This PDF file contains the front matter associated with SPIE Proceedings Volume 11423, including the Title Page, Copyright information, and Table of Contents.

The sensor bias estimation problem is crucial in autonomous driving systems for perception and target tracking. This work considers the bias estimation for two collocated synchronized sensors with slowly varying, additive biases. The differences between the two sensors’ observations are used to eliminate the target state. Consequently, the bias estimation is independent from the target state estimation. The biases’ observability condition is met when the two sensors’ biases are Ornstein-Uhlenbeck stochastic processes with different time constants. A Maximum-Likelihood measurement fusion technique is introduced for the bias-compensated observations. Simulation results, for several scenarios with various bias model parameters, prove the consistency of the estimator. It is shown that the uncertainties of biases are significantly reduced by the estimation algorithm presented. The sensitivity of the proposed algorithm is also tested with mismatched filters.

This article focuses on Group Target Tracking (GTT) to counter swarms of drones using Random Finite Sets (RFSs) and Random Matrix (RM) approaches. Tracking swarms of drones is analog to tracking extended targets that are characterized by their continuously evolving shape and composition. Extended target tracking for groups of targets finds various applications in the literature because detecting and tracking each individual target of a group is computationally demanding and unnecessary if the group itself can be modeled. Elliptic shapes offers a suitable representation for most groups, and their inference is quite inexpensive with the random matrix approach. Indeed, they are efficient when coupled with random finite sets based filters, which represents the current state of the art for Bayesian multi-target tracking. In this work, a practical implementation of a labeled Poisson/Multi-Bernoulli filter using random matrices for group target tracking is proposed. This study compares several random matrix prediction and update algorithms with and without random finite sets based filters on several dataset.

This paper concerns measurement extraction for targets in electro-optical (EO) sensors. The goal is to be able to estimate the location and intensities of two targets in a focal plane array (FPA). Work has been done previously on extracting single targets, as well as two targets of equal intensity. The current work extends this to two targets with unequal and unknown intensities. We consider point targets that deposit energy in the FPA according to a Gaussian point spread function (PSF) with parameter σPSF . The measurement extraction for the targets is performed using a Maximum Likelihood method using two different models: a two target model for resolved targets, and a one target centroid model for unresolved targets. We present the Cramer-Rao Lower Bound (CRLB) on estimation accuracy for both models and provide simulations that confirm of our method’s efficiency (i.e., that this lower bound is achieved). Our results show that a target separation of about 1:15σPSF is needed for efficient extraction.

The problem of selecting a target of interest for interdiction in the presence of several spurious tracks that are meant to confuse the defense has been around for several decades. The spurious tracks are from the objects released from the target of interest and they move forward at the same speed as the target of interest. They separate due to a release velocity orthogonal to the forward motion. The main means of carrying out the discrimination between the target of interest and the spurious tracks discussed in the literature is using some features, which, however, are not always available. The present work considers this problem when the extraneous tracks “look” the same as the target of interest for the sensor tracking them, i.e., they have no distinguishing features. It is shown that the history of the track kinematics — the evolution of the tracks — can be used via “track segment association” to select the track of the target of interest from the several tracks in the field of view of the sensor. One of the challenges of this work is that, with limited resolution capability, the observations from the sensor are unresolved when the extraneous targets start separating. In this work, the data association and tracking are handled separately from track segment association, which reduces the complexity of the problem and is shown to have timely and reliable results in the simulations.

A Signal of Interest (SOI) is a signal the operator has decided to record for further analysis. This is driven by mission requirements, known anomaly characteristics, or unidentifiable signals. Currently on our radar detection system, identifying SOI or anomalies is reliant on the system operator’s knowledge and skill, a method highly susceptible to human error. The objective was to find a way to provide the system operator with improved awareness by automating identifying SOIs or anomalies with machine learning and artificial intelligence techniques. By applying data science processes and techniques such as density-based clustering algorithms and artificial neural networks to our data, we successfully proved the daily emitter and frequency distribution in the Hampton Roads area has a strong consistent subset of emitter traffic, identified anomalies based on this fingerprint, and implemented this algorithm in an application which provides a graphic that highlights anomalies and SOIs.

In this article, we explore the role and usefulness of neuro-fuzzy logic in the context of automatically reasoning under uncertainty about complex scenes in remotely sensed data. Specifically, we consider a first order Takagi- Sugeno-Kang (TSK) adaptive neuro-fuzzy inference system (ANFIS). First, we explore the idea of embedding an experts knowledge into ANFIS. Second, we explore the augmentation of this knowledge via optimization relative to training data. The aim is to explore the possibility of transferring then improving domain performance on tedious but important and challenging tasks. This route was selected, versus the popular modern thinking of learning a neural solution from scratch in an attempt to maintain interpretability and explainability of the resultant solution. An additional objective is to observe if the machine learns anything that can be returned to the human to improve their individual performance. To this end, we explore the task of detecting construction sites, an abstract concept that has a large amount of inner class variation. Our experiments show the usefulness of the proposed methodology and it sheds light onto future directions for neuro-fuzzy computing, both with respect to performance, but also with respect to glass box solutions.

Formation control of unmanned aerial vehicles (UAVs) has many applications including target tracking, surveillance, terrain mapping, precision agriculture, etc. Although many centralized control methods (single command center/computer controlling the UAVs) exist, there are no standard decentralized control frameworks in the literature. In this paper, we present a novel UAV swarm formation control approach based on a decision theoretic approach. Specifically, we pose the decentralized swarm motion control problem as a Decentralized Markov Decision Process (Dec-MDP). Here, the objective is to drive the swarm from an initial geographical region to another geographical region where the swarm must lie on a certain geometrical surface (e.g., surface of a sphere). As most decision theoretic formulations suffer from the curse of dimensionality, we adapt an approximate dynamic programming method called nominal belief-state optimization (NBO) to solve the formation control problem approximately. We perform simulation studies in MATLAB to validate the performance of the algorithms.

In this paper, we propose an efficient approach to improve moving object detection accuracy which is adaptive to changes in object size according to the altitude of the aircraft. The proposed algorithm can effectively detect moving objects with various pixel-scale from 8x8 to 100x100 in full-HD motion imagery. It consists of two stages which are detection and fusion. At the first stage, two algorithms are performed simultaneously: One is one-stage object detection network for detecting large objects, and the other is optical flow method for detecting small moving objects. In the second stage, results from the first stage are fused with a Ground Sample Distance (GSD) of imagery. We have conducted experiments using aerial imagery taken at a height between 130 meters and 400 meters. We evaluated the detection performance of our method in terms of precision, recall and normalized multiple object detection accuracy (N-MODA). Through experiments, we proved that the superiority of the proposed method.

Recent advances in the development of artificial intelligence and machine learning (AI/ML) techniques have shown great potential for enhancing the modeling and characterization of materials science issues. AI/ML techniques revolutionize big data analytics through signals and imaging detection, segmentation, and characterization; data fusion processes of association, estimation, and prediction, and modeling of deformation, structural, and materials awareness. Unfortunately, the dominance of AI/ML applications may hinder fundamental understanding of driving parameters in complicated material properties, including the impact of local chemistry and energy influences on nucleation, phase evolution, and bonding. As mathematical complexities are modeled using AI/ML approaches the interaction of driving mechanisms and underlying physics-based and first-principles understanding is often omitted in the final modeling of material behavior. However, microscopy and advanced characterization techniques may help clarify the underlying physics by providing critical validation for mathematical assumptions used in AI/ML models and determining model inter-parameter relationships. Sensing-based characterization is critical for operations using deep learning and/or clustered neural networks where complex interactions between microstructural features strongly impact each other, and driving mechanisms that control material response. The paper addresses several considerations for applying machine learning techniques to fundamental material problems, and the role for parameter validation through characterization. The future of AI/ML materials awareness includes procedural potential applications, advanced analytical tools, and coordinated research discovery thrusts.

The fusion of data obtained in different electromagnetic ranges is an important task for many areas of research. The combining data is necessary for security systems (when searching for people in difficult weather conditions (snow, fog, rain, dust), automated control systems (auto-driving, UAVs), etc. The process of data analysis involves identifying base features. It includes the search and selection of borders, salience maps (human attention cards), angles, analysis of color gradients, etc. Most often, the detected features highlighted in images recorded in one range do not coincide with data obtained in other ranges. This is due to the fact that different electromagnetic ranges operate with different physical characteristics of objects in frames. The paper presents an approach based on the search and analysis of the basic descriptive characteristics of objects and the search for their correspondences on images of the same object, recorded in different electromagnetic ranges. As such data, the directions of the gradients are revealed, the search for the boundaries and angles of objects, the selection of locally stationary regions, the search for the center of mass of objects, the identification of the middle lines of stationary regions with included structures. The search for features is carried out on the basis of data obtained at various scales. Simplification of images is carried out on the basis of an algorithm for analyzing stationary regions and replacing the current intensity with an average. On the set of test data obtained in the visible range, near and far-infrared range, depth maps, the applicability of the proposed approach is shown. As an example of the applicability of this approach, an example is shown of stitching a pair of images obtained in different electromagnetic ranges.

This work examines, from a theoretical perspective, novel aspects of processing the frequency-modulated continuous wave (FMCW) radar returns from a fast quasicylindrical target moving along a relatively flat trajectory. There are various realworld scenarios in which such objects (as threats) approach victim vehicles. Pinpointing, in time and space, when a flattrajectory quasicylindrical object has penetrated a safety perimeter around an intended victim vehicle is important for activating just-in-time or pre-positioned countermeasures. With radar-enabled countermeasures, the resolution of the radar and its tracking algorithm are determinative of the power and indiscriminateness that the countermeasure needs to defeat a fast-flying threat in the vicinity of the “sacred” perimeter. A smaller location error and smaller timing error enable lowerpower, more precise, and less dangerous countermeasures to be deployed. In this work we examine a method that can potentially reduce the uncertainty in the target’s observed location to less than ten (and closer to five) centimeters at critical instants. With its conventional tracking having established an incoming missile’s flight path, a small FMCW radar can switch into a “lying-in-wait” mode for “end-zone” observation, in which a pre-designed resonant discrete-time filter can be shifted back and forth along the frequency spectrum to determine the instant at which the flying target has breached the safety perimeter. This paper discusses one way in which this can be accomplished, examines the theoretical underpinnings of the method, and makes a preliminary assessment of the uncertainty that can be expected. Simulation results are reported.

We present a weak matching algorithm for interval graphs, to detect recurrent patterns in multimodal temporal data, with feature time series extracted by nonnegative tensor factorization (NTF). NTF enables latent feature extraction as well as uniform representation of multimodal observables. This work builds on our previous work introducing an interval graph representation framework for multi-sensor data. Salient data regions and their relationships are represented by temporal interval graphs, where observables are captured as time intervals (nodes), and temporally proximate nodes are related by edges. Comparing events is then posed as a subgraph matching problem. However, subgraph matching is notoriously difficult (NP-complete) with polynomial algorithms for only very restricted families of graphs. Even in these cases, perturbations to graph structure from missing or extra nodes and edges can lead to brittle matching results. Indeed, realworld sensing involves noisy environments where extraneous or missing observables interfere with event interval graph structures. To cope with these challenges, we propose a proxy representation of interval graphs via their shortest and longest paths and compare graphs by matching their path sets. We describe an attributed path matching scheme that is robust to inclusions and exclusions of nodes by adapting the longest common subsequence algorithm using dynamic programming for attributed path matching. We demonstrate the efficacy of interval graph analysis of tensor features on real-world multimodal sensor data where we investigate the detectability, similarity, and distinguishability of three sets of known events based on ground truth. We illustrate our results with match matrices and ROC curves.

During the 2019 SPIE DSS conference, panelists were invited to highlight the trends and use of artificial intelligence and machine learning (AI/ML) for information fusion. The common themes between the panelists include leveraging AI/ML coordinated with Information Fusion for: (1) knowledge reasoning, (2) model building, (3) object recognition and tracking, (4) multimodal learning, and (5) information processing. The opportunity for machine learning exists within all the fusion levels of the Data Fusion Information Group model.

We investigated the difference in performance on an implicit learning task between humans and machines in the auditory domain. Implicit learning is the process of ingesting information, such as patterns of everyday life, without being actively aware of doing so and without formal instruction. In pattern and anomaly detection, it is desirable to learn the patterns of everyday life in order to detect irregularities. In addition, we also considered how affect or emotion-like aspects interacts with this process. In our experiments, we created a synthetic pattern for both positive and negative sounds using a Markov grammar, which we then asked a machine-learning algorithm or humans to process. Results indicated that the generated pattern is a trivial task for even a simple RNN. For a similar but more complex task, humans performed significantly better under the condition of positive affect inducing sounds than they performed with negative sounds. Possibilities for the outcomes are discussed, along with other potential methods to compare human and machine implicit learning performance.

A recent report on Hostile Social Manipulation stated that the tools and techniques being employed in targeted social media campaigns, sophisticated forgeries, cyberbullying and harassment of individuals, and distribution of rumors and conspiracy theories pose a “potentially significant threat to U.S. and allied national interests”. This creates a challenge for an Intelligence, Surveillance and Reconnaissance (ISR) force comprised of platforms, sensors, networks and personnel that have been focused on overseas contingency operations for nearly 20 years. Recent defense strategy emphatically stated, “Inter-state strategic competition, not terrorism, is now the primary concern in U.S. national security.” Joint guidance on Operations in the Information Environment (OIE) emphasized the need for a better characterization and assessment of the informational as well as physical, and human aspects of the security environment in order to identify and leverage interdependencies between them. Carley and Beskow wrote, “Social cybersecurity is an emerging subdomain of national security that will affect all levels of future warfare, both conventional and unconventional, with strategic consequences”4 and characterized a set of 16 socio-cyber security forms of maneuver in the BEND model. This paper will introduce the BEND model, providing real-world examples of maneuvers, and highlight emerging multi-source analytics for discernment of threat.

This research focuses on parametric word recognition applications using the orthogonal signal decomposition method. Despite the popularity of the statistical approach in speech recognition, the parametric approach is very common for simple word applications due to its less computational costs. Higher recognition rates can be obtained by incorporating numerous elements in the feature vector, but this requires greater computational costs, which may not be suitable for applications that demand rapid decision making with a system on a budget. In this investigation, the feature vectors were constructed by means of singular values from the orthogonal signal decomposition method and their effectiveness was tested in realtime word recognition applications. When the theoretical foundation was established and confirmed, it was validated in a program using a sound card.

In this paper, we compare the performance of different phase modulated frequency hopping (FH) multiple-input multiple-output (MIMO) radar waveforms. The communication symbols are embedded into the FH waveforms under the auspices of dual-function (DF) radar and communications system. In the proposed scheme, each embedded symbol is represented by a sequence of phases. The phase modulated sequence is embedded through multiplication with the radar hops and is transmitted through a MIMO radar platform. Using this scheme, we consider information embedding in MIMO FH radar implementing different types of phase modulated sequence, including phase shift keying (PSK), differential phase shift keying (DPSK) and continuous phase modulation (CPM). We analyze the corresponding ambiguity functions (AFs) in terms of reducing or increasing range and Doppler sidelobes compared to the FH radar without any signal embedding. The spectral sidelobe levels and the complexity of the demodulator at the communication receiver are also examined. It is shown the proposed embedding scheme for DPSK and CPM minimizes the frequency leakage outside the radar signal bandwidth. To demonstrate the effectiveness of proposed scheme, we compare its performance with existing embedding schemes, all for FH radars. It is shown that our approach, in its significant reductions of range sidelobes, permits high data rate transmission. This is made possible because the FH radar can now afford using duplicate hopping code values without the penalty of incurring high correlations, or range sidelobe levels.

We present the system integration and validation tests of a compressive Multi-Mission Electro-Optical Sensor (MMEOS). With the unique algorithm implementation, the sensor exhibits exceptional agility enabling both multispectral (MS) sensing for wide area situational awareness and hyperspectral (HS) sensing for target recognition and identification. The sensor enables seamless mission changes on-the-fly via only software configuration of the operational parameters such as spatial, spectral and temporal resolutions based on mission requirements.

Signal attributes such as angle of arrival (AoA), time of arrival (ToA), signal amplitude, and phase can be used by a set of receivers (detectors) to perform location fingerprinting (LF), whereby the location of a wireless source is determined. In validating new approaches for location fingerprinting, it is useful to simulate these attributes for the subset of signals that intersect detectors. However, given indoor settings with a complex architecture, it is computationally expensive to simulate multipath propagation while preserving detailed signal information. Moreover, this cost can be unnecessary since determining whether an LF approach is promising may not require tracing all rays that impact the detector. Here, we report on our preliminary efforts to design and test a MATLAB-based simulation tool for wireless propagation that addresses this issue. Our approach builds upon well-known ray-tracing techniques, but innovates via an algorithm designed to obtain a sizable subset of rays that intersect a detector, along with the AoA, ToA, signal amplitude, and phase for each such ray. Finally, we employ our tool in conjunction with a neural network-based method for location fingerprinting, demonstrating the intended use case for our simulation tool.

The backscatter of a target illuminated by a stepped-frequency radar represents the frequency response of the target as seen by the radar and it can be modeled as a sum of exponentials. The inverse Fourier of the backscatter represents the target impulse response. A time-frequency analysis (using Wigner-Ville) of the target backscatter should yield features that are directly related to target scattering centers. These features can then be used for target recognition purposes. This concept is first tested on a synthetic radar target that consists of five scattering centers. Next, the frequency response of real backscatter of commercial aircraft models is examined using time-frequency analysis. The instantaneous frequency and the group delay as extracted using the pseudo Wigner-Ville distribution are used as target recognition features. The classification results are compared with those obtained using an optimal maximum-likelihood classifier. The performance of time-frequency extracted features as target recognition tools is investigated in all scenarios of complete or partial azimuth knowledge, and additive noise. The time-frequency analysis of the frequency response of the synthetic target is examined and compared with that expected using a closed-form solution.

When it entered into the era of big data, Earth observing systems developed into a new stage, namely characterized by low cost, multi-national, multi-sensor and multi-modal with varying spatial and spectral resolutions confronting new challenges and opportunities. Climate data records from multiple data sources are used to infer seasonal and interannual variations which will advance and promote the development of data fusion methods. Compressed sensing is a new framework in which data acquisition and data processing are merged. It provides a new fantastic way to handle multiple observations of the same field view from complementary remote sensing instruments, allowing us to recover information at very low signal-to-noise ratio. We will particularly point out that a Compressive Sensing based framework is flexible enough for combining the two measurement systems by fusing the data from the two satellites, NASA Orbiting Carbon Observatory -2 (OCO-2) and the JAXA Greenhouse gases from Orbiting Satellites (GOSAT) to calculate the interannual Net XCO2 variability over land for three latitudinal regions, Alaska/Canada, United States and the Amazon/Brazil. The OCO-2 design is optimized for sensitivity to XCO2 variations, with an unprecedented combination of spatial resolution (about 3km) with narrow nadir coverage, while GOSAT provides broader spatial coverage (10km) with wider scanning coverage. There are different temporal degradations of both instruments over time because GOSAT was launched in 2009 and OCO-2 was launched in 2014. Both instruments infer CO2 concentration from high-resolution measurements of reflected sunlight and use similar inversion algorithms to retrieve CO2 concentrations. Both are passive satellites providing on-orbit global measurements of the greenhouse gas, XCO2, for the years 2015 -2018. The results of the CS data fusion framework show that the fused data have Root Mean Square Error (RMSE) varying from 1.31 ppm to 4.12 ppm compared with original data, depending on the region of study and gridding resolution. Validation of fused data compared with AmeriFlux station towers observations shows RMSE of 2.68 ppm.

The initial development of two First-Person Perspective Video Activity Recognition Systems is discussed. The first system, the First Person Fall Detection or UFall, can be used to recognize when a person wearing or holding the mobile vision system has fallen. The problem of fall detection is tackled from the unique first-person perspective. The second system, the directed CrossWalk System (UCross), involves detection of the user movement across a crosswalk and is intended for use in helping a low vision person navigate. In both cases, the user is wearing or holding the camera device for purposes of monitoring or inspection of the environment. This first-person perspective yields unusual fall data and this is captured and used for the creation of a fall detection system. For both systems Machine Learning is employed using video input to trained Long-Term Short-Term (LSTM) Networks. These first-perspective video activity recognition systems use the Tensorflow framework [1] and is deployed using mobile phones for proof of concept. These applications could be useful for low vision people and in the case of fall detection for senior citizens, police, construction and other inspection-oriented jobs to help users who have fallen. The success and challenges faced with this unique first-person perspective data are presented along with future avenues of work.

High blood pressure has been one of the main causes for cardiovascular health problems like heart attacks, aneurysms, or even strokes. About 32% American adults, have high blood pressure and only about half (54%) of people have their condition under control [9]. The main objective of this project is to analyze, design, implement and test a blood pressure monitor which can transmit data in real time via radio frequencies. The paper includes all the analysis performed for each of the subsystems in the block diagram. Also, a diagram of each of the electronic circuits with the values obtained during the analysis. Results of the implementation and testing were included in the report.

Modern Prosthetics for the rehabilitation of hand amputees has been improved recently, but the value of prosthetics for the freedom of movement and adoption in the daily life of amputees is still substandard. The Electromyography (EMG) signals are generated by human muscle systems when there are any movements and muscular activities. These signals are detected over different areas from the skin surfaces and each movement corresponds to a specific activation pattern of several muscles. In this research, multi-channel EMG measurements were performed with electrodes placed on involved arm muscles. Since deltoid, bicep brachii, pectoris major, and flexor digitorium muscles can almost independently move human arm with adjustable contraction forces, the surface EMG signals from these muscles were utilized to recognize different arm movements. The EMG signals were digitally recorded and processed using digital filters, feature extraction methods, and classification algorithms. For feature extraction, the envelopes were extracted from the signal waveforms. To reflect the moving average activities, the root means squares (RMS) operation and normalization were successively utilized as initial signal processing method. Afterward, an activation vector containing normalized RMS signals was obtained in realtime. For machine learning, the activation vectors were utilized to train a real-time support vector machine (SVM) classifier to recognize different muscle EMG signals and their respective motion commands. A detailed analysis using SVM reveals more than 98% accuracy for recognition and successful classification of different motion commands after training. The effectiveness of the proposed method was verified through several experiments.

Pregnant women with conditions such as hypertension, diabetes, anemia, obesity, among others; have more possibility to be diagnosed with a high-risk pregnancy. Women with this type of pregnancy should visit their gynecologist up to two times a week depending on their condition, to monitor contractions and the fetal heartbeat, just to know the wellbeing of the fetus. Mothers have no other way to monitor the health and the contractions because the equipment is very limited and involves high costs. Constant Monitoring of the fetus could help to detect early symptoms and anomalies that can be a sign of premature childbirth and other fetal complications that could be of major concern. By creating a low-cost contraction monitor that measures the duration and frequency of contractions we could detect some symptoms or abnormalities that may indicate symptoms of early miscarriage or some other problem with the fetus. This device will save the data and would be able to alert the patient if an anomaly in the contractions occurs. When the abnormality is detected the doctor receives a text message with the information of the patient so he can give her a recommendation on what to do. The device will also save all the data so the doctor can analyze and determine the status of the fetus. The idea of this device is to help detect early symptoms of possible complications during pregnancy and so that both the mother and the fetus can enjoy a healthier pregnancy. The data recollected can also be useful to support investigations related to fetal conditions and abnormalities.

Neuromorphic computing has made tremendous advances in computing efficiency by modeling computers after the brain primarily by applying a spiking model of information. Spiking models inherently maximize efficiency in noisy environments by placing the energy of the signal in a minimal time. However, many neuromorphic computing models ignore time delay between nodes, choosing instead to approximate connections between neurons as instantaneous weighting. With this assumption, many complex time interactions of spiking neurons are lost. Here, we show that the coincidence detection property of a spiking-based feed-forward neural network enables mirror symmetry. Testing this algorithm exemplary on geospatial satellite image data sets reveals how symmetry density enables automated recognition of man-made structures over vegetation. We further demonstrate that the addition of noise improves feature detectability of an image through coincidence point generation.

It is commonly believed that having more white pixels in a color filter array (CFA) will help the demosaicing performance for images collected in low lighting conditions. We present a comparative study to evaluate the performance of demosaicing for images collected in realistic low lighting conditions using two CFAs: the standard Bayer pattern (aka CFA 1.0) and the Kodak CFA 2.0 (RGBW pattern with 50% white pixels). Using a data set containing 10 images collected in low lighting conditions, we observe that having more white pixels does help the demosaicing performance. However, some cautions are needed in quantifying the performance.

Comparisons of IR absorption spectra calculated using density functional theory (DFT) and measured IR transmission spectra for three nitrosamines are presented. These nitrosamines are major environmental contaminants. In general, identification of target molecules by cross correlation of spectra can be accomplished using signal templates having patterns associated with known materials. DFT calculated IR spectra can provide reasonable templates for filtering of IR spectral measurements associated with different types of detector schemes and complex spectral-signature backgrounds. The comparisons of spectra presented here demonstrate that DFT calculated IR spectra can be cross correlated with measured spectra to determine the presence of target molecules.

This paper is generally concerned with mathematical formalisms to support theory and algorithm developments of multiple hypothesis tracking (MHT), as a class of solutions to multiple target tracking (MTT) problems based on targetwise detections. In particular, this paper presents a new perspective on random set (RFSet) formalism to support a form of MHT, in which an unknown number of targets is modeled by a RFSet of continuous-time stochastic processes, rather than a single stochastic process defined on the space of finite sets in a given target state space, while generally multiple sensors provide noisy and cluttered target detections without any explicit indications of origins. The focus is on a clearcut approach to avoid any complication resulting from diagonal sets in direct-product spaces when a space of finite subsets of a state space is defined as its quotient space, instead of a subspace of the space of closed subsets in the state space with Fell-Matheron topology.

The solution to the problem of recognizing human actions on video sequences is one of the key areas on the path to the development and implementation of computer vision systems in various spheres of life. Such areas as video surveillance systems, monitoring, contactless control interfaces, video processing as a preliminary stage of processing, etc. Most of the approaches published in the literature can be divided into two groups: approaches based on constructing a global descriptor or a description of local points, unrelated to the human skeleton; and approaches based on the construction of the feature points (joint) of the human skeleton. In most cases, only the second group of methods use depth sensors to obtain clear information about the human skeleton. While additional sources of information (such as depth sensors, thermal sensors) allow you to get more informative features, and thus increase the reliability and stability of recognition. In the article, we present the algorithm, combining information from visible cameras and depth sensors based on the PLIP model (parameterized model of logarithmic image processing) close to the perception of the human visual system, and the development of a global descriptor characterizing the action taking place in the frame. The proposed algorithm takes advantage of the fusion of various modalities that provide the construction of a more informative descriptor. Experimental results showed the effectiveness of the proposed algorithm.