects of various wavelengths, optical configurations, and short exposure imaging on angle of arrival fluctuation statistics. The result of the development is an atmospheric turbulence simulation technique that is physics-based but less computationally intensive than phase-based or deflector screen approaches. In these cases, the statistical approach detailed in this paper provides the user with an opportunity to obtain a better trade-off between accuracy and simulation run-time. The mathematical development and reasoning behind the changes to the previous statistical model will be presented, and sample imagery produced by the extended technique will be included. The result is a model that captures the major turbulence effects required for algorithm development for large classes of mitigation techniques.

Low- to mid-altitude temporal/spatial tracking requirements

We describe the development, experimentation, collected data, and results of research designed to gain an understanding of the temporal and spatial image collection guidelines for tracking humans. More specifically, we seek a quantitative understanding of the relationship between human observer performance and the spatial and temporal resolutions. We measure performance as a function of the number of video frames per second, the imager spatial resolution, and the ability of the observer to accurately determine the destination of a moving human target. Our research is restricted to data and imagery collected from typical modern, low- to mid-altitude, persistent surveillance platforms using a wide field of view. The ability of the human observer to track a human target unaided was determined by the observers completion of carefully designed perception experiments. In these experiments, the observers were presented with simulated imagery from the U.S. Army Night Vision and Electronic Sensor Directorates EOSim urban terrain simulator. The details of the simulated targets and backgrounds, and the design of the experiments as well as their associated results, are included in this treatment.