SAR Traffic Monitoring Using Time-Frequency Analysis for Detection and Parameter Estimation

Kurzfassung

In recent years many powerful techniques and algorithms for joint time-frequency analysis have been developed and successfully applied to ISAR imaging and SAR ground moving target indication (GMTI). Most of the proposed algorithms have their origin in the military field. In contrast to military applications where often the number of moving targets in the scene of interest is small (e.g. ISAR imaging of vessels), for civil road traffic monitoring applications the number of moving targets might be quite large, even in a relative small scene containing e.g. a autobahn or a major road section. The large number of moving vehicles in real road traffic scenarios and additionally the possibly small spacing between them implicate that RAW or range compressed data based algorithms have to cope with overlapping or crossing range histories of clutter suppressed moving target signals. Separating these signals from each other in the “range compressed data domain” for detection, motion parameter estimation and imaging purposes is still challenging. Instead of range compressed data also an already processed and clutter suppressed SAR image can be used for the same purposes, as known from the literature. However, by picking out successively azimuth lines from the complex SAR image one has compulsory to cope with multi-component linear frequency modulated (LFM) “azimuth” signals. For the SAR traffic monitoring application presented in the paper, it is essential that multiple moving targets signals can be detected and separated from each other. As already mentioned above, the signal range histories as well as their corresponding time-frequency histories may cross each other. For the detection of multi-component LFM signals and simultaneous Doppler slope estimation several methods were proposed in the literature. Whereas the relationship between the rotational Wigner-Radon transform and the fractional Fourier transform (FrFT) is known, a performance comparison between the FrFT and the matched filter bank (MFB) principle especially using real data is missing. Therefore in the paper we compare the FrFT and the MFB for detection and Doppler slope estimation purposes using simulated and real clutter suppressed data. The used real data were acquired by DLR’s new airborne SAR system F-SAR, operated in a aperture switching four-channel mode in X-band, during several traffic monitoring campaigns conducted in 2007. Furthermore, for separating the detected moving target signals from each other, filtering in the fractional Fourier domain is used. During the target separation procedure also residual range cell migration can be removed iteratively by using the estimated Doppler slopes. The problem of separating along-track velocity and across-track acceleration in the motion parameter estimation steps is addressed. It will be shown that a separation between velocity and acceleration in many cases is possible, if the SCNR of the target signal is not too small.