Method and apparatus for detection of moving objects by sar images

Ί8Π ΓΠ31!Μ1 fl1»Jt!SK3 ΓΠΪ3 ni’187 ’ib’S1? ΠΒ’ΒΖΙ Γ3Ί5?ϊί SAR 310»

METHOD AND APPARATUS FOR DETECTION OF MOVING OBJECTS BY SAR IMAGES

Applicant: Galileo Avionica S.p.A

Our ref.: 5053\4

"METHOD AND APPARATUS FOR DETECTION OF MOVING OBJECTS BY SAR IMAGES"

The present invention relates to a method and an apparatus for detection of moving objects by SAR images. SAR (Synthetic-Aperture Radar) is a microwave pulse-transmission radar system carried by an aircraft.

The use of microwaves enables acquisition of images at any moment, irrespective of whether it is day or night and irrespective of the meteorological conditions, and consequently SAR is widely used in environmental monitoring, in military applications, in geological mapping, and in other applications .

The energy of the radar pulse, transmitted by an antenna, is reflected by ground surface in all directions, and part thereof is captured by the same transmitting antenna. This signal, generally referred to as "echo", is then processed for generating a sequence of images. The sequence of images is obtained by integrating the radar signal over a given time interval, referred to also as time of observation.

The image resolution in the direction of pointing of the antenna (range direction) depends exclusively upon the bandwidth of the microwave pulses transmitted. The image resolution in the direction perpendicular to the direction of pointing of the antenna (azimuth direction or cross-range direction) depends upon the time of observation, i.e., the time of reception of the echo signal. In order to obtain a high resolution image, quite long observation times are necessary, which generally depend upon the distance observed and upon the velocity of the radar platform, and may even be of the order of tens of seconds. Consequently, recognition and location of possible moving objects in the scene observed are very imprecise. In a way similar to what occurs with an ordinary camera, in fact, a moving object gives rise to a "wake" in the image acquired. Obviously, the wake is all the more evident, the longer the time of observation and the faster the object is moving.

The aim of the present invention is hence to provide a method and an apparatus for detection of moving objects by SAR images that will be able to overcome the described drawbacks of the prior art and that at the same time will be fully efficient and reliable.

According to the present invention, a method and an apparatus for detection of moving objects by SAR images are provided, as defined in Claim 1 and in Claim 11, respectively.

For a better understanding of the invention, some embodiments thereof will now be described, purely by way of non-limiting example and with reference to the attached drawings, wherein: - Figure 1 schematically shows an aircraft equipped with an apparatus for detection of moving objects by SAR images according to the present invention, and moving surface targets; - Figure 2 shows a simplified block diagram of the apparatus of Figure 1; - Figure 3 shows a more detailed block diagram of a part of the apparatus of Figure 2; - Figure 4 shows a schematic representation of quantities regarding the present invention; - Figure 5 shows a more detailed block diagram of a part of the block diagram of Figure 2; - Figure 6 is a flowchart of a procedure of a method according to one embodiment of the present invention; - Figure 7 shows a schematic representation of quantities regarding the method according to the present invention; and - Figure 8 is a flowchart of a procedure of a method according 3 to a different embodiment of the present invention.

Figure 1 shows, by way of example, an aircraft 1, which is moving in a pre-determined direction A with a velocity VA and is equipped with a synthetic-aperture radar (SAR) apparatus 2 according to the present invention. Figure 1 moreover illustrates objects 3 moving on a portion of the Earth's surface illuminated by the SAR apparatus 2.

As shown in Figure 2, the SAR apparatus 2 comprises a singlechannel radar transceiver 2a, provided with a processing unit 4. The device 2 generates two-dimensional SAR images in the range and azimuth (or cross-range) co-ordinates and moreover recognizes and locates objects moving on the scene being observed, such as for example the surface targets 3. SAR images are images in a numeric format, defined by a matrix of points or pixels, each of which has a respective value indicating the reflectivity of a corresponding portion of the ground surface observed.

Figure 3 shows the processing unit 4 implementing the method according to the present invention. In particular, the processing unit 4 comprises an image-generation module 5 and a detector module 6, which operate in parallel. The image-generation module 5 is configured for supplying, starting from the radar signal RP, a sequence of high resolution images IA given by the aperture TA, in a known way.

In detail, the image-generation module 5 comprises, cascaded to one another, an analog-to-digital converter 7, an in-phase and quadrature sampler 8, a filtering and decimation module 9, and a fast-processing module 10 operating in the time domain, which form a· pre-processing line 5a. The image-generation module 5 further comprises an image-generation module 11, arranged downstream of the pre-processing line 5a. The analog-to-digital converter 7 receives at input the radar signal RP captured by the radar transceiver 2a and converts it into a numeric format. The in-phase and quadrature sampler 8 extracts an in-phase signal I and a quadrature signal Q from the digitized radar signal RP and supplies them to the filtering and decimation module 9. The fast-processing module 10 operates in the fast-time domain (i.e., in the time domain associated to processing in the range direction) and, in a known way, supplies a pulse-repetition frequency signal PRF starting from the in-phase signal I and from the quadrature signal Q, which are filtered and decimated. In particular, the fast-processing module 10 is configured for carrying out a procedure for compensating the motion of the SAR apparatus 2 (i.e., of the aircraft 1) with respect to the portion of the ground surface illuminated. Finally, the image-generation module 11 produces a sequence of two-dimensional high resolution images IA, by coherent integration of the pulse-repetition frequency signal PRF.

In greater detail, the high resolution images IA present a resolution in the range direction or range resolution Rr and a resolution in the azimuth (or cross-range) direction or azimuth (or cross-range) resolution RA that are substantially the same as one another. Hereinafter, by "range direction DR" is meant a direction that is radial with respect to the radar transceiver 2a, and by "azimuth (or cross-range) direction DA" is meant a direction perpendicular to the range direction. Normally, the azimuth direction is parallel to the motion direction A of the aircraft 1.

The coherent integration time of the pulse-repetition frequency signal PRF necessary for obtaining an azimuth resolution RA that is equal to the range resolution Rr is referred to as "synthetic aperture" or "aperture time" TA. In practice, considering the velocity VA of the aircraft 1, the aperture time TA is given by the extension or aperture that a radar antenna would need to have to obtain an azimuth 5 resolution RA and a range resolution Rr that are the same as one another.

The detector module 6 is configured for detecting and locating the objects 3 moving on the portion of the ground surface illuminated by the SAR apparatus 2. In particular, the detector module 6 supplies a list of the moving objects 3 with the respective co-ordinates CM and velocity VM (module and direction) starting from the pulse-repetition frequency signal PRF at output from the fast-processing module 10.

In greater detail, the detector module 6 comprises a subaperture generator 12, an image-difference module 13, a pattern-recognition module 14, a validation module 15, and a motion-estimation module 16.

The sub-aperture generator 12 generates a sequence of low azimuth resolution images ISA obtained by coherent integration of the pulse-repetition frequency signal PRF for a subaperture or sub-aperture time TSA shorter than the aperture TA. By way of .example, the sub-aperture TSA is approximately 0.25 s; normal values for the aperture TA are generally higher than 1 s and can reach 8-10 s.

The image-difference module 13 receives on its input the sequence of low azimuth resolution images ISA and performs a pixel-by-pixel difference between consecutive low azimuth resolution images ISA, supplying a sequence of difference images ID on its output.

The pattern-recognition module 14 receives on its input the sequence of difference images ID and processes it for identifying specific patterns typically associated to the movement of an object (hereinafter referred to as "doublets" D) . In practice, the pattern-recognition module 14 processes each difference image ID to recognize the doublets D present and supplies a list of the recognized doublets D to the validation module 15. The validation module 15 carries out a validation procedure for recognizing series of doublets D effectively corresponding to respective moving objects 3 and for eliminating any possible artefacts due to false alarms or to residues of the other-than-perfect motion compensation applied by the fast-processing module 10.

The motion-estimation module 16 receives the sequence of the validated doublets Dv from the validation module 15 and processes them so as to reposition the moving objects 3 in a map, associating to each of them a position CM (in range and azimuth co-ordinates) and a velocity VM (modulus and direction).

As indicated previously, the detector module 6 uses the low azimuth resolution images ISA produced by the sub-aperture generator module 12. In a low azimuth resolution image ISA, the azimuth resolution RA is degraded with respect to a high resolution image IA, since the low azimuth resolution image ISA is obtained considering shorter times of coherent integration. However, the trace of moving objects 3 is much more concentrated and hence enables more precise location.

In addition, the detector module 6 works on the difference images ID (between consecutive low azimuth resolution images Isa) > which contain particularly significant information as regards the moving objects 3. In fact, the portions of the difference images ID for immobile objects are substantially zero: in two consecutive low azimuth resolution images ISA, in fact, the reflectivity values corresponding to immobile objects remain unvaried and are hence eliminated by the difference operation. As shown in Figure 4, instead, a moving object 3 produces, in the difference image ID, a doublet D, i.e., a waveform that comprises two regions 19, 20 close to one another, each having a homogeneous sign but opposite to

I the sign of the other. In addition, pre-determined morphological characteristics of the first region (19; RP) and of the second region (29; RN) are equal, except for a predetermined tolerance. In particular, the regions 19, 20 have respective peaks 21, 22 of substantially the same amplitude and of opposite sign (hereinafter referred to also as "dual" peaks) . The peak 21, which is positive, indicates a position Pi of the object in the most recent low azimuth resolution image ISA, whereas the peak 22, which is negative, indicates a position P2 of the object in the least recent low azimuth resolution image ISA. In addition, the distance between the dual peaks 21, 22 of the doublet D indicates the space covered by the moving object 3 in an interval of time equal to the sub-aperture TSA. Consequently, each doublet D also supplies information regarding the velocity VM of the corresponding moving object 3.

The pattern-recognition module 14 carries out a first selection, eliminating all the components of the difference image ID that cannot be put down to doublets D. A second, more accurate, selection is carried out by the validation module 15, to eliminate components of the difference image ID that may be regarded as doublets D, but are caused by anomalous fluctuations not produced by the moving objects 3. For this purpose, the validation module 15 carries out a procedure for tracking each doublet D identified, in particular, by verifying the congruence of the movement starting from a series of consecutive difference images ID. For example, a doublet D is validated and associated to a moving object 3 if it persists in a minimum number of consecutive difference images ID, and the positions of the dual peaks 21, 22 are compatible with the velocities previously estimated (i.e., if a minimum number of consecutive

difference images ID contains a series of doublets D compatible with the movement of an object having the velocity VM estimated on the basis of the positions of the dual peaks 21, 22 of each doublet D of the series) . Doublets D that do not meet the validation requirements are not recognized as moving objects and are eliminated.

The tracking procedure moreover supplies an accurate estimation of the velocity VM of the moving objects associated to the validated doublets Dv.

Finally, the motion-estimation module 16 receives from the validation module 15 the sequence of the validated doublets Dv, processes them so as to reposition the moving objects 3 in a map, and supplies a list of the moving objects 3 present on the portion of the ground surface illuminated by the SAR apparatus 2, associating to each of them a position CM and a velocity VM.

Figure 5 shows in greater detail the pattern-recognition module 14, which comprises a peak-search module 17 and a doublet-search module 18.

The peak-search module 17 analyses pixel by pixel each difference image ID received at input and generates a list of all the peak values that exceed a pre-determined threshold, associating them to respective sets of parameters, amongst which the co-ordinates, the amplitude, and the sign.

The doublet-search module 18 scans a neighbourhood Δ of predetermined extent around each peak in order to seek a further peak having substantially the same amplitude and opposite sign (dual peak) . If the search is successful, the doublet-search module records the presence of a doublet D. The doublets D identified (still not validated) are candidates to being associated to moving objects 3.

The procedure implemented by the doublet-search module 18 is represented in the flowchart of Figure 6.

Initially (block 23) , the doublet-search module 18 selects a peak (n-th peak) from the list of the peaks present in the difference image ID. Next (block 24), the doublet-search module 18 seeks a further possible dual peak of opposite sign in the neighbourhood Δ of the peak selected.

At this point (block 25), if in the neighbourhood Δ no possible dual peak has been identified, the search for doublets D is resumed starting from block 23, with selection of a new peak ((n+l)th peak) the

If, instead, in the neighbourhood Δ a possible dual peak has been identified, having a sign opposite to that of the first, the doublet-search module compares the amplitudes of the first peak selected and of the possible dual peak (block 26). If the two peaks have different amplitudes, the search for doublets D is resumed starting from block 23 with the selection of a new peak ((n+l)th peak). If the peak selected and the possible dual peak have substantially the same amplitude, the presence of a doublet D is recognized, and the data regarding the doublet identified are stored (block 27) (for example, the coordinates and amplitudes of the peaks are stored).

Finally (block 28), if the difference image has been completely scanned, there is supplied the list of the doublets •D identified, accompanied by corresponding parameters. Otherwise, scanning proceeds starting from block 23, with the selection of a new peak ((n+l)th peak).

The method and apparatus according to the present invention advantageously enable estimation of the position CMt and velocity VM of a moving object in a much more precise way than what is possible using conventional methods that consider complete synthetic apertures. 10

In addition, the method and device according to the present invention are able to identify also those moving objects that possess a component of velocity exclusively according to the azimuth direction DA.

Finally, the method described can function effectively using a single-channel radar transceiver and is very light from the computational standpoint.

According to a different embodiment of the invention, the recognition of the patterns is carried out according to a procedure that will be described hereinafter, with reference to Figures 7 and 8. In this case, the patterns that are sought in the difference images ID are a positive region RP and a negative region RN, which are dual with respect to one another. The positive region RP and the negative region RN are portions of a difference image ID, in which the values of the pixels have a homogeneous sign, a positive sign and a negative sign, respectively. By "dual" it is meant, in this case, that the positive region RP and the negative region RN have substantially the same shapes and dimensions, except for predetermined tolerances, and are both comprised in a confidence region RC, which has also a pre-determined extension (indicated by a dashed line in Figure 7) . In practice, the presence of dual positive regions RP and negative regions RN indicates the movement of an extended object.

With reference to Figure 8, the pattern-recognition module 14 is configured for initially searching for peak values PV in a difference image ID (block 30; see also Figure 7). Then (block 31), a threshold filter is applied to portions of the difference image ID for generating a filtered image IF. In greater detail, for each peak value PV a threshold TH is defined, that is equal to a pre-determined fraction of the peak value PV itself (for example, 1/2; note that each threshold TH has the same sign as the respective peak value 11 PV) . Then, using windows W of pre-determined dimensions (indicated with a dashed-and-dotted line in Figure 7), centred on each peak value PV, the pattern-recognition module 14 calculates the pixels IF(x,y) of the filtered image IF on the basis of the values of the corresponding pixels ID(x,y) of the difference image ID in the following way (x and y indicate the co-ordinates in the range and azimuth directions, respectively): Γ 0 if 1 Id(x, y) 1 < TH 1 IF(x,y) = 1 if Id(x, y) > TH > 0 < -1 if Id(x, y) < TH < 0

Once the operation of filtering around all the peak values PV previously identified is completed, the values of the remaining pixels of the filtered image IF are set egual to the values of the corresponding pixels of the difference image ID.

Next (block 32), the filtered image IF is examined, and positive regions RP and negative regions RN having homogeneous and non-zero values are identified.

The morphological characteristics of adjacent positive regions RP and negative regions RN are then compared (block 33), considering a tolerance margin, in order to identify dual positive regions RP and negative regions RN. For example, the comparison is carried out by extraction and comparison of salient morphological characteristics (maximum and mean dimensions, area, perimeter, distribution about the centroid and the like; the tolerance is fixed in terms of maximum percentage difference admissible between corresponding characteristics).

When dual positive regions RP and negative regions RN are identified (output YES from block 34), the recognition of a pattern indicating a moving object 3 is completed (block 35) . 12

If a positive region RP and a negative region RN examined do not dually correspond, the recognition fails (output NO from block 34). 5 After the recognition procedure has been completed on all the positive regions RP and negative regions RN (output YES from block 36), a validation procedure (block 37) Is carried out, substantially as previously described. In this case, the validation envisages that the presence of dual positive 10 regions RP and negative regions RN is verified in a predetermined number of consecutive difference images ID. The velocity can be estimated, for example, on the basis of the positions of the centroids of the dual positive regions RP and negative regions RN. 15

Finally, it is evident that modifications and variations can be made to the device and method described herein, · without thereby departing from the scope of the annexed claims. , crrixan rwzn arn pnow pnszn irn nr “|»ob ,Ρ’ηη ηχΰ- πώοβπβ mzrna π^πιώώ np’ioz .zruwan ms rnp’nn pmi1? oxnm □ιηπη Pi? _____88°4003 25 JU am 0631 :u 40200 .(moiB nzrnn) cras&'an nwa