Carol T. Christou

Vehicle detection and localization using Unattended Ground Magnetometer Sensors

This analysis was completed as part of a larger Modeling and Simulation effort to estimate algorithm-level Measures of Performance (MOP), such as the probability of detection (PD) and the probability of identification (PID) of a vehicle or person transiting through an area of interest. The present work focuses on MOPs for Unattended Ground Magnetometer Sensors, which may be used to detect passing vehicles and estimate their bearing relative to the magnetometer position. In the first phase of the analysis, we concentrate on the probability of detection as a function of vehicle speed and distance (i.e., point of closest approach (CPA)) from the sensor. In the second phase, we try to localize the vehicle by extracting its relative bearing with respect to the magnetometer from the two orthogonal induced magnetic field measurements. The derivations are based on the assumption that a road vehicle may be approximated as a prolate homogeneous ellipsoid, as well as the assumption of uniform linear motion. Results show that, for speeds below 30 MPH, the maximum detection ranges (for PD = 0.5) are on the order of 40 meters for two-axis fluxgate magnetometers and for the operational parameters used in this analysis.

Vehicle acoustic classification in netted sensor systems using Gaussian mixture models

Acoustic vehicle classification is a difficult problem due to the non-stationary nature of the signals, and especially the lack of strong harmonic structure for most civilian vehicles with highly muffled exhausts. Acoustic signatures will also vary largely depending on speed, acceleration, gear position, and even the aspect angle of the sensor. The problem becomes more complicated when the deployed acoustic sensors have less than ideal characteristics, in terms of both the frequency response of the transducers, and hardware capabilities which determine the resolution and dynamic range. In a hierarchical network topology, less capable Tier 1 sensors can be tasked with reasonably sophisticated signal processing and classification algorithms, reducing energy-expensive communications with the upper layers. However, at Tier 2, more sophisticated classification algorithms exceeding the Tier 1 sensor/processor capabilities can be deployed. The focus of this paper is the investigation of a Gaussian mixture model (GMM) based classification approach for these upper nodes. The use of GMMs is motivated by their ability to model arbitrary distributions, which is very relevant in the case of motor vehicles with varying operation modes and engines. Tier 1 sensors acquire the acoustic signal and transmit computed feature vectors up to Tier 2 processors for maximum-likelihood classification using GMMs. In a binary classification task of light-vs-heavy vehicles, the GMM based approach achieves 7% equal error rate, providing an approximate error reduction of 49% over Tier 1 only approaches.

PIGLT: A Pollen Identification and Geolocation system for forensic applications

The Department of Homeland Security, Science and Technology Directorate (DHS/S&T) is exploring the feasibility to geolocate pollen grains found on goods or people for compliance with U.S. import laws and criminal forensics. A multidisciplinary team built the Pollen Identification and Geolocation Technology (PIGLT) system to help users identify pollen samples and perform geolocation. Identification is performed using either traditional family, genus and species information, or a morphological ID system based on an existing database of herbaria samples. As the user makes morphological decisions, visual aids help exclude pollen taxa that lack given attributes. The user systematically lowers the number of matches until the number is small enough for visual identification. PIGLT has ~5 images per sample, but experiments with Z-stack imagery may positively affect human identification. Given grain identities, geolocation proceeds using distributions developed using Maxent. The database is implemented in PostgreSQL and the userinterface uses Django, a high-level Python Web framework.

Geolocation analysis using Maxent and plant sample data

A study was conducted to assess the feasibility of geolocation based on correctly identifying pollen samples found on goods or people for purposes of compliance with U.S. import laws and criminal forensics. The analysis was based on Neotropical plant data sets from the Global Biodiversity Information Facility. The data were processed through the software algorithm Maxent that calculates plant probability geographic distributions of maximum entropy, subject to constraints. Derivation of single and joint continuous probability densities of geographic points, for single and multiple taxa occurrences, were performed. Statistical metrics were calculated directly from the output of Maxent for single taxon probabilities and were mathematically derived for joint taxa probabilities. Predictions of likeliest geographic regions at a given probability percentage level were made, along with the total corresponding geographic ranges. We found that joint probability distributions greatly restrict the areas of possible provenance of pollen samples.

Automated pollen identification system for forensic geo-historical location applications

The use of pollen grain analysis for forensic geo-historical location has been explored for several decades, yet it is not widely adopted in the United States. We confirmed significant improvement in geographic precision, i.e., from 2.5×107 to 1.2×105 km2, by simultaneously applying flowering plant data from four different taxa at the genus and species levels. Moreover, when we calculated precision using collected pollen data, we found that co-occurring, pairwise genus-level distinctions based on expert-provided indicator taxa resulted in average precision values of 4° and 4.5° in latitude and longitude, respectively - corresponding to roughly 1.8×105 km2. We also applied computer vision techniques to identify morphologically similar pollen grains, which resulted in grain-identification error rates of 2.18% and 6.24% at the genus and species levels, respectively, surpassing previously published records. Collectively, our results demonstrate that algorithmic identification of species-specific pollen morphology, founded on established computer vision techniques, when combined with species-level pollen distribution, has the potential to revolutionize the scope, accuracy, and precision of forensic geographic attribution.

An Innovative method to determine multi-system performance for the detection of clandestine tunnels

The threat posed by underground clandestine tunnels has been a growing concern for law enforcement and national security. Cross-border tunnels have been used by smugglers with the intention of avoiding border security for trafficking people, drugs, firearms, and other illegal materials. The ability to detect these tunnels is vital to achieving effective border control. This paper describes the development of an innovative method to model and assess the performance of various sensor systems in the geological region of their intended use, and to determine the best sensing modalities and equipment to operate in that region. The method includes: 1) Investigation and characterization of the regional representative geologic and geophysical properties of the shallow subsurface soil and environmental conditions along the southern US border; 2) Sensor performance modeling and simulation studies for various sensor systems components/configurations, tunnel characteristics, surface and subsurface environmental and soil conditions; and 3) Validation and verification of the performance via tunnel detection testbed development and demonstration. The results of these combined efforts will be used to develop and implement an integrated sensor performance characterization suite to assist in identification of the most suitable methods and/or equipment to detect tunnels in a variety of locales. A case study illustrating our approach applied to an area along the southern border using available field data to characterize the sensor performance indicates the methodology can yield accurate predictions of sensor performance in various geologies and indigenous environmental noise. For the simulations to be useful, more work is planned to improve the accuracy of the sensor models, the precision of the geophysical databases, and to overcome the long execution times required for the models to run.

Simulation of vehicle acoustics in support of netted sensor research and development

The MITRE Corporation has initiated a three-year internally-funded research program in netted sensors, the first-year effort focusing on vehicle detection for border monitoring. An important component is developing an understanding of the complex acoustic structure of vehicle noise to aid in netted sensor-based detection and classification. This presentation will discuss the design of a high-fidelity vehicle acoustic simulator to model the generation and transmission of acoustic energy from a moving vehicle to a collection of sensor nodes. Realistic spatially-dependent automobile sounds are generated from models of the engine cylinder firing rates, muffler and manifold resonances, and speed-dependent tire whine noise. Tire noise is the dominant noise source for vehicle speeds in excess of 30 miles per hour (MPH). As a result, we have developed detailed models that successfully predict the tire noise spectrum as a function of speed, road surface wave-number spectrum, tire geometry, and tire tread pattern. We have also included realistic descriptions of the spatial directivity patterns for the engine harmonics, muffler, and tire whine noise components. The acoustic waveforms are propagated to each sensor node using a simple phase-dispersive multi-path model. A brief description of the models and their corresponding outputs is provided.

Netted sensors-based vehicle acoustic classification at Tier 1 nodes

The MITRE Corporation has embarked on a three-year internally-funded research program in netted sensors with applications to border monitoring, situational awareness in support of combat identification, and urban warfare. The first-year effort emphasized a border monitoring application for dismounted personnel and vehicle surveillance. This paper will focus primarily on the Tier 1 acoustic-based vehicle classification component. We discuss the development and implementation of a robust linear-weighted classifier on a Mica2 Crossbow mote using feature extraction algorithms specifically developed by MITRE for mote-based processing applications. These include a block floating point Fast Fourier Transform (FFT) algorithm and an 8-band proportional bandwidth filter bank. Results of in-field testing are compared and contrasted with theoretically-derived performance bounds.

Simulation of human migration based on swarm theory

The analysis documented in this paper was completed as part of a larger multi-component simulation effort to evaluate the efficiency of sensor networks in detecting and identifying human subjects remotely. Such an assessment necessitates the accurate simulation of movement of people along roads, known paths or over a more open terrain. The individuals are assumed to be traveling in groups of varying size and usually with a designated leader. The mass motion toward a “goal” or away from an “obstacle” is modeled using swarm theory, which successfully applies basic physical principles to the simulation of emergent collective behavior of biological organisms. Individuals are modeled as particles whose velocities are updated using either dynamic or kinematic schemes. In the present analysis, swarming techniques were found to reproduce quite well the group movement and behavior of individuals over predefined surface conditions, as determined by driving forces, social interactions and environmental conditions.

An integrated method to detection, data association and tracking of multiple broadband signals

The present work explores a new method of integrated detection, localization, and tracking of multiple broadband signals directly from array data, without the requirement of distinct data association. The method is based on Maximum A-Posteriori probability concepts and combines Maximum Likelihood direction finding techniques with Kalman Filter theory. Implicit data association is given by a Nonlinear Programming scheme that simplifies the solution of a constrained optimization problem. Assuming Markov Motion and random Gaussian signals and noise, diverse kinematic scenarios for both synthetic and real data sets were investigated. Full data batch, semi-sequential and fully sequential variants were developed in element space, beamspace and windowed element space. The method was found to work well down to a signal-to-noise ratio of -10 dB, and for highly dynamic scenarios. An alternating projection method was used for contact state initialization and signal enumeration.