The Night Vision & Electro-Optics Laboratory (NV&EOL) is generating a tactical target imagery data base to use for the development and evaluation of image processing algorithms. The data base contains both digital and analog data taken by a variety of image sensors. The digital image dta sets, presently part of the data base, are described as to the scenario, sensor, meteorological data, format and other pertinent information as available. The procedures used to collect the data, generate the digital form and prepare the ground truth of the imagery are explained. Plans for future expansion of the digital imagery are discussed.

Infrared imagery from various terrain backgrounds have been collected by the Environmental Research Institute of Michigan and analyzed to determine the statistics of both radiometric and spatial features. This paper describes some of the characteristics in the form of histograms, equivalent ellipse statistics, and power spectra for several infrared bands. Data are included for nine wavelength regions of common interest. These were acquired over a variety of terrains during the four seasons and at various times of the day and night. Additionally, the industrial target data collected as part of the development of a DARPA image data base for autonomous terminal homing technology evaluation is described. The terrain backgrounds data were collected with the ERIM M-7 scanner, a multispectral scanner which operates in several wavelength bands in the visible and infrared portions of the electromagnetic spectrum. The terrain backgrounds imagery were collected with the scanner looking both downward and in a direction 35° below the horizontal. The industrial target imagery was collected using an ERIM-modified Reconofax X scanner. The target imagery was collected at depression angles of 6, 10, 20, 75, and 90 degrees.

The Air Force Armament Laboratory (AFATL) is currently involved in programs to develop increased capability against massed armor threats. The requirement for small air-to-sur-face anti-armor weapons capable of detecting targets in a high clutter background is recognized by Air Force exploratory development agencies. This paper describes the effort now under way to evaluate the performance of candidate infrared seekers against armor targets in various backgrounds. To this end, a target/background model is being built using measurements made of tactical targets and backgrounds under a variety of conditions. An in-frared scene, generated by the model, will be presented to candidate seekers which will exercise target detection and tracking algorithms as scene range and aspect angle change according to a flight dynamics model. The philosophy of generating realistic infrared scenes from measured data will be discussed as well as techniques required to represent armor targets.

The performance of many infrared seekers and sensor systems operating in the far infrared spectral region of 8 to 12 microns is critically dependent on the magnitude of thermal signatures (i.e., thermal contrasts, or, equivalently, differences in thermal radiance values between various scene components). This paper describes the Thermodynamic Armored Vehicle and Environment Thermal Signatures (TAVETS) Model for estimating the day-to-day diurnal variations in thermal radiance and signatures for armored vehicles and selected scene components. Because precipitation effects were not present in the selected field test data analyzed, the modeling accuracies reported in this paper are valid only for moderately good weather.

Fire products such as smokes and heat which are introduced into the atmosphere as a result of burning grasses and vehicles are identified and factors which influence their amounts and concentrations are discussed. When possible, factors affecting the optical properties of the fire perturbed atmosphere such as smoke particle size density distributions and optical density are presented. The impetus for the study lies in the need to assess the performance of electro-optical sensors in an adverse environment and the paper is aimed toward identifying factors to be considered in overall system performance modeling.

Qualitative comparisons of electro-optical (EO) sensor field performance measurements and model predictions are presented for a variety of nonideal weather situations including fog, haze, rain, high absolute humidities, and high temperatures observed at Wright-Patterson AFB, Dayton, Ohio. Commercially available 3-5pm and 8-l2pm AGA Thermovision imagers and an RCA silicon charged coupled device (CCD) TV camera are being used at the Air Force Avionics Laboratory (AFAL) Targeting Systems Characterization Facility (TSCF) to observe natural and man-made targets. The imagery are analyzed digitally to determine how much weather degrades EO sensor performance. Initial comparisons of field sensor perfor mance measurements and model predictions indicate that the AFAL Sensor Performance Model (ALSPM) correctly predicts sensor performance degradation trends due to nonideal weather. Future field sensor measurements and validation efforts will attempt to quantify ALSPM prediction accuracy and isolate critical input data.

A new technique developed at General Dynamics Convair Division permits rapid calculation of sensor performance including full statistical treatment of atmospheric and target conditions. The most common statistical method in use relies on the Monte Carlo technique in which thousands of single point calculations are made with data from a random number generator. The Monte Carlo method is replaced by a closed form calculation requiring orders of magnitude less computer time. The probability density function of atmospheric attenuation is used along with the appropriate distribution of target characteristics. The probability density function of the video signal-to-noise ratio is calculated directly using simple numerical integration. Performance criteria can then be applied to the curve for either observer or autonomous operation. Sample results are presented of FLIR and laser 3-D imager performance against targets in European atmospheric conditions.

This paper describes the design analysis and results of performance test of a rosette scan infrared sensor which can be used as a thermal imager as well as a radiation seeker for search and tracking. The fundamental properties of a sensor are discussed, and the relationship between sensor parameters has been made clear. The effect of bandwidth on the noise equivalent irradiance(NEI) is also investigated by using a simplified model. On the basis of these analyses, the sensor which responds in the 3 to 5 μm spectral band has been designed, fabricated and tested. It has been verified that the minimum NEI over the total field of view could be realized by optimizing the bandwidth at the center of the field of view. The ability as a thermal imager has also been demonstrated by displaying rosette-scanned thermal images on a CRT monitor, and the detection capability of a target on a display has been evaluated by measuring the minimum detectable temperature difference(MDTD).

A serial scan thermal imaging system with a multielement matrix type detector is discussed. The parallel TDI signals are converted into a serial signal by using CCD delay lines in the system. By this process, high horizontal scan efficiency is realized, and the configuration of the high rotating scanner is miniaturized. In the system NETD depends on the scanning parameters, so scanning method satisfying simultaneously the requirements that NETD should be decreased and horizontal scanner should be miniaturized is discussed. On the other hand, ability of a target recognition by a thermal imaging system is frequently described by MRTD. Optimization of MRTD is also discussed and the optimum MRTD is obtained when the product of IFOV and a target spatial frequency is equal to 0.5 (cycle) . Finally the concrete system design adapted to midrange (order of a few kilometer) vehicle size target and some results of the performance evaluation of a prototype system manufactured are presented.

A compact, lightweight, reliable, high performance Mini-FLIR designed for autonomous operation will be described. The heart of the unit is the scanner which includes the mechanical raster scanner, the detector array, and the electronics. The basic scanner unit is a cylinder 4 inches in diameter by 5 inches long, weighing only 3.5 lb. (1.59 Kg). It operates in the 8 to 12 pm spectral region and has a field of view (FOV) of 30 by 40 degrees. It can be combined with different optical systems to provide a wide range of capability for a variety of military and commercial applications. Mechanical scanning of the FOV is accomplished with a 525-line raster at a field/frame rate of 60/30 per second. Serial scan techniques are employed, using two multi-element arrays of mercury cadmium telluride (HgCdTe) detectors, with the two arrays time=-hared and combined through an acoustic delay line. The number of detectors can be chosen for the degree of sensitivity required. Thermal compensation of the optical system is provided to maintain the FLIR in focus over a wide temperature range. Pictures of the Mini-FLIR as packaged with multiple FOV optics for remotely piloted vehicle (RPV) and missile seeker applications will be presented.

An expendable, autonomous vehicle by definition and implication will require small, low-cost sensors for observation of the outside world and interface to smart, decision-making avionics. This paper describes results of several interrelated CCD camera projects directed toward achieving such an integrated sensor package. A shuttered high resolution CCD detector combined with a CCD analog frame store memory is described. This system results in a full resolution frame rate reduced, deinterlaced image. This image data is suitable for transform or differential pulse code data compression as well as various other 3 X 3 element operators directed at extracting image intelligence for on-board decision-making.

The analysis of an airborne line-scan surveillance system based on a linear-array solid state sensor mounted parallel to the longitudinal axis of the aircraft with a scanning mirror providing cross-track area coverage, resulted in a simple kinematic model which can be useful in the systematic design of such systems. Practical implementation of the model using state of the art technology was considered prior to the actual design and construction of a laboratory prototype and simulation system. Experimental evaluation of the system confirmed that a digital data rate of 4-8 Mbit per second is feasible for acquisition and real-time display of visual information at a ground area coverage rate of almost 200 km2/hr for relative flight altitude of 3000 ft and velocity of 80 knots, attaining a ground pixel size of 0.25 m. A digital image processing facility was interfaced to the system providing offline processing capabilities for selected image-frames. Online analog and digital processing was performed to provide compensation for errors due to image motion and aircraft attitude, as well as acquisition errors due to uneven illumination. Restoration of some geometrical distortions was also performed.

Key to the successful deployment of autonomous vehicles envisioned under future defense system programs is the ability for automated target identification. For true vehicular autonomy, this automation of the decision-making function should be extended to include incompletely known problem environments, perhaps with less than perfect supervisory information being available for training and design of the target recognition processor. Driven by these real-world requirements, this study develops a panoramic view of the entire learning environment scenario and delineates the most suitable techniques/algorithms available for addressing the identification problem under this spectrum of environments. This brings into focus the state-of-the-art in this field and serves as an information base for defining the directions for further developments required to meet the needs of autonomous vehicle missions.

Every improvement in infrared detector sensitivity afforded by modern technology amplifies apparent background "clutter", and complicates the task of designing target/background discrimination algorithms for autonomous surveillance systems. To the extent that background clutter rather than detector sensitivity limits over-all performance, the sensitivity inherent in the detector is not being fully exploited. In this paper we design adaptive threshold point target detectors that operate in the background-shot-noise-limited regime against structured backgrounds.

Autonomous operation of tracking systems implies the need for abilities which are predicated on discriminating a target from nontarget objects and noise. Intelligent discrimination is most often hampered by lack of explicit in-formation, hence requiring elaborate processing to extract implicit information for an electronic "brain." The fact that it is now possible to construct sensors which provide not only brightness information on a two-dimensional map, but also motion and range on equivalent maps, bypasses the aforementioned processing to directly provide sufficient explicit information for the discrimination of targets. The only processing then required is to combine the information in a composite map such that targets stand out clearly from nontargets. This is particularly true of airborne targets where range and motion are generally quite distinct relative to their background. This general philosophy pertains as well to ground targets, in that sensing directly those characteristics which distinguish them from other objects greatly facilitates the tasks of locating and identifying such targets. For ground targets, sensors designed to sense characteristics such as straight lines, smooth surfaces, color, symmetry and motion would, by use of appropriate composite mapping, detect and acquire predefined classes of targets with high levels of discrimination.

Signals from a solid state electro-optical line sensor, which samples a two-dimensional image brightness function in time and space, can be digitally processed to extract the ground velocity vector of relatively slow, autopilotlcontrolled aircraft such as mini-RPVs. This sensor can be rotated into the direction of motion by a stepping motor which is controlled by a computational unit using simple easily realizable algorithms to keep the sensor in alignment with the velocity vector as well as to compute its magnitude. Together with other instruments already installed onboard the aircraft, this combination of sensor and computational unit may form an instrumentation setup which can be used in passive, autonomous navigation systems. Computer simulated experimental runs proved that a sufficient degree of directional sensitivity and overall accuracy can be attained with the proposed method.

The cross-covariance function is commonly used to locate conjugate imagery in digital stereo-pair images. Statistical analysis of the covariance function leads to expressions for the probability of correlation error, P , and the estimated mean-square-error over a region of imagery. These expressions depend only on the image auto-correlation function, the match window, and the number of correlator taps. The presence of distortion between images is modelled by comparinc a function f(x) with its distorted form f(h(x)). The effect of distortion on the expected cross-covariance function is shown to be a convolution with the image auto-ccvariance function. The result indicates that the choice of the true correlation oeak does not necessarily imply correct image conjugacy. The effect of distortion on the error prediction expressions is discussed.

The problem of matching two images of the same scene is important in many applications including navigational guidance, georgraphic map matching, environmental studies and diagnosis of diseases with medical x-rays. In the general case, the images were produced by completely different sensors such as radar and optical sensors at different viewing geometries. Prior to the matching, digital signal processing must be made to line up the two images geometrically and to match their intensities. Objects of interest represented by subimages in one scene can be located in the other using hierarchical search techniques by a digital computer. Scene matching performances can then be presented in terms of the probability of a match as a function of the probability of false fix.

The feasibility of a 1000:1 bandwidth compression ratio for image transmission has been demonstrated and described operationally in a previous paper. For such a system to be practical, it must be flexible enough to deal with wide variations in scene environment. Such flexibility has not been achieved in previous computer vision systems. This paper describes an organization and control methodology for achieving system flexibility. The methodology was successfully incorporated and tested as part of the intelligent bandwidth compression system to provide the system with the capability to deal with diverse scene variations.

The need for handoff capability arises when a target is acquired with an aircraft sensor (such as forward looking infrared (FLIR)) and it is desired to transfer the target position data to a second imaging sensor associated with an on-board missile, or possibly an-other aircraft. Westinghouse has developed scene-matching algorithms which can accomplish this task. The autonomous handoff problem places several demands on the scene-matching algorithm. For reasons of survivability and successful attack, target transfer should be accomplished within a fraction of a second. For tracking purposes a precision of less than one pixel (TV line) is required. Because of the boresight alignment characteristics of the two sensors, handoff must be possible if the initial misregistration of the two sensor fields of view is as much as half of the second sensor field of view (FOV). A laboratory demonstration of han•off capability has been implemented by Westinghouse as a result of an Army problem. The paper will describe the handoff algorithms, the test procedure, and the results achieved to date.

Conventional target tracking approaches rely on numerical correlation over successive frames on a window around the target. They are, therefore, sensitive to partial obscuration and changes in target and background appearances. Further, multiple target tracking requires replication of the hardware. In this paper, we present a multiple target tracking approach based on a dynamic scene model, derived from the analysis of a time sequence of imagery. Simulation results demonstrate multiple target tracking in cluttered backgrounds and in imagery from fast moving platforms. The approach is implementable as an integral part of the Honeywell target screener system. Several multitarget tracking application scenarios eminently suited to this approach are identified.

Supertracker represents a programmable parallel pipeline computer architecture that has been designed to meet the real time image processing requirements of auto-cueing target data processing. The prototype bread-board currently under development will be designed to perform input video preprocessing and processing for 525-line and 875-line TV formats FLIR video, automatic display gain and contrast control, and automatic target cueing, classification, and tracking. The video preprocessor is capable of performing operations full frames of video data in real time, e.g., frame integration, storage, 3 x 3 convolution, and neighborhood processing. The processor architecture is being implemented using bit-slice microprogrammable arithmetic processors, operating in parallel. Each processor is capable of up to 20 million operations per second. Multiple frame memories are used for additional flexibility.

This paper describes a digital simulation which uses measured infrared backgrounds and targets for the evaluation of various scan and clutter reduction concepts. Measured infrared background images were digitized, edited, and placed in mass storage of a CDC Cyber 172 computer. Software was developed to combine the small 4° x 4°) measured images into a larger scene (, 12' x 12') suitable for most IR homing seeker simulations. Programs were written which perform the scene scanning process, the summation of pixels within a small IFOV, and simulation of the signal processing. Examples are given of the relative signal at the scanner's aperture, the Fourier transform of this signal, and the filtered output obtained via an inverse transform. A comparison of the scan ing noises for a mid-IR versus an LWIR band are shown for a few sky/cloud backgrounds. The peak signal-to-peak background noise in the mid-IR band is also given for a tank located in a few example terrain back-grounds.

A critical problem in target detection is to develop algorithms for autonomous target acquisition at long ranges. When the angular extent of a target is so small that identification is impossible, then detection of object motion can be an important method for acquiring and tracking military targets. Moving target identification (MTI) may be difficult however for down-looking airborne sensors where the apparent motion of back-ground clutter can mask the movement of any objects in the scene. To get around this problem, an algorithm has been developed which compares successive frames from a passive imaging sensor and estimates the apparent background motion. The velocity model is applied to an earlier image to predict the appearance of the most recent image. The predicted and measured images are compared and the portions of the background which did not move in the expected manner are identified as moving objects. Several sequences of FLIR images are included where the algorithm has successfully extracted targets from cluttered moving backgrounds.

Under contract to the Army's Night Vision and Electro-Optics Laboratory, Westinghouse has been investigating the design, test, and implementation of a set of algorithms to perform intelligent target tracking and intelligent target homing on FLIR and TV imagery. The intelligent tracker will monitor the entire field of view, detect and classify targets, perform multiple target tracking and predict changes in target signature prior to the target's entry into an obscuration. The intelligent tracking and homing system will also perform target prioritization and critical aimpoint selection. A system concept was developed, several frame-to-frame tracker designs are compared, and several example scenarios from work performed on the NV&EOL base are analyzed. One scenario involves the reacquisition of a partially obscured target under substantially changed conditions which preclude the use of standard change detection techniques. This type of analysis, in conjunction with an analysis of the intelligent tracker functions, and an analysis of AAH, RPV, and PGM scenarios containing an intelligent tracker served as a basis for the system concept.