This PDF file contains the front matter associated with SPIE Proceedings Volume 11166 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

The study of using a cyclone collector with a liquid recirculation phase for collecting microparticles of TNT (trinitrotoluene) explosive, located in the form of traces on solid and fibrous surfaces was conducted in this work. Fragments of fibrous nonwoven material, as well as flax fabric fragments were used as a basis on which TNT powder was randomly distributed. The TNT sample area was about 900 cm² . The cyclone collector with recirculating liquid phase originally designed for aerosol particles collection was used as a device for pre-concentration of explosives particles. Inlet air velocity was equal to 500 cm³ /min. Quantitative measurements of collected TNT amount were provided by gas chromatography – mass-spectrometry method. Experiments to compare the efficiency of cyclone collector sampling to the currently used methods of explosives microparticles collection - by wiping of surfaces and by dry filtration - were carried out. Cyclone collector showed high collection efficiency, amounting to 91% by weight of originally TNT spread on the surface that is more than 3 times higher than efficiencies of the wiping method (24% of the total mass of the trace) and the dry filtration method (23% of the total mass of the trace).

The features of laser ionization of trinitrotoluene and hexogen at registration of ionization products by field asymmetric ion mobility spectrometer were investigated in this paper. The optimal range of intensity to obtain the maximum ion current was determined, and it was shown that composition of ions of explosives is changed with the change of intensity. The values of intensity and radiation at which it is possible to obtain spectra of nonlinear ion mobility with different composition of RDX and TNT ions were determined, which is promising for increasing the selectivity of registration. It was shown that ion peaks of TNT and RDX at laser ionization are characterized by low nonlinear mobility coefficients in comparison with a radioactive source based on Ni⁶³, which indicates a more selective mechanism of laser ionization of these explosives at atmospheric pressure.

It is known that luminescent optical sensors are perspective for detection towards nitroaromatic compounds that are a basis of a many kind of explosives. Operation of these sensors is based on quenching luminescence, which is caused by photo-induced electron transfer from a luminophore (donor) to a nitroaromatic molecule (acceptor). The conjugated polymers, small molecule dyes and metal-organic frameworks are used as a sensitive luminophores currently. One of the methods to improve these sensors is embedding the luminophore into porous matrix with properties of photonic crystal, which may be a porous silicon (pSi) Bragg mirror or a microcavity (MC). The PPV derivatives polymers are usually used as the sensitive luminophores for embedding into pSi matrix. However, there is a task to find an optimal set of luminophores to develop a highly sensitive and selective sensor. In this work we investigate embedding of 5- triphenylamino-4-(triphenylaminothiophen-2-yl)-pyrimidine (HEM-461) into pSi MC and examine the sensitivity of obtained structures. The pSi MC were fabricated using a standard electrochemical etching process. The eigenmode of the pSi MC had a width of 4-6 nm. The samples were oxidized to stabilize the surface chemical properties and to prevent quenching of luminescence of the embedded luminophores after fabrication. The embedding of the dye into the pSi MC was performed at excess pressure. Well known conjugated polymer MDMO-PPV was used as a reference. In this work, we compared the photophysical properties of MDMO-PPV and HEM-461 in solution and into pSi MC. The luminescence parameters and resistance to heat have been studied. Comparative studies of sensitivity of MDMO-PPV and HEM-461 to trinitrotoluene in liquid and gaseous phases have been carried out. It was concluded that pSi MC with embedded HEM-461 is a promising structure for developing sensors of nitroaromatic compounds.

Testing the environment for bio-aerosols is an important feature of biosafety in the modern world. It is often necessary to collect aerosols from large areas in a short time, which requires outstanding collection efficiency, sufficiently high flow rate of incoming air and ability to maintain the viability of the collected samples. The paper presents the results of creating an effective sampling device capable of operating at airflow rates of 4000 liters per minute. The device consists of two functional parts - a virtual impactor and a cyclone collector with liquid phase of deposition. We present all the necessary calculations and algorithm to simulate parameters of the impactor. The sampling device was tested using dry and liquid dispersed particles with a diameter of 0.5 to 5 μm. We demonstrated that at a flow rate of about 4000 l / min, the efficiency of collecting of particles is more than 20% of the total aerosol mass, and at a flow rate of more than 300 l / min, this value exceeds 60 %. The proposed device supports the viability of the collected microorganisms. The paper also presents the results of testing the device at infrastructure objects. The device is portable, with easy settings for sampling and cleaning, and can be controlled remotely over a network

Polarimetric and Multi-spectro-polarimetric imaging systems have a high potential for different applications such as biomedical, surface evaluation and roughness, remote sensing and imaging through turbid mediums. The liquid crystal variable retarders (LCVRs) are propitious candidates to combine a polarization control devices in polarimetric imaging systems due to their exceptional properties such as retardation tunability using low voltages and minimal power consumption, in addition to being cost-effective and compact. In this work, we present a new broadband ellipso-polarimetric camera[1] based on tunable achromatic waveplate (AWP)[2,3] consisting of LCVRs. This AWP can be tuned between three modes: quarter, half and full waveplate in wide spectral band. Also, the operation band can be tuned with the three modes. By capturing images at the three modes of the AWP, the system can extract various EP parameters: two Stokes parameters (S1 and S3), sin(2ψ) and cos(Δ). Particularly we show that cos(Δ) images are superior over many known polarimetric imaging quantities. The performance of the imaging system was demonstrated in different applications such as tissue for biomedical imaging, fingerprints and its residuals on the surface, old coins and remote sensing. In addition to the spatial and polarimetric information, integrating a narrowband LC tunable filter[4] into imaging systems provides the ability to extract the spectral information which helps to characterize the object better. In this work, we shall present the performance of Multispectral camera and Multi-spectro-polarimetric camera based on our developed LC devices for different applications. References: 1. M. J. Abuleil and I. Abdulhalim, "Broadband ellipso-polarimetric camera utilizing tunable liquid crystal achromatic waveplate with improved field of view," Opt. Express (2019), accepted. 2. I. Abdulhalim and M. J. Abuleil, "Tunable achromatic liquid crystal waveplates," U.S. patent 10,146,095 B2. 3. M. J. Abuleil and I. Abdulhalim, "Tunable achromatic liquid crystal waveplates," Opt. Lett. 39(19), 5487 (2014). 4. M. Abuleil and I. Abdulhalim, "Narrowband multispectral liquid crystal tunable filter," Opt. Lett. 41(9), 1957 (2016).

The purpose of the research is to develop a methodology for testing the camouflage fabrics using hyperspectral data. The fabrics used in the research have been overexploited and have been subjected to a variety of chemicals. The research is aimed at answering how and how often the masking fabrics should be examined, what processes eliminate the fabric from their camouflage functions and at what time the regular use disqualifies a given fabric from further use. The research was conducted in the laboratory conditions. For the study, VNIR hyperspectral camera was used, produced by Headwall. In the presented study, selected camouflage fabric samples were used. They were exposed to long-lasting effects of various atmospheric conditions, such as sunrays, rain, snow, etc. The time of exposure of textiles and the type of conditions was monitored. In addition, samples were used that were subjected to the action of various chemicals, e.g. chlorine. Chemicals applied in the research are widely used for cleaning fabrics. The influence of chemicals other than those commonly used in textile maintenance has not been verified. Based on the hyperspectral imagery data, the spectral reflectance characteristics of processed samples were retrieved. Next, developed spectral indices were applied in order to determine the degree of ageing. Additionally, hyperspectral images of used material samples have been subjected to a classification process using spectral reflectance characteristics of brand new materials. This allowed to determine the divergence between the samples and determining the degree of destruction of samples, depending on the ageing processes used. Trends were set, and the conditions and activities were eliminated from the use of fabrics. Finally, based on the obtained results, methodologies for testing the camouflage fabrics and determining their usefulness were developed.

Border management and security challenges are increasing considerably in recent years. One of the major concerns is counterfeiting and fraudulent use of identity and other travel documents for crossing border controls. This poses serious threats and safety concerns worldwide, considering the scenario of terrorism and illegal migration across the world. Hence, advanced technologies with improved security features becomes essential to strengthen border security and to enable smooth transits. In this paper, we present a novel dual waveguide based invisible fluorescence security feature with lifetime discrimination and a simple validation system. Molecular fluorescence and lifetimes from the rare earth doped waveguides can be used as additional security features in the identity documents. The validation system consists of a modulated excitation source and fast photo-diodes which helps in the simultaneous detection of multiple security features from the fluorescence waveguides. The rare earth doped fluorescence waveguides are embedded into the identity document as micro-threads or tags which are invisible to the naked eye and are only machine readable. Rare earth fluorescence materials have higher sensitivity and selectivity as they absorb only specific ultraviolet (UV) or visible (VIS) wavelengths to create corresponding fluorescent emissions in the visible or infrared wavelengths. Herein, we present the results based on the fluorescence and fluorescence lifetime spectroscopic studies carried out on the terbium (Tb) and dysprosium (Dy) doped waveguides. The different emission wavelengths and lifetimes of these rare earth elements is a key differentiating feature, providing selectivity and security to the detector systems.

Identification of chemical species using eye-safe laser pulses is becoming of wide interest for both biology and security purposes. Many instruments are designed to this purpose exploiting the high selectivity and sensitivity of the Raman spectroscopy. A laser pulse is sent to a target, and its Raman echo (usually from the Stokes band) is collected by some optics, dispersed and analyzed by a detector. Being the Raman cross sections usually very small, when compliance with the regulations about exposure to laser radiations is requested, each step of the acquisition chain must be optimized in order to lose less photons as possible from the target to the detector. In this work we will discuss some of these aspects with a main focus on the maximization of the laser dose and the detector signal-to-noise.

The work is devoted to the influence of scattering of terahertz (THz) radiation by hexogen particles (RDX) in powdery samples on their transmission and reflection spectra. A terahertz radio-vision installation with spectral resolution was used to determine experimentally THz spectra of RDX. For samples with small RDX particles (the typical particle size is 100 μm), characteristic peaks at 0.8 THz and 1.06 THz are observed in absorption spectra despite scattering, that can be used to identify this substance. For large hexogen particles (a typical particle size is 450 μm), experiments and numerical simulation showed that even the most intense peak at 0.8 THz is not observed in absorption spectra, and the spectra are mainly due to the scattering effect and its depending on the wavelength of radiation. The reflection spectra of RDX layers (particle size is about 100 μm) qualitatively differ from the reflection spectra of RDX crystals and are formed as a result of absorption during propagation of THz radiation in the particle layer. Thus, the substance can be identified by absorption spectra in a reflection scheme.

Considerable progress with unmanned aerial vehicles (UAVs) has led to an increasing need for counter-UAV systems to detect present, potentially threatening or misused drones. Therefore, a UAV detection algorithm has been developped recently for day and night operation. Whereas high resolution VIS cameras enable to detect UAVs in daylight in further distances, surveillance at night is performed with a short wave infrared (SWIR) camera. The proposed method is based on temporal median computation, structural adaptive image differencing, codebook-based background learning, local density computation, and shape analysis of foreground structures to perform an improved near range change detection for UAVs. Areas with moving scene parts, like leaves in the wind or driving cars on a street, are recognized to minimize false alarms. This paper presents a significant improvement with respect to some of the most challenging tasks in this field, e.g., increasing the UAV detection sensitivity in front of trees with waving leaves, false alarm minimization, and avoiding the background model update problem. The provided results illustrate the reached performance in a variety of different situations.

Intelligent robotic autonomous systems (unmanned aerial/ground/surface/underwater vehicles) are attractive for military application to relieve humans from tedious or dangerous tasks. These systems require awareness of the environment and their own performance to reach a mission goal. This awareness enables them to adapt their operations to handle unexpected changes in the environment and uncertainty in assessments. Components of the autonomous system cannot rely on perfect awareness or actuator execution, and mistakes of one component can affect the entire system. To obtain a robust system, a system-wide approach is needed and a realistic model of all aspects of the system and its environment. In this paper, we present our study on the design and development of a fully functional autonomous system, consisting of sensors, observation processing and behavior analysis, information database, knowledge base, communication, planning processes, and actuators. The system behaves as a teammate of a human operator and can perform tasks independently with minimal interaction. The system keeps the human informed about relevant developments that may require human assistance, and the human can always redirect the system with high-level instructions. The communication behavior is implemented as a Social AI Layer (SAIL). The autonomous system was tested in a simulation environment to support rapid prototyping and evaluation. The simulation is based on the Robotic Operating System (ROS) with fully modelled sensors and actuators and the 3D graphics-enabled physics simulation software Gazebo. In this simulation, various flying and driving autonomous systems can execute their tasks in a realistic 3D environment with scripted or user-controlled threats. The results show the performance of autonomous operation as well as interaction with humans.

The Swedish National Forensic Centre, NFC, has been developing methods for 3D modeling of crime scenes since 2016. Documentation in 3D opens up new possibilities for visualization, documentation and forensic analysis and live streaming 360 cameras might open new possibilities for telepresence in a wide variety of situations. In 2017 NFC, Voysys and Linköping University were granted funding from Visual Sweden to develop a prototype system and explore how telepresence could be used for witness walkthroughs. The project was successful and a fully functional system was developed and tested with forensic technicians, investigators and persecutors. The participants all agreed that the system delivered a far more immersive experience than videos and images, and also compared to live video such as Skype or Facetime. We believe that immersive telepresence can play an important role within law enforcement in the future but that further development and more studies are needed.

Nowadays, there is wide spectrum of areas of use of small unmanned aerial vehicles (UAV, drones). Their low cost, flexibility of use, low service cost and independence on infrastructure make them ideal carriers of different cargo. This fact brings many significant safety risk and threats. Especially aerial survey, flight zones disruptions and transport dangerous cargo. Above mentioned threat can be eliminated by: disabling navigation and control features, taking control over the steering and object destruction. One of the possible typically military solutions of that situation, is destruction by shooting. The most common way of solving the above task, is the use of barrel weapons of individual shooter or small military units. Therefore, we focused our effort on preventing the UAV mission accomplishing by its shooting destruction. To accomplish the task, in addition to detection and identification, we consider the distance determination as a fundamental problem. For the purposes of this paper, the distance determination we consider as a “localization”. Comparison of UAV localization options is the issue of this paper. Two localizations methods have been defined. Localization option include active method (ILR – impulse laser rangefinder) and passive method (triangulation method). Using mathematical modelling we proved, that ILR measurement is possible. To determine a distance with a defined probability we designed specific measurement frequency. For a predetermined type of UAV and ILR we identified the maximum distance that can be measured. In the second part of this paper, we analyzed the accuracy of localization using the triangulation method. The accuracy of localization using this method is dependent on the squared of the UAV distance. There are two significant advantage of this method: measurements can be made continuously and multiple objects in the field of view can be measured at the same time. This article is part of a planned publishing activity dealing with the issue

We present a study of border surveillance systems for automatic threat estimation. The surveillance systems should allow border control operators to be triggered in time so that adequate responses are possible. Examples of threats are smuggling, possibly by using small vessels, cars or drones, and threats caused by unwanted persons (e.g. terrorists) crossing the border. These threats are revealed by indicators which are often not exact and evidence for these indicators incorporates significant amounts of uncertainty. This study is linked to the European Horizon 2020 project ALFA, which focuses on the detection and threat evaluation of low flying objects near the strait of Gibraltar. Several methods are discussed to fuse the indicators while taking the uncertainty into account, including Fuzzy Reasoning, Bayesian Reasoning, and Dempster-Shafer Theory. In particular the Dempster-Shafer Theory is elaborated since this approach incorporates evaluation of unknown information next to uncertainty. The method is based on belief functions representing the indicators. These functions show a gradual increase or decrease of the suspiciousness depending on input parameters such as object speed, size etc. The fusion methods give two output values for each track: a suspect probability and an uncertainty value. The complete dynamic risk assessment of detected flying objects is evaluated by the automatic system and targets with probabilities exceeding a certain threshold and appropriate uncertainty values are presented to the border control operators.

In the past years, small unmanned aerial vehicles have increasingly become a hard to defend against threat to both military and civilian infrastructure. Both DIY and COTS UAVs are difficult to detect in a realistic environment, particularly in a cluttered and noisy one such as a port facility. In this context the use of active detection systems such as radar, and lidar is limited since it should not adversely interfere with the normal operation of the port, in particular the existing harbour sensors. This leads our interest towards passive multi-sensor detection systems such as electro-optic (EO) and acoustic monitoring. This work investigates the capability of passive acoustic systems to detect small commercial UAVs within the context of a harbour. We use a machine learning approach to detection using real-world data. We collected audio signatures of several different types of commercial off-the-shelf UAVs both in a quiet environment and in a variety of complex real environment. For this we used a directional 4-microphone array composed of readily available audio components. This setup limited our experiment to the audible spectrum, in which motor and propeller noise are the main characteristics used to distinguish the UAV from the background sounds. We studied machine learning algorithms typically applied to this category of problems, and implemented a Gaussian Mixture Model (GMM) classifier using the Mel-Frequency Cepstrum Coefficient (MFCC) as a feature representation of the audio data, and apply this to the data collected during our measurement campaigns.

Today, unmanned aerial systems (UAS) are being used for a growing range of useful applications. This has been drastically increased during the last few years. The market is growing and a wide range of affordable and capable products is available. However, also the misuse of such systems is growing and has become a big problem for public security, private areas and industrial facilities. Therefore, products for detection, classification and countermeasures are an upcoming market though a complete solution is not yet present. At the Fraunhofer IOSB, a research project has been set up in 2017 with its aim to develop an interactive detection and analyzing tool that helps the security staff in situation analysis and decision making during the handling of threatening situations caused by the misuse of drones. Beginning with different measuring devices and data fusion of these channels in order to generate a situational picture the systems handle a comprehensive risk management while calculating the chances of countermeasures. The user is supported by situational awareness tools and simulation/prognosis of possible scenarios during the complete process. The objective is the optimal cognition for the users and to assist during the solution finding. There is no automatic decision and no automatic reaction without humans in the decision loop.

We present a new approach to an optical UAS detection system that confirms several requirements specified by the authorities. Our UAS detection system consists of a ground unit and a folding mirror located in the air. The ground based unit contains high resolution camera system mounted on a pan tilt unit. Therefore, the lens of the camera system can be actively aligned to a convex mirror which is located in vertical distance over the ground unit on the lower side of a raised captive balloon. The focal length of the ground based lens, the altitude of the balloon and the curvature of the mirror define the field of view towards the ground. The recorded video streams are processed in a vision system using change detection and other algorithms to alert a UAS intrusion. We will outline the design parameters of the optical system, the requirements and the implementation of the mechanical system as well as the active alignment system and show first results of outdoor operation

We address the problem of controlling EO/IR sensors carried by UAVs, in a context where multiple UAVs are employed to perform a surveillance mission. Swarm-based ISTAR can deliver commanders a more complete situational picture, but controlling a swarm becomes increasingly complex with each additional vehicle. A practical system requires a degree of autonomy to avoid overwhelming operators with detail and/or to avoid requiring too many operators. We present a sensor tasking algorithm which chooses a sequence of tasks for an EO/IR sensor located on an air vehicle. The tasking problem is formulated as a Markov Decision Process (MDP), in which each task is associated with an expected reward stream that would be accrued if the task were put into effect. The size of the MDP state space increases exponentially with the number of candidate tasks, so generic MDP solutions are inapplicable. Instead we propose a solution using Gittins indices. Our problem includes time sensitive tasks, where the probability of receiving a reward decays with system time. The problem also includes tasks that are not time sensitive. The combination of time-sensitive and non-time-sensitive tasks is challenging for an index policy. Our contribution is to create a combination of index algorithms that approximately solves the MDP for a combination of time-sensitive and non-time-sensitive tasks. EO/IR stands for Electro-Optic / Infra-Red. UAV stands for Unmanned Aerial Vehicle. ISTAR stands for Intelligence, Surveillance, Target Acquisition and Reconnaissance.

This paper describes a multimodal vision sensor that integrates three types of cameras, including a stereo camera, a polarization camera and a panoramic camera. Each sensor provides a specific dimension of information: the stereo camera measures depth per pixel, the polarization obtains the degree of polarization, and the panoramic camera captures a 360° landscape. Data fusion and advanced environment perception could be built upon the combination of sensors. Designed especially for autonomous driving, this vision sensor is shipped with a robust semantic segmentation network. In addition, we demonstrate how cross-modal enhancement could be achieved by registering the color image and the polarization image. An example of water hazard detection is given. To prove the multimodal vision sensor’s compatibility with different devices, a brief runtime performance analysis is carried out.

Image enhancement of pre-processed and colored fingerprints such as ninhydrin-processed fingerprints in a complexpatterned background is important in crime investigation. Contrast adjustment, which is one of the most common image processing methods usually has a limitation in enhancing such fingerprints because of interference superposing patterns of the background. We propose three image processing methods using color information or spatial frequency information of an RGB material image. The first method is a hue-based method, which converts hue values of each pixel in CIELAB color space to corresponding brightness. The second method is a PCA-based method, which conducts principal component analysis on a color data of the material image and reconstructs three principal component images. Both proposed methods achieve more enhanced fingerprint images than the contrast adjustment gives. The PCA-based method works well even when the hue-based method does not. As the third proposed method, remaining background periodic patterns are removed for fingerprint enhancement by spatial frequency filtering using the Fast Fourier Transformation. According to estimated frequency components of background periodic patterns, we altered the width of frequency removing region of such background patterns, suggesting that there is an optimal width for each material for fingerprint enhancement. Also, we tested four different edge profiles of the frequency removing region and checked that an edge profile with gradual change tends to reduce the effect of suppressing the background periodic patterns compared to an edge profile with sharp change under the equivalent width of removing region of frequency components.

In intelligent video surveillance, cameras record image sequences during day and night. Commonly, this demands different sensors. To achieve a better performance it is not unusual to combine them. We focus on the case that a long-wave infrared camera records continuously and in addition to this, another camera records in the visible spectral range during daytime and an intelligent algorithm supervises the picked up imagery. More accurate, our task is multispectral CNN-based object detection. At first glance, images originating from the visible spectral range differ between thermal infrared ones in the presence of color and distinct texture information on the one hand and in not containing information about thermal radiation that emits from objects on the other hand. Although color can provide valuable information for classification tasks, effects such as varying illumination and specialties of different sensors still represent significant problems. Anyway, obtaining sufficient and practical thermal infrared datasets for training a deep neural network poses still a challenge. That is the reason why training with the help of data from the visible spectral range could be advantageous, particularly if the data, which has to be evaluated contains both visible and infrared data. However, there is no clear evidence of how strongly variations in thermal radiation, shape, or color information influence classification accuracy. To gain deeper insight into how Convolutional Neural Networks make decisions and what they learn from different sensor input data, we investigate the suitability and robustness of different augmentation techniques. We use the publicly available large-scale multispectral ThermalWorld dataset consisting of images in the long-wave infrared and visible spectral range showing persons, vehicles, buildings, and pets and train for image classification a Convolutional Neural Network. The training data will be augmented with several modifications based on their different properties to find out which ones cause which impact and lead to the best classification performance.

Deep learning is a widely utilized approach specifically for computer vision applications. Visual recognition is one of the applications utilizing deep learning. Several challenges limit the performance of visual recognition methods. One of the most important challenges is the insufficient number of labeled data in the datasets. To overcome this challenge, the recent studies propose sophisticated methods which require high computational resources, which may create another problem. That is, the implementation of such algorithms on mobile devices is quite challenging. Especially, these issues are encountered in surveillance systems that utilize the drones and/or CC-TVs. To solve these problems and obtain high accuracy, the network should be able to extract both representative and discriminative features from such a small amount of data. In this paper, we propose a generative adversarial semi-supervised training method for visual recognition. Experiments are performed to evaluate a lightweight deep convolutional neural network as a classifier network that is trained by the proposed method and a conditional/unconditional generator networks that are examined in adversarial training.

Over the past years, tracking in the visible domain has seen rapid growth by exploiting Deep Neural Network (DNN) based methods, whereas, tracking in the Thermal Infrared (TIR) domain has seen a small interest. In this comparative study, we address tracking in a TIR maritime context for surveillance applications. Towards this end, we first compare the performances of traditional Single Object Trackers (SOTs) and recent DNN-based SOTs on a TIR maritime data set. Following this, we examine the sequences of the TIR data set causing difficulties for trackers and identify problematic attributes. Firstly, We use a group constituted of recent state-of-the-art DNN-based trackers and another group constituted of traditional trackers not employing DNN-based methods, and measure performance using the following metrics: Intersection over Union (IoU), center error, success rate, and robustness. Furthermore, we rank the trackers by taking into account their scores on IoU and robustness. The presented study shows that recent trackers exploiting DNNs methods for tracking perform on average better: over 24% on IoU and over 14% on robustness than their counterparts not utilizing DNN in their tracking process. Moreover, despite the provided improvement by using DNN-based trackers, a failure case analysis shows that clutter, occlusion handling, low-resolution and scale change of the target, are visual attributes that still remain challenging, requiring further improvement.

To perform a data protection concept for our mobile sensor platform (MODISSA), we designed and implemented an anonymization pipeline. This pipeline contains plugins for reading, modifying, and writing different image formats, as well as methods to detect the regions that should be anonymized. This includes a method to determine head positions and an object detector for the license plates, both based on state of the art deep learning methods. These methods are applied for all image sensors on the platform, no matter if they are panoramic RGB, thermal IR, or grayscale cameras. In this paper we focus on the whole face anonymization process. We determine the face region to anonymize on the basis of body pose estimates from OpenPose what proved to lead to robust results. Our anonymization pipeline achieves nearly human performance, with almost no human resources spent. However, to gain perfect anonymization a quick additional human interactive postprocessing step can be performed. We evaluated our pipeline quantitatively and qualitatively on urban example data recorded with MODISSA.

Images or videos recorded in public areas may contain personal data such as license plates. According to German law, one is not allowed to save the data without permission of the affected people or an immediate anonymization of personal information in the recordings. As asking for and obtaining permission is practically impossible for one thing and then again, manual anonymization time consuming, an automated license plate detection and localization system is developed. For the implementation, a two-stage neural net approach is chosen that hierarchically combines a YOLOv3 model for vehicle detection and another YOLOv3 model for license plate detection. The model is trained using a specifically composed dataset that includes synthesized images, the usage of low-quality or non-annotated datasets as well as data augmentation methods. The license plate detection system is quantitatively and qualitatively evaluated, yielding an average precision (AP) of 98.73% for an intersection over union threshold of 0.3 on the openALPR dataset and showing an outstanding robustness even for rotated, small scaled or partly covered license plates.

In recent years advances in machine learning methods such as deep learning has led to significant improvements in our ability to track people and vehicles, and to recognise specific individuals. Such technology has enormous potential to enhance the performance of image-based security systems. However, wide-spread use of such technology has important legal and ethical implications, not least for individuals right to privacy. In this paper, we describe a technological approach to balance the two competing goals of system efficacy and privacy. We describe a methodology for constructing a “goal-function” that reflects the operators preferences for detection performance and anonymity. This goal function is combined with an image-processing system that provides tracking and threat assessment functionality and a decision-making framework that assesses the potential value gained by providing the operator with de-anonymized images. The framework provides a probabilistic approach combining user preferences, world state model, possible user actions and threat mitigation effectiveness, and suggests the user action with the largest estimated utility. We show results of operating the system in a perimeter-protection scenario

Machine learning has recently made great progress in object classification, and especially object classification in images. In some cases, machine learning has even shown better ability than the human. There is a great potential to partially, or fully, automate sensor data analysis in surveillance systems. The automation could greatly facilitate the human operator’s work in finding critical information. After the success of object classification in images, the next step is to try to achieve similar progress in behaviour recognition using similar approaches. The paper includes a brief state-of-the-art for automatic behaviour recognition, and especially behaviours related to surveillance. The focus is on approaches based on deep learning. The paper presents an experiment with a specific deep learning algorithm, namely the long short-term memory (LSTM). For behaviours in surveillance applications, actual training data are rare. One approach for overcoming this is to use a deep learning algorithm that is pre-trained for a nearby application, and then transferred to the current application. Another is to expand the amount of training data using simulations. A difficulty with simulations is to create data that have similar characteristics as real sensor data, that is, include relevant noise and uncertainties. The paper briefly discusses the case of simulated data for the development of a deep learning method.

Suspicious human behaviors can be defined by the user, and in long distance imaging it may include bending the body during walking or crawling, in contrast to regular walking for instance. State-of-the-art methods using convolutional neural networks (CNNs) dealt in general with “clean” signals, in which the object of interest is relatively close to the camera, and therefore fairly clear and easily distinguished from the surrounding environment. This makes it easier to capture detailed information regarding the object and its action. However, in relatively long distance imaging (few kilometers and above) additional difficulties occur which affect the performances of these tasks, since the captured videos are likely to be degraded by the atmospheric path that cause blur and spatiotemporal-varying distortions. Both of these degradation types may reduce the ability for action recognition. These effects become more significant for longer imaging distances and smaller sizes of the objects of interest in the image. The images of objects in imaging through long distance are usually relatively small, and hence, the range of actions that can be resolved is more limited, particularly under strong atmospheric effects. In this study, we perform action localization by first applying optical flow unique processing, and also using a variant of SSD (Single Shot MultiBox Detector) to regress and classify detection boxes in each video frame potentially containing an action of interest.

Recently a high demand for fast, cheap and easy to use selective and sensitive methods for trace detection of explosives developed. Beside other techniques like sensitive mass detections SERS (Surface-enhanced Raman spectroscopy) is a very powerful tool to enhance signals in the order of 10⁴ - 10¹⁰ for Raman-active analytic molecules. SERS based on the interaction between surface plasmon of metal nanoparticles (10 – 300 nm) like gold, silver, aluminum and gold/palladium alloy with traces of contaminants like explosives. Here we present our results for different explosives on different available and self-made SERS-substrates with varying parameters, like laser excitation (266 nm, 532 nm, 633 nm, 785 nm and 1064 nm), laser power, spot size and local energetic efficiency to acquire the best practice for an easy, effective and safe treatment of explosive traces.

Detection systems for explosives have to be tested and validated with realistic samples. However, at present, well defined and characterized and homogeneous test materials for the development and validation of trace detection systems are still missing. Drop-on-demand and inkjet printing technology has become a promising method of producing standardized chemically contaminated test materials with high precision, accuracy, scalability, and flexibility to allow for the inexpensive, high throughput production. Especially for the preparation of samples for optical stand-off technologies the drop-on-demand technique leads to samples of higher quality. Test samples with contaminations ranging from nanograms to micrograms have been prepared and analysed over several weeks of storage. The influence of plot parameters on the morphology and durability of printed samples of various common explosives have been investigated on different substrates. Our results show that this technique leads to samples of higher quality compared to already existing preparation methods of reference samples. For the evaluation of deposition uniformity, quantity and reproducibility samples on several different surfaces, with different solvents and areal density distributions were prepared and analysed. HPLC measurements were used to quantitatively evaluate the durability of printed samples. The morphology of the test samples has been characterized by optical microscopy and confocal Raman-microscopy. This poster focuses on the development of a proposed standardized methodology for the characterization of particulate trace explosive samples of all types. Micrographs were processed with ImageJ (National Institutes of Health), an open source image processing program designed for scientific multidimensional images, which has already been applied to the shape analysis of particles in various research areas and for different materials. The main problem is to find the right threshold to apply to segment the image into features of interest (particles) and background. It is certainly a goal of digital image processing to establish protocols that eliminate variability, but currently it is not possible to use only one threshold with a fixed value for a set of images of samples of various explosives. Manual methods have several limitations like reduced reproducibility, high user bias, tedious and time-consuming fiddling-around finding an "appropriate" cut-off value, etc. The open source software that is in use to process the images provides several methods to automatically apply a threshold. Based on the morphology of the particles and the applied threshold, the obtained results can be really different and, an incorrect selection of the automatic threshold could provide either an overestimation or an underestimation of the parameters to be evaluated, particle size distribution and area coverage.

For Lidar technology that can identify the location of the target substances and measure spatial distribution, the establishment of that technology is required so that it can comprehensively provide remote measuring hazardous substances that cause harm to human bodies, such as toxic substances and combustible substances. Hazardous substances exist in a very wide range of forms, for example, chemical species, physical conditions, and organisms or inorganisms. In addition, substances developed for the purpose of attacking the human body, represented by nerve agents, exhibit their effects by a small amount. Therefore, in order to realize remote sensing of hazardous substances, it is necessary to apply an excellent measurement principle that can respond to the diversity and the detection of trace components of these objects. The Raman effect is a useful phenomenon that enables identification of many individual substances, but the extremely weak response has led to significant limitations in applicable fields. In this study, we conducted basic experiments for the realization of remote sensing technology of hazardous substances based on the resonance Raman effect. The resonance Raman effect is a phenomenon in which the intensity of Raman scattering light is greatly enhanced by excitation with light of a wavelength corresponding to the electronic transition energy of the target substance. The presence of electronic transition energy of substances can be confirmed by observing the ultraviolet absorption spectra. Many hazardous substances exhibit ultraviolet absorption in the deep ultraviolet wavelength region of 300 nm or less. Therefore, in this study, we constructed a resonance Raman spectrum measuring device capable of wavelength sweeping in the deep ultraviolet wavelength range, selected SO₂ and NH₃, typical corrosive gases, as target substance, and verified experimentally the enhancement of Raman signal intensity by resonance Raman effect.

Forensic evaluation of crime scenes normally involves examination of textile fibers, to find out the association between an individual and a crime scene, or between a suspect and a victim. The forensic samples normally include a mix of various types, sizes (micro to nano - scale) and shapes of natural and synthetic fibers, which are very difficult to differentiate/identify. Various sophisticated analytical instruments are being used to carry out the examination of these fibers. They involve various microscopy and spectroscopy based techniques, most of which are very complex and highly sensitive. Further, they may require a series of sample preparation steps to get high selectivity and are highly time consuming. Here we report a fluorescence microscopy based synthetic (plastic) fiber detection method using Nile Red (NR) dye, which provides high selectivity for synthetic fibers. The methodology involves the use of NR dye which selectively stains the fibers collected on filter papers following separation from samples/soils and water. The selectivity of NR towards the fibers is due to their non-polar property. Binding with NR makes the fibers fluoresce when viewed under a fluorescence microscope. This selectivity of NR for fibers makes the identification of fibers lot easier and less timeconsuming in forensic samples when compared to the more commonly used optical microscopy (where the presence of naturally-occurring substances of similar size can result in more errors). The paper will discuss optimisation of various parameters and method validation for detection of synthetic fibers and microplastics from soil samples. As an example, our method has shown to provide distinct clarity for the analysis of microfibers. The potential for the application of the method for faster forensics analysis will be discussed.