Ship-based automatic detection of small floating objects on an agitated sea surface remains a hard problem. Our
main concern is the detection of floating mines, which proved a real threat to shipping in confined waterways
during the first Gulf War, but applications include salvaging,search-and-rescue and perimeter or harbour defense.
IR video was chosen for its day-and-night imaging capability, and its availability on military vessels.
Detection is difficult because a rough sea is seen as a dynamic background of moving objects with size order,
shape and temperature similar to those of the floating mine. We do find a determinant characteristic in the
target's periodic motion, which differs from that of the propagating surface waves composing the background.
The classical detection and tracking approaches give bad results when applied to this problem. While background
detection algorithms assume a quasi-static background, the sea surface is actually very dynamic, causing
this category of algorithms to fail. Kalman or particle filter algorithms on the other hand, which stress temporal
coherence, suffer from tracking loss due to occlusions and the great noise level of the image.
We propose an innovative approach. This approach uses the periodicity of the objects movement and thus its
temporal coherence. The principle is to consider the video data as a spacetime volume similar to a hyperspectral
data cube by replacing the spectral axis with a temporal axis. We can then apply algorithms developed for
hyperspectral detection problems to the detection of small floating objects.
We treat the detection problem using multilinear algebra, designing a number of finite impulse response
filters (FIR) maximizing the target response. The algorithm was applied to test footage of practice mines in the
The main application of a scatterometer is the determination of the wind speed and direction at the sea surface. This
is achieved by measuring the radar backscattering coefficient in three different directions and inverting these measurements
using a geophysical model function (GMF). The scientific value of the data is directly related to the quality of the
There are currently two european C-band scatterometers operating, one on-board the ERS-2 spacecraft launched in
1995 and the other on-board METOP-A, launched in 2006. The similarity of the two scatterometers is an opportunity
to ensure the continuity of more than 15 years of global scatterometer measurements. To achieve the consistency of the
backscattering coefficients data sets, required for long-term climate studies, an accurate cross-calibration is vital. The
cross-calibration is made possible since the two spacecrafts operate simultaneously from 2006 up to now.
As the backscattering coefficients measured by the scatterometers depend on acquisition time, location on the ground
and on the geometry of the measurements (incidence and look angle), a direct comparison of measurements made by both
instruments is practically impossible.
In particular cases, models can be used to cope with measurement differences. On the rain forest, assumed to be
time-invariant, homogeneous and isotropic, the backscattering coefficient depends only on the incidence angle, and the
constant gamma model can be used to cope with the incidence angle effects. On some ice covered areas (e.g. Greenland
and Antarctica), assuming that the ice surface is isotropic, the ice line model can be used. It is a function of incidence
angle and ice age and depends on the location. On the ocean, which is inherently not stable in time, the CMOD5 GMF
is used. CMOD5 relates the observed backscatter to the geophysical parameters which are the wind speed and wind
direction. Using the last model, measurement biases can be assessed making simultaneous observations unnecessary.
The coastal zone is an extremely dynamic system. Variations in the concentration of its major constituents occur rapidly
over space and time. This is in response to changes in bathymetry and tidal forces coupled with the influences of fronts,
upwelling zones and river inflow. Today's research on the functioning of estuarine and coastal ecosystems, as well as
attempts to quantify some of their biogeochemical fluxes are based on highly time consuming and costly sea campaigns
and laboratory analyses.
On September 2002, an airborne campaign using CASI sensor covered part of the Scheldt estuary (Belgium-
Netherlands coastal zone). A 13 sampling stations field survey was realised in order to cover as quickly as possible the
wide range of water quality encountered from the mouth of the estuary to the outer limit of the plume. Correlation was
searched between classical ground truth measurements and the rich information provided by numerous CASI-SWIR
spectral bands carefully chosen. These relations were not sufficient enough to derive synoptic view of the spatial
distribution of many biogeochemical parameters in the Scheldt estuary and plume.
In this research we found that some biogeochemical parameters of interest in estuaries and plumes that were retrieved
using imaging spectroscopy techniques as the MF (Matched filtering) and the MTMF (Mixture Tuned Matched
Filtering) are very encouraging. We showed that using those spectra based processing techniques we could accurately
obtained the concentration distribution of suspended particulate matter (SPM) and particulate organic matter (POM),
that we could not retrieved using the classical statistical techniques. Moreover, using the imaging spectroscopy
techniques we significantly improved the coloured dissolved organic matter (CDOM) concentration classification,
relatively to the results derived using the multiple regression technique.
The primary measurement objective of the Advanced Scatterometer ASCAT, a spaceborne real aperture C band
radar, is the determination of wind fields at the ocean surface. Unlike AMI instruments on-board ERS satellites,
ASCAT uses long transmit pulses with linear frequency modulation (chirps) allowing the application of low peak
transmission power while retaining a high SNR. A pulse-compression is performed on the received signal.
This paper will focus on the impact of the use of pulse compression in particular on the location accuracy
of the samples in presence of external perturbations. An eventual location error has important consequences on
the normalization as well as on the geolocation of the measured data.
In the first Gulf War, unmoored floating mines proved to be a real hazard for shipping traffic.
An automated system capable of detecting these and other free-floating small objects, using readily available sensors such as infra-red cameras, would prove to be a valuable mine-warfare asset, and could double as a collision avoidance mechanism, and a search-and-rescue aid. The noisy background provided by the sea surface, and occlusion by waves make it difficult to detect small floating objects using only algorithms based upon the intensity, size or shape of the target. This leads us to look at the sequence of images for temporal detection characteristics. The target's apparent motion is such a determinant, given the contrast between the bobbing motion of the floating object and the strong horizontal component present in the propagation of the wavefronts. We have applied the Proesmans optical flow algorithm to IR video footage of practice mines, in order to extract the motion characteristic and a threshold on the vertical motion characteristic is then imposed to detect the floating targets.
High resolution sonars are required to detect and classify mines on the sea-bed. Synthetic aperture sonar increases the sonar cross range resolution by several orders of magnitudes while maintaining or increasing the area search rate. The resolution is however strongly dependent on the precision with which the motion errors of the platform can be estimated. The term micro-navigation is used to describe this very special requirement for sub-wavelength relative positioning of the platform. Therefore algorithms were designed to estimate those motion errors and to correct for them during the (ω, k)-reconstruction phase. To validate the quality of the motion estimation algorithms a single transmitter/multiple receiver simulator was build, allowing to generate multiple point targets with or without surge and/or sway and/or yaw motion errors. The surge motion estimation is shown on real data, which were taken during a sea trial in November of 2003 with the low frequency (12 kHz) side scan sonar (LFSS) moving on a rail positioned on the sea-bed near Marciana Marina on the Elba Island, Italy.
The assessment of the performances of ground-penetrating radar (GPR)
in humanitarian demining is an important problem. These performances
are related to the relative strength of the target radar response with respect to that of the soil. Many parameters influence both responses. The physical and geometrical parameters that influence the target signature include the soil electromagnetic (EM) constitutive parameters, the target depth and orientation with respect to the soil surface, the antenna height and the target EM and geometrical properties.
This work presents a numerical parametric study of the soil and target radar signatures. The advantages of the numerical approach are: it allows for a separate study of the influence of each parameters on the radar responses, it is fast, cheap, generic with regards to hardware, and finally it is not prone to experimental errors and hardware failures or misuse. Moreover it is always possible to link the numerical experiments with a particular hardware by characterizing this latter. However, a number of simplifications, such as modeling the soil as a planar multilayered medium, are introduced to keep the problem tractable.
This study yields surprising results, such as for example the possibility of considering the target in homogeneous space for computing its signature, as soon as it is a few centimeters deep. The target considered in the numerical experiments is a dielectric cylinder representing an AP mine, with diameter 6 cm and height 5 cm, and εrt=3. These values are chosen to approach as much as possible the physical properties of the M35BG AP mine, which is small and therefore difficult to detect.
This paper analyzes the effect of the soil on the response of a metal detector (MD). The total response is first decomposed in a direct coupling between the transmitter and the receiver, the mine contribution and the soil contribution. The mine contribution is further related to its free space signature by introducting a number of transfer functions (TFs). Those TFs characterize the effect of the soil on the field propagation, from the transmit coil to the mine and back to the receiver, and on the mine signature. The expressions derived are quite general. However the TFs and other quantities of interest can only be computed if the scattering problem has been solved. For this it is usually necessary to resort to numerical techniques. Such techniques are computationally expensive, especially to analyze the various effects of the soil as they require to compute the solution for a large set of parameters. Therefore, we propose to model a buried mine by a multilayered sphere. From outside to inside, the layers represent the air, the soil, the mine explosive and the mine metallic content. Further, the analytic solution for such a multilayered sphere is used to compute the mine and soil responses, the mine free space signature and the various TFs as a function of the parameters of interest such as the soil electromagnetic (EM) properties or the mine depth. Finally, the validity domain of a number of practical approximations is discussed.
From the beginning of its mission, in 1995, the ERS-2 satellite has recorded an important
set of data. The performance and accuracy of its instrument provide precious
information for the scientific community. The experience acquired during
10 years has led the European Space Agency (ESA) to plan a reprocessing activity of the
entire set of the available scatterometer data.
This reprocessing activity will use the enhanced on-ground processing1
and calibration2 chains.
In this paper, the calibration strategy for the scatterometer data reprocessing
from the beginning of the mission is presented.
It consists in looking for
a calibration area (rainforest, ocean or ice) which would allow a highly accurate tuning of the
antenna patterns (already tuned within the specifications).
Although other uses have emerged, ERS scatterometer data is operationally used to measure wind speed at the surface of the oceans. The wind speed and direction can indeed be inverted from the measured backscattering coefficients provided the measurements were performed over sea. While a land-mask can be used to reject measurements made over land, operational constraints make the use of an externally-provided ice-mask unpractical. It is thus desirable to discriminate between measurements made over sea and measurements made over ice using the backscattering coefficients alone. Due to
operational constraints, a temporal averaging of the measurements
is not feasible. Several methods have been proposed to discriminate between sea and ice. These are based on measuring the distance
between the measurements made and a model. An ice model and a wind model are available. Measurements located far from the ice model were most likely not performed over ice and similarly, measurements close to the wind model were most likely performed over sea. However, for particular values of the incidence angles, these models are very close to each other, which leads to classification errors. In this paper, we propose to enhance the criterion of the distance to the wind model by taking into account the wind direction. This permits a better discrimination between ice-and sea-measurements. The enhanced criterion is implemented using a neural-network. The other methods proposed in the literature are also implemented in the same neural-network framework, which permits an easy comparison of their relative performances. Finally, the various methods are combined in a Bayesian framework.
The scatterometer on-board of the ERS spacecraft is range-gated, in opposition to Seasat/Nscat scatterometers that are Doppler filtered. It can thus admit some Doppler frequency shift in the returned echo. However, the bandwidth of the filter on-board of the ERS satellite limits the Doppler shift that can be tolerated on the returned echo data. For these reasons, the ERS spacecraft needs to be yaw-steered in order to minimize the residual Doppler shift. Due to a malfunction of several of the on-board gyroscopes used to govern the attitude of the ERS-2 spacecraft, precise yaw steering of the spacecraft cannot be guaranteed anymore. This paper reviews modifications to the existing processing chain that are needed to be able to process the data acquired by the ERS-2 spacecraft in degraded attitude mode, so called zero-gyro mode (ZGM). The main modification is the introduction of the on the fly estimation of the yaw angle as input to a model of the acquisition geometry. This paper also details how the yaw angle is estimated from the received raw data by measuring the residual Doppler frequency shift. Finally, several improvements such as the impact of an increased spatial resolution of the backscattering coefficients are also discussed.
Due to gyroscopes malfunctions, the ERS-2 spacecraft cannot accurately be yaw steered. Moreover, the actual yaw angle of the spacecraft is unknown. Since the yaw angle is not known in advance and not periodic, the look-up tables-based original scatterometer processor is not able to compute accurate values for the backscattering coefficients from the measurements made. This implies the need for an upgraded wind scatterometer ground processor in order to obtain accurate backscattering coefficients. Moreover, the upgraded processor includes several other enhancements. This paper presents the results of the validation of the upgraded processor. The validation of the geometrical model is performed by comparing geometric parameters such as the incidence angle and the sub-satellite track heading. The radiometric performance of the upgraded processor is first evaluated with data acquired in nominal attitude, by comparing the backscattering coefficients. The radiometric performance of the upgraded processor is then further assessed with data acquired in degraded attitude. This is only possible over the calibration test site and over other selected land areas.
Humanitarian demining is a very dangerous, cost and time intensive work, where a lot of effort is usually wasted in inspecting suspected areas that turn out to be mine-free. The main goal of the project SMART (Space and airborne Mined Area Reduction Tools) is to apply a multisensor approach towards corresponding signature data collection, developing adapted data understanding and data processing tools for improving the efficiency and reliability of level 1 minefield surveys by reducing suspected mined areas. As a result, the time for releasing mine-free areas for civilian use should be shortened. In this paper, multisensor signature data collected at four mine suspected areas in different parts of Croatia are presented, their information content is discussed, and first results are described. The multisensor system consists of a multifrequency multipolarization SAR system (DLR Experimental Synthetic Aperture Radar E-SAR), an optical scanner (Daedalus) and a camera (RMK) for color infrared aerial views. E-SAR data were acquired in X-, C-, L- and P- bands, the latter two being fully polarimetric interferometric. This provides pieces of independent information, ranging from high spatial resolution (X-band) to very good penetration abilities (P-band), together with possibilities for polarimetric and interferometric analysis. The Daedalus scanner, with 12 channels between visible and thermal infrared, has a very high spatial resolution. For each of the sensors, the applied processing, geocoding and registration is described. The information content is analyzed in sense of the capability and reliability in describing conditions inside suspected mined areas, as a first step towards identifying their mine-free parts, with special emphasis set on polarimetric and interferometric information.
The MsMs project is a major campaign to collect calibrated and well-documented data, suitable for use by workers developing advanced multisensor algorithms for antipersonnel mine detection. The data, together with a full description of the site layout and measurement protocols, are publicly available via the internet site http://demining.jrc.it/msms. Measurements are made on a test lane consisting of 7 plots of different soils, each 6m by 6m, populated with surrogate mines, calibration objects, simulated clutter and position markers. There are 48 targets in each plot, configured identically for all plots. A first report was presented last year. Since then, laser acoustic vibrometer and magnetometer data have been added and the metal detector and thermal infrared data have been augmented. The database has been reformatted to make it more uniform and user-friendly and to remove typographic mistakes. The test site remains essentially unchanged, apart from some equipment upgrades, and is available for further data collection. In particular, the targets have not been moved, so as to provide stable surrounding soil conditions representative of mines left undisturbed for long periods post-conflict. This presentation will describe the new data and data format, the status of the upgrades and the outlook for the future.
We discuss the problem of detecting minelike shapes in data coming from mine detection sensors that can provide images, such as an imaging metal detector, an infrared camera or a ground-penetrating radar. Firstly, we show a way for selecting possibly dangerous regions that should be further analyzed, i.e. to which shape analysis methods should be applied. Then, two shape detection methods are presented, both based on the randomized Hough transform. Most of the mines are of a cylindrical shape, so, due to some burial angle, they appear elliptical in 2D images that are taken parallel to the ground. Thus, one of the two presented methods deals with the detection of elliptical shapes. The other method is developed for detecting the hyperbolic signatures of mines in B-scans (vertical data slices into the ground) of ground-penetrating radar data. Finally, pieces of information that can be extracted from detected ellipses and hyperbolas are discussed, and two ways are suggested for their further use towards determining whether a particular selected region contains a mine indeed or not. Both methods are illustrated on real data.
The presented work aims to automatically register high-resolution polarimetric SAR images with each other and other types of images. A digital topographic map is used as an aid for the registration. SAR images are very different from visual or infrared images. The idea is to identify, for each type of image, objects present on the map and easily detectable in the image. Detecting these objects in the image and matching them between map and image provides a first registration. Several object detectors were developed for the subsequent stages of the registration. Each of these detectors is briefly described. The actual registration uses a hierarchical method. First the SAR image is converted into ground range. Then a rough registration between image and map is obtained based on the position of forests and/or built-up areas. A voting method is used to find the parameters of a simple transformation model and to match the objects between map and image. The third step finds the parameters of an affine transformation based on the objects matched by the voting method. To improve the registration, objects with low 3D structure, e.g. roads and rivers, are used. The method for detecting these in SAR images yields an incomplete results leading to ambiguities for the optimal local displacement. Optimisation methods are used to overcome this problem and yield the parameters of a global transformation model. The accuracy of the registration is now within the accuracy of the map. Once the different images are registered with the map, the results of edge detectors are used to refine the registration between them.
The aim of this article is to explore new methods to enhance the results of automatic interpretation of SAR images by combining images acquired from different viewing directions (multi-aspect SAR images). Using the combined information extracted from multi-aspect images allows to resolve problems of obscurance, by for instance the borders of a forest, to increase the resolution and to augment the confidence in detection as compared to detection in single images. The article focuses on high-resolution polarimetric images for the automatic interpretation of an airfield scene. Specifically for this type of images we have developed a set of new image interpretation tools such as edge detectors and bar (line) detectors, both based on multi-variate statistics. These detectors are briefly described in the article. The main part of the proposed article will focus on how the use of multi-aspect images can enhance the results of these detectors. The multi-aspect images are supposed to be accurately registered. It is thus possible to warp them into a common coordinate system. Because the spatial resolution of a SAR system is usually not the same in range and azimuth, it is sometimes better to detect objects in each image separately and fuse the results of the detection at the object level. This is particularly true for the detection and delimitation of the buildings. On the other hand, edge detectors can benefit from combined information on a pixel-level. In particular edge detectors based on multi-variate statistical methods can be applied on registered images, thus increasing the confidence level of detection and reducing the false alarm rate, by combining the information at a low level. For edge detectors we will compare results of combining the information available from multi-aspect polarimetric images at different levels. In particular we will compare the results of applying them directly to the registered image set with these obtained when applying them on each individual image and fusing the results at the object level or intermediate (edge-strength) level. Similar investigations will be presented for the bar detectors. Results will be shown on a set of polarimetric L-band images of an airfield.
Migration is a common name for processing techniques that try to reconstruct, from the dat recorded at the surface, the reflecting structures in the sub-surface. Most of the existing migration techniques do not take into account the characteristics of the acquisition system and the ground characteristics. We propose a novel migration method, applicable on Ground Penetrating Radar (GPR) images, that integrates the time domain model of the GPR in the migration scheme. We calculate by forward modeling a synthetic 3D point spread function of the GPR, i.e. a synthetic C-scan of a small point scatterer. The 3D point spread function, containing system characteristics like the waveform of the excitation source, the combined antenna footprint and the impulse response (IR) of the antennas, is then used to deconvolve the recorded data. Results of this migration method on real data obtained by an ultra-wideband GPR system show that the migration method is able to reconstruct the top contour of small targets like AP mines, in some cases even the correct dimensions. The method is also capable of migrating oblique targets into their true position. The migration scheme is not computational intensive and can easily be implemented in real time.
In this paper, two methods for fusion of mine detection sensors are presented, based on belief functions and on voting procedures, respectively. Their application is illustrated and compared on a real multisensor data set collected at the TNO test facilities under the HOM-2000 project. This set contains data acquired by metal detector, infrared camera and ground penetrating radar. The data acquisition and preprocessing are briefly described. For some typical cases presented in this data set, the characteristics extracted and used by both methods are discussed, as well as the answers given by each method and possible causes of potential differences in results. Also, it is shown how the different voting schemes compare to belief functions modeling in various situations, based on the knowledge that is put into the belief functions. Since the roots of the two methods are different, i.e. belief functions involve expert knowledge while voting is a simple approach, the explanations involve these differences. Problems that arise when comparing and evaluating different methods are also addressed. Finally, it is shown that both of the methods have their advantages and drawbacks, depending on the measurement and operational conditions. This paper is a result of a joint work at three European institutions towards a common goal: humanitarian demining.
The objective of the Joint Multi-sensor Mine-signatures (MsMs) campaign is to organize and execute an experimental campaign for collecting data of buried land-mines with multiple sensors. These data sets will then be made widely available to researchers and developers working on sensor fusion, signal processing for improved detection and identification of land-mines, assessing the role of the operator in the detection process, etc. The outdoor test facility of the Joint Research Facility of the European Commission, located at Ispra (Italy), houses the test minefield. Six test strips of 6 by 6 m consisting of different soil types (cluttered grassy terrain, loamy soil, sandy soil, clay soil, soil with high content of organic matter, and ferromagnetic soil) are complemented with one reference test strip of 6 by 6 m consisting of pure sand. The list of objects buried in the minefield includes mine simulants of three different dimensions with either a low or a high metal content, reference targets for position referencing and calibration checking, and clutter objects including empty bullet cartridges, metal cans, barbed wire, stones, wood, plastic boxes, etc. This test minefield is going to be left intact for a long period, in order to be able to perform multiple runs on it. For the test campaign of the year 2000, the core sensors were a metal detector, a ground penetrating radar, a microwave radiometer, and thermal infrared imagers. Later, other (more experimental) detectors will also be tested on the same test minefield. The first data sets are in the process of being released right now.
Automatic contour detection in SAR images is a difficult problem due to the presence of speckle. Several detectors exploiting the statistics of speckle in uniform regions have been already presented in literature. However, these were mainly applied to multi-look low-resolution imagery. This paper describes two new CFAR contour detectors for high-resolution single-look polarimetric SAR images. They are based on multi-variate statistical hypothesis tests. Failing of the test indicates the presence of an edge. A test for difference in means on log-intensity images and difference in variance on complex (SLC) images are used. Both tests take into account the interchannel covariance matrix which makes them a powerful tool for contour detection in multi-channel SAR images. Spatial correlation jeopardizes the CFAR character of the detectors. This problem is often neglected. In this paper its influence on the detectors is studied and eliminated. The localisation of detected edges is improved using a directional morphological filter. Different methods to fuse the results of the two detectors are explored and compared. Results obtained on a polarimetric L-band E-SAR image are presented. Most contours are well detected. Narrow lines on a uniform background remain undetected. Although the detector was developed to detect edges only between uniform areas, it also detects edges between textured and uniform areas.
In this paper, ideas for modeling humanitarian mine detection sensors and their combination within Dempster- Shafer framework are presented. Reasons for choosing this framework are pointed out, taking into account specificity and sensitivity of the problem. This work is done in the scope of the HUDEM project, where three promising and complementary sensors are investigated, so detail analysis is performed in case of fusing the data from them. A way for including in the model influence of various factors on sensor and their result ins discussed as well and will be further analyzed in the future. The application of the approach proposed in this paper is illustrated on the case of sensing metallic objects, but it is possible to modify it for other situations.
In this paper, the time domain modeling of an indoor impulse UWB GPR systems, built in the scope of the HUDEM project, is presented. For an impulse UWB system, a time-domain modeling is an obvious choice. We explain how the antennas can be characterized by their normalized impulse response. By considering the antenna as a convolution operator, we get a mechanism for modeling the whole radar system as a cascade of linear response, which gives a lot of advantages and possible application. In our research it is used to express the radar range equation in the time-domain, to optimize the antenna configuration and to calculate the point-spread function of the UWB GPR at a given depth. The point-spread function can be used for migration by deconvolving it from the collected data. In this way the migration method takes into account the characteristics of the radar system. Finally, results of this migration method on data obtained by our UWB GPR system are shown.
A standard pulsed induction metal. detector is used to image buried metallic objects by scanning an area of interest. It is shown that, under specific hypotheses, the output image is the result of the convolution of a target function with a kernel depending on the incident magnetic field. Several hypotheses are considered, leading to different kernel shapes and different interpretations of the target function. As the detector imaging function is a low-pass filter, shape's details spread out and the resulting raw image are blurred, Since a high-pass restoration filter must be used to deconvolve the raw images, care must be taken to avoid a strong amplification of noise. The imaging filter is computed using a numerical simulation of the incident magnetic field. Finally, the restoration filter is computed using the Wiener approach. Results are shown for a couple of metallic pieces.
In this paper, the design and the modeling of an indoor impulse UWB GPR systems (1 GHz - 5 GHz), built in the scope of the HUDEM project, is presented. For an impulse UWB system, a time-domain modeling is an obvious choice. We explain how the antennas can be characterized by their normalized impulse response. By considering the antenna as a convolution operator, we get a mechanism for modeling the whole radar system as a cascade of linear responses, which gives a lot of advantages and possible application. In our research it is used to express the radar range equation in the time-domain, to optimize the antenna configuration and to tune signal- processing algorithms. The deconvolution of the signal source and antenna impulse responses is an ill posed operation. In this paper we present a method for decomposing an A-scan in a linear combination of wavelets, using the Continuous Wavelet Transformations -- by properly choosing the mother wavelet. This technique can also be used to reduce the amount of data for further processing. Finally results obtained by our UWB GPR system are shown. Advantages and shortcomings are discussed.
The aim of this paper is to propose a strategy that uses data fusion at three different levels to gradually improve the performance of an identity verification system. In a first step temporal data fusion can be used to combine multiple instances of a single (mono-modal) expert to reduce its measurement variance. If system performance after this first step is not good enough to satisfy the end-user's needs, one can improve it by fusing in a second step result of multiple experts working on the same modality. For this approach to work, it is supposed that the respective classification errors of the different experts are de-correlated. Finally, if the verification system's performance after this second step is still not good enough, one will be forced to move onto the third step in which performance can be improved by using multiple experts working on different (biometric) modalities. To be useful however, these experts have to be chosen in such a way that adding the extra modalities increases the separation in the multi-dimensional modality-space between the distributions of the different populations that have to be classified by the system. This kind of level-based strategy allow to gradually tune the performance of an identity verification system to the end-user's requirements while controlling the increase of investment costs. In this paper results of several fusion modules will be shown at each level. All experiments have been performed on the same multi-modal database to be able to compare the gain in performance each time one goes up a level.
An approach to the long range automatic detection of vehicles, using multi-sensor image sequences, is presented. The algorithm was tested on a database of six sequences, acquired under diverse operational conditions. The vehicles in the sequences can be either moving or stationary. The sensors also can be moving. The presented approach consists of two parts. The first part detects targets in single images using seven texture measurements. The values of some of the textural features at a target position will differ from those found in the background. To perform a first classification between target- and non-target pixels, linear discriminant analysis is used on one test image for each type of sensor. Because the features are closely linked to the physical properties of the sensors, the discriminant function also gives good results to the remainder of the database sequences. By applying the discriminant function to the feature space of textural parameters, a new image is created. The local maxima of this image correspond to probably target positions. To reduce the false alarm rate, any available prior knowledge about possible target size and aspect ratio is incorporated using a region growing procedure around the local maxima. The second part of the algorithm detects moving targets. First any motion of the sensor itself need to be detected. The detection is based on a comparison of the spatial cooccurrence matrix within one image and the temporal cooccurrence matrix between successive images. If sensor motion is detected, it is estimated using a multi-resolution Markov Random Field model. Available prior knowledge about the sensor motion is used to simplify the motion estimation. The motion estimate is used to warp past images onto the current one. Moving targets are detected by thresholding the difference between the original and warped images. Temporal and spatial consistency are used to reduce false alarm rate.
This paper presents the discussion and results in the field of automatic face identification. The implementation possibilities are presented, and the retained choices are motivated. The objective is to identify the person whose image is available from a grey-level camera. The approach is to extract characteristics that will be classified according to extracted characteristics of a database. One section is devoted to the importance of a proper acquisition method, based on profile images. Several sections are more technical and deal with the profile extraction, the computation of the curvature and the way characteristics are derived. This is naturally followed by practical results. Finally, some prespectives are listed to let the present work be integrated in a practical application where several hundred people must be identified.
It is clear that the compression of Meteosat radiometric data is not a classical problem. A first challenge consists in image coding with the intention to preserve the quality of measurements. This is quite different from image coding with the intention to preserve the visual quality. The latter problem is extensively addressed in the literature. The way in which the problem of the errors is addressed is quite different in the two approaches. When the visual quality is concerned, visual criteria are used and the error amplitude can be very large in some places. On the other hand, if the measured features are extracted from radiometric data like meteorological images, it is necessary that the reconstruction errors do not exceed some threshold depending on the required precision. A second challenge concerns the construction of progressive coding scheme which allows the progressive transmission of the image data, avoiding the artifacts of a block coding scheme. In the present work, only the measurement data compression problem has been considered and the tests were realized accordingly. However, these methods also perform quite well when the visual quality is addressed. This paper presents a progressive the incoming data in the scanning order. The comparison with the JPEG standard shows that the progressive wavelet method always performs better when the distortion is concerned.
In this paper, an approach to the automatic detection of vehicles at long range using sequences of thermal infrared images is presented. The vehicles in the sequences can be either moving or stationary. The sensor can also be mounted on a moving platform. The target area in the images is very small, typically less than 10 pixels on target. The proposed method consists of two independent parts. The first part seeks for possible targets in individual images and then merges the results for a subsequence of images. The second part of the algorithm specifically focuses on finding moving objects in the scene.
This paper presents an expert system for target recognition from short distance infrared images. This expert system first identifies the position and orientation of the target. The number of possible positions and orientations is extremely large, therefore, the solution space cannot be fully explored. A heuristic search algorithm is used instead to guide the expert system toward the solution in an efficient manner. Bayesian estimation theory and fuzzy logics are used to derive the knowledge combination rules used by the heuristic search algorithm. These rules are generic enough to be applied to other types of expert systems and to data fusion problems. Once the position and the orientation of the target have been found, the expert system simply tries to match parts of the image of the target with templates under the same position and orientation in order to identify the target.
In this paper, different methods are presented for the detection of important directions on thermal infrared images. These methods are compared to the well-known Hough method which performs poorly for these types of images. These algorithms are especially designed for an automatic target recognition system, but could as well be used for other types of application.
A generic method for target recognition is presented. The stress is put on the methods based on the neural networks and more specifically on the adaptive resonance theory (ART) models. This type of artificial neural network (ANN) has the advantage of being unsupervised and adaptive: it is indeed able to acquire and adapt its long-term memory taking into account the context evolution. ART networks very quickly recognize classes that are already known, they also learn new images very fast. Two versions of ART are investigated: ART1, which only works with binary data, and ART2, which is working with analog data. In practice, ART1 seems to need larger images than ART2 to achieve the same efficiency, but is obviously faster. A preprocessor has been developed whose output is invariant to translation, rotation, and scale changes of the input. The most important feature of this preprocessor is its ability to preserve visual interpretation, which is not the case for the more classical methods using Fourier-like and log/polar transforms.