 |
 |
 |
 |
| |
|||||
|
|
|
 SAR Imaging and Detection of Moving Targets Authors: R.C. DiPietro, R.P. Perry, R.L. Fante, and C. Teng  1.0 Introduction Future Intelligence, Surveillance, and Reconnaissance systems that exploit both moving target indication (MTI) and advanced synthetic aperture radar (SAR) systems to identify critical ground and air targets will require new approaches for processing and integrating sensor, computer, and target recognition technologies to support the technical and operational challenges that exist for these applications. MTI systems provide location and tracks of moving air/ground vehicles. SAR systems provide two-dimensional radar images of ground targets which allow further classification. Ground MTI radar is the sensor of choice to perform a wide area search. Most currently fielded systems do not automatically initiate and maintain tracking of specific vehicles. MTI can be used to track moving targets but not to identify them. Therefore, conventional MTI radar cannot sort out specific vehicle types from other vehicles present in typical traffic. As a result, sensor operators can be swamped by the volume of tracks, cannot separate critical targets from neutrals and look-alike vehicles, which can result in a high number false alarms. In addition, the emerging Unmanned Aerial Vehicle (UAV) platforms and related systems will be capable of gathering very large amounts of data from multiple sensors (MTI/SAR radar, infra-red IR, etc), which will aggravate the problem. To increase the number of identified and killed critical targets, a system will have to perform both moving and stationary target identification. For example, an exposed Transportable Elevated Launcher (TEL) typically may stand for less than a half hour but will usually spend much longer times in transit. When not ready to launch or in transit, the TEL is hidden. This was a major problem in the Gulf War as missiles launched from such mobile platforms were quite effective and difficult to locate and neutralize. Present deployed MTI systems can support rapid revisit times (necessary to find briefly exposed vehicles) but cannot ID targets. Moving target SAR imaging can provide this capability and thus expands the opportunity to perform target identification for periods of vehicle movement. This is important because targets often hide in caves, bunkers, and tunnels, which are out of radar sight. Currently, a TEL is mostly susceptible before and after launch movements. Furthermore, MTI can be used for cueing and the MTI track reduces the vehicle aspect uncertainties and makes rapid computer-based automatic target recognition (ATR) possible to aid the image analyst in selecting critical battlefield targets. Also, moving vehicles have restricted configurations and camouflage is limited, making ATR somewhat easier. In recent years high-resolution SAR imaging has been successfully demonstrated for non-moving ground targets. The SAR community is currently researching the imaging and identification of moving targets because of the importance of moving target imaging for wide area surveillance systems with limited revisit times. DARPA's Moving Target Exploitation (MTE) program objectives are to develop and demonstrate the synergistic combination of moving target tracking, High Range Resolution (HRR) MTI, and moving target SAR imaging to detect, track, and identify high value ground moving targets. This article discusses a new and innovative SAR moving target imaging algorithm developed on a MITRE sponsored research (MSR) program led by Richard Perry. 2.0 Moving Target Imaging Issues and MITRE's new Approach A SAR image is formed by processing many radar pulse returns from a target. Each pulse provides information as to the range to the target, and the pulse-to-pulse variations at a given range provides the necessary information to extract the azimuth target position. A moving target will pass through many range resolution cells during this data collection process (which may be in the order of many seconds) producing a blurred image using conventional ground focused SAR image formation techniques. The major innovation of MITRE's new approach to moving target imaging involves a data preprocessing technique known as Keystone mapping. The collected radar return data forms a two-dimensional sampled rectangular array in the radar pulse time (azimuth samples) / range spectral frequency domain. This data is processed into a Keystone shape (hence the name) by this mapping by performing a one-dimensional interpolation of the azimuth samples as shown in Figure 1.
Figure 1 Keystone Mapping The net effect is to correct for the linear component of the range migration without specific knowledge of the targets velocity and direction of travel. This linear component of motion is the main contributor to blurring, and eliminating it allows enough detail in the partially focused image to proceed with final focusing using more conventional techniques. This "divide and conquer" approach to focusing is carried out in three steps. The first step isolates the partially focused moving targets from the static clutter background for further focus processing. This is necessary because focus parameters will, in general, be different for each mover in the scene. Step two of the moving target focusing involves a two-dimensional phase correction of the extracted mover to remove residual quadratic phase errors associated with the target motion and keystone remapping. This requires an estimate of the normal acceleration of the target. This estimate can be obtained from unfocused image frames. It may also be estimated by simply iterating the single acceleration parameter to optimize a prominent point response in the unfocused target which is the preferred approach. The acceleration estimate may also be obtained by fitting a quadratic function to the phase associated with a prominent target point. Some errors in the acceleration estimate can be tolerated as these can be corrected in the third step of the imaging process. The acceleration estimate is also important to properly scale the final target image (provide accurate sample spacing in both the range and crossrange coordinates of the corrected image). To that end, MITRE uses a secondary Keystone technique (now applied to the quadratic time samples) to provide this estimate. The third step in the focusing process involves removal of residual cubic and higher order defocusing phase errors in the target image. These terms could be obtained by extending the techniques discussed in step two by estimating higher order motion derivatives of the targets. This has been done using the phase function curve fit and inverse filtering approaches. Another approach is to use standard nonparametric autofocus techniques such as Sandia's phase gradient autofocus algorithm or the ERIM-developed prominent point family of algorithms. Additional prominent point focusing techniques can also be used to provide correction phase estimates. These latter techniques appear to provide the best results. The detection and imaging techniques described here have been demonstrated on several target types using real SAR data collected as part of DARPA's MTE program. Figure 2 shows final SAR images of a TEL surrogate, tractor trailer, and M813 truck. In this example the vehicles were moving at 17 mph down a straight road over a 6-second collection dwell. The final 2 ft resolution images correlate well with their non-moving equivalents. The scaled image of the TEL surrogate moving along a circle of 80 meters in diameter with the same speed is shown in Figure 3. The scaled size is very comparable to the actual size of the TEL surrogate.
Figure 2. TEL, Tractor Trailer, and M13 Truck SAR Images Using Keystone Imaging
Figure 3. Scaled Image for TEL Surrogate on a Circle Using Keystone Imaging 3.0 Conclusions A new method has been presented to generate synthetic aperture radar images of moving vehicles that does not require specific knowledge of the vehicle's motion. The method uses a unique processing kernel which we call Keystone mapping to eliminate the effects of linear range migration for all vehicles regardless of their speeds and direction of travel. The results in an enormous saving in computation over "brute force" approaches that require a different matched filter for each possible target track. For more information, please contact Robert DiPietro using the employee directory. |
||||||||||||||||||
Solutions That Make a Difference.® |
|
|
