FR2609174A1 - Device for measuring displacement, speed or acceleration - Google Patents

Abstract

Device for measuring displacements, and/or speeds and/or accelerations, in two dimensions, and in rotation, for a vehicle from images of the roadway taken in succession by at least one camera mounted on the said vehicle. The image signals CAM output from the camera are sent to a computational unit UP through shift registers RD21, 22 and buffers B21, 22, taking from the image signals those corresponding to a window, and, to a set of shift registers RD1 mounted in series. The outputs of digital correlators CN4 are connected to the computer UP through a device for calculating the maximum in the sum of the correlation coefficients corresponding to the displacement of the vehicle during the time period in question. Application to land vehicles, in particular, to regulating and limiting speeds and fuel consumptions, to antilock, antislip and antiskid systems associated with steering control, and to automated position finding.

Description

DEVICE FOR DETERMINING DEPLACEMENT, SPEED OR ACCELERATION
The present invention relates to devices for determining displacements and / or speeds and / or accelerations with the aid of optical collectors finding their application particularly on land vehicles.

 The prior art connatt very many devices sensors displacements, speeds or accelerations can be mounted in land vehicles, air or sea. These devices use the most varied techniques and it would be difficult to give a summary.

 It may be emphasized, however, that most apparatuses use at least once in their measurement chatter mechanical elements necessarily related to wear problems. And if clean things have been done to limit this type of inconvenience, it has usually been to the detriment of accuracy. But even in the simplest and most traditional cases, that of automobile speed indicators, whereas the measurement ring should give a high precision, parameters such as inflating tires can greatly modify the measurements and introduce errors.

 The problem that the invention seeks to solve is that of an accurate measurement of the speed of a vehicle along one or two axes and possibly also taking into account the rotations. In addition, it has been sought, in accordance with the invention, to carry out precise, continuous measurements with reliable equipment not involving mechanical, pneumatic, hydraulic or other elements which cause wear phenomena.

 However, current technologies offer electronic solutions that achieve both high accuracy, ease of processing information displacement, speed or accelerator collected for various applications (recording, limitation and speed control and consumption, anti-slip or anti-lock systems, etc.).

 It has been proposed in the prior art (WO 84/01223 of CRANE Co.) a system for determining the speed of an aircraft with respect to the ground, raising an instantaneous profile of the ground, then a second covering part of the first, the comparison two profiles to deduce the relative positional shift and this by taking into account the altitude, to obtain the corresponding distance, and time elapsed, to obtain the speed. This system uses photodetectors.

 The applicant hereby filed the French application 85/09621 which concerns a device for measuring the speed and position of a mobile relative to the ground. It includes an electro-optical sensor targeting the ground whose video signals are processed by a computer: correlation means compare two by two the successive images and the collected data are stored in memory to allow the determination of a maximum of correlation between images from which displacement and speed are deduced; a central control device synchronizes the operation of the various elements (shooting, data integration and correlation of images). This application envisages the use of DTCs (charge transfer devices) for the correlation of one-dimensional images and matrices for two-dimensional images.

 Knowing the velocity vector by measuring the velocity along two axes and determining the rotation makes it possible, by integration, to define the path of the vehicle, which is very useful in all the techniques currently developing and affecting the cartographic tracking of the movements.

 These technologies using charge transfer (DTC) allow the realization of very small cameras giving scanned images (luminous intensity emitted by a point called pixel and marked by its coordinates in the matrix of sensors with N lines and M columns).

It is therefore possible with such cameras to observe from the vehicle the unfolding of the images and the relative displacement of objects or landmarks to deduce the speed.

 Note that the frontal or rear sight gives a delicate evaluation of the speed and requires in practice two cameras, the binocular vision indeed allowing the driver his direct appreciation of distances and speeds.

 The lateral vision gives a considerable amount of information, but at extremely variable depths thus complicating the selection of landmarks. The lower vision offers the advantage, for motor vehicles to see scrolling a soil whose distance to a camera linked to the vehicle, will vary very little, compared to the side landscape. The distance varies only because of the unevenness of the road less and less frequent and increasingly weak, and the vertical travel dO to the suspension of the vehicle. By designing a suitable optics and working under suitable shooting conditions, a depth of field can be obtained such that the extreme values of distance from the ground to the optics are included between the limits of this depth of field.

 It is therefore conceivable that between two successive images taken by the camera A charge transfer device, the displacement, in the field of the lens and thus on the photosensitive device, of a set of characteristic elements of the ground is measured. running under the vehicle.

 If the speed of the vehicle is very Flevee and that the time separating two successive images is such that no point of the ground seen on one image is found on the next one, it is possible to work with two camperas, the second one stinging in the time the passage of the characteristic points of the soil raised by the first.

In fact, the speed of shooting in DTC or DCC technique is such that one can usually be satisfied with a single camera. It is necessary to take precautionary measures. If, for example, a speed of 200 km / h is considered, this corresponds to 55 mm / msec. However, the image is taken during an exposure time which may be of the order of one hundred nanoseconds per pixel. If we do not try to obtain an image comprising a large number of pixels, we can arrive to a fast enough image which will not eliminate the presence of a slight blur for a vehicle traveling at high speed. In fact, experience has shown that this is not a disadvantage, since in the phase of correlation between successive images, we find, with the shift dO at speed, the same image elements with an identical or almost identical blur. . In addition, experience has also shown that for the purposes of speed measurement:
(a) we can reduce the number of information and processing times
corresponding to a low number of bits per pixel:
black / white image and minimum gray level number (O and 1).

(b) one can be content, if need be, not with a DTC matrix but with a
Linear DTC array to the extent that one does not seek to measure the
speed only along one axis (the axis of movement of the vehicle, in
general).

(c) we can accept a relative blur, if moving during shooting
image does not exceed approximately one fifth of the length
of the image taken in the sense of speed and preferably a
tenth.

This nevertheless means that for more complete measurements (speed along two axes and rotation, which does not do the prior art cited above), we must use a matrix N lines M columns, that is to say AT
N x M pixels.

 Before going into the description of the devices according to the invention, it should be remembered that the DTCs work in successive phases: each pixel (of the English "picture-element") receives a given amount of light, and puts a signal corresponding to the light intensity reviewed (two levels in the simplest case); the signals of the pixels are simultaneously transferred into a transfer register (end of the so-called integration operation) and then to one or more shift registers which send to the processing data by point, and / or by line or column and / or per frame and / or per image depending on the device chosen. Storage and processing are therefore generally sequentially, which is very fast compared to conventional vidfo cameras. According to the invention, it is possible to use a direct image memory, which considerably limits the transfer time and allows a good correlation between one acquired image and the next one.

 Indeed, conventional DTC devices operating o maximum frequencies of 10 to 20 MHz, which corresponds to a data output time of 100 to 50 usec (for a line of 1024 pixels).

 If, therefore, it is possible without any particular problem to use the conventional shift registers for a linear sensor, it will be understood that for the matrix sensor and the two-dimensional image, it is desirable to limit the sequential offsets and transfers and to use memories. direct images for correlation.

 In the case of the matrix sensor, the problem of the correlation is to find trace and position in each image of at least one zone of the previous image. According to the invention, a correlation window can be used. This electronically defined window covers for example a tile of the image divided into nine tiles (three lines three columns).

In this case, we choose the central tile on the side of the image that has just entered the field of the camera. If the vehicle moves substantially axially, the subject found in the tile of the first image will follow the axis of the assembly and traverse the next two tiles substantially along the axis followed. In a second phase, the correlation window will go through the second image to find the new position of the image corresponding to the surface of the tile chosen in the first image. If the velocity vector is axial, the window must be in the same axis of the second image with the corresponding offset due to the speed (and displacement). If the velocity vector has a lateral component, the window will recover its image with the corresponding lateral shift. If in addition there is rotation, the window will have to turn from the corresponding angle.

This means that the window must go through each image to find a trace of the tile that it had retained from the previous image.

 In FIG. 1, there is shown in a first image cut out nine squares, in b the following image with axial velocity vector (Vx), in c the following image with velocity vector V with lateral component Vy and in FIG. next image with lateral component Vy and rotation d. On the four drawings, the hatched tile corresponds to the electronic window.

 It is therefore expected that the window must explore each image, which can be done according to a mud of realization of the invention in 2 ms.

 We can thus understand the interest that could represent in the future memories such as optical optical memories (optical RAY) that record directly the image and are readable as a conventional random access memory.

 It is also necessary to specify the general conditions of use of a correlation window.

The dimensional choice of the window depends on:
the number of pixels of the sensor,
- extreme speeds to measure
- the exposure time of the sensor,
- the resolution of the camera.

The resolution is the ratio between the observed area (that is, that of the object corresponding to the field) and the number of pixels of the sensors. The resolution is therefore measured in surface units. When the object is immobile, the resolution hereafter referred to as static AS is the smallest area observable by the camera.

When the object is moving (in relative motion) for a speed
linear V and a time at, the image becomes fuzzy, each pixel receives a
average illuminance information from the RS surface but from the RS surface (1 + V.at). The smallest observable detail must therefore cover a surface greater than the static resolution RS.

We will take the following notations for the calculations below:
Vmin x minimum speed to be measured;
Vmax = maximum speed To be measured; Hymen - minimum speed variation detected or accuracy on speed; time = acquisition time; = = time between two shots;
RS = static resolution;
RD = dynamic resolution.

Dynamic resolution is considered in the worst case, ie for Vmax. If one places oneself in the simplest case of the linear displacement, during the exposure time rt, the length traveled is l = V. # t therefore
Lmax V At (1)
Lmax Vmax
Referring to FIG. 2 which represents the field (hatched area) at times t and t 4 At (beginning and end of exposure, the overlap zone corresponds to the dynamic resolution RD, the displacement Lmax between times t and T + rt being as defined above Vmax.At.

 The field must therefore cover a pattern longer than Lmax. If we consider that the resolution must be of the order of the minimum length to be observed, then RD is of the order of Vmax. # T.

The correlation window must therefore have a length estimate in number of pixels to:
F - k. RD (2)
RS formula in which k is the number of basic information contained in the window.

Between two successive images we move from D = V. # t.

If one wants a precision on the speed of aVmjn, it is necessary that:
AV. . Has
my is greater than 1 (where p is the number of pixels of the pattern).

 Thus, an elementary difference in velocity will correspond to a displacement of the observed pattern on p pixels. Of course, p = 1 is a lower limit to undesirable minimum precision.

If p pixels corresponding to the smallest variation of speed observed, to measure the maximum speed, it will be necessary:

ou encore:

The observed field length therefore involves a sensor length in pixels of:

or:

If we consider that the images succeed one another at times that are consistent with the time of shooting (successive shots taken),

 The number of pixels in the window must therefore be at least equal to N.

It will be recalled that this is true in the hypothesis of linf-areas and therefore unidinnsional measurements.

 The present invention is based on these principles studied for linear measurements, but had to reconsider them taking into account that the measurements are no longer limited to one-dimensional space (linear measurements) but correspond to a two-dimensional space ( measurements in the plane according to two axes of orthogonal preference) and in rotation.

It will be noted here that if, in absolute terms, the correlation between two successive images has to take account of Cartesian x and y coordinates and rotations e> the importance of the correlation operations can be reduced insofar as, in the case of the automobile, for example, the displacement along the axis of the vehicle is generally greater (considered between two successive shots) than the lateral displacement, the angle of rotation between two successive images being, in general, weak even in the event of steer and / or significant skidding if the time interval between successive images is small. If in the following and to simplify and clarify, reference is made to DTC or DCC matrices, optical fields and electronic windows all substantially square, we can actually be satisfied with relatively narrow rectangular surfaces
relative to the length taken in the direction of the longitudinal axis of the vehicle.

To better understand the technical characteristics and
advantages of the present invention, there will be described examples of
realization being understood that these are not limiting as to
their mode of implementation and the applications that can be made. In addition to the figures previously described, reference will be made to FIG.
schematically represents a block diagram of a device according to the invention and in FIG. 4 which schematically represents field, image and
window.

A matrix DTC sensor camera is connected to a conventional set of pretreatment control, adaptation and management of the information provided by the sensor, all, not shown in Figure 3, being connected by Sd output to the CAM input Figure 3. The data
from the camera and its auxiliary circuits are fed from two sets of shift registers. The charge of these data to the first set RD1 is made through two gates ETI1 and ET11 and an YES gate controlled in VALII by a nonrepresented validation device 6 alternatively attacking the gates ET1 and ET12 with a logic inverter II. shift register Ru11, RD12, ... RD1N corresponds to the recording of the signals of one of the N rows of the matrix sensor. These various elements numbered 1 AN are mounted in series so that during the sequenced transmission, the data from the sensor pass through the N elements successively.

 This set for putting an image into memory is shifted under the control of a clock CLK1 and a validation device VAL12 (not shown) connected to each element RD11 to RD1N-through the gate ETl3.

Each of the elements is capable of recording the data relating to M pixels. It will be noted here that the sensor may have a matrix, for example, at I x J pixels (with I = J in the case of a square matrix) and record in the shift registers RD11, RD11 only M x N pixels with
M s I and N s J, if the capacity of the elements of the trade and especially the use that one wants to make according to the invention can allow a reduction of the number of useful information. As it was pointed out above in the case of a motor vehicle, the speeds and therefore the most significant displacements are longitudinal, speeds and lateral or rotary displacements are generally plu: weak, hence the use of optical fields narrower in the transverse direction than in the longitudinal direction
In parallel with the set of shift registers RD1, the data coming from the sensor are fed to the shift registers RD21 and RD22 associated with buffers B21 and B22 themselves connected to the data bus of the computer UP which is not shown in FIG.

 This set RD2 / B2 is controlled by a device comprising two gates ET21 and ET22 validated alternately in VAL2 by means of a logic inverter 12 and receiving the pulses of a clock CLK2.

 The data coming from the camera CAM are thus shifted alternately in the registers RD21 and RD22. When one is loaded, the data is directed to the other. The unloading of the registers is thus carried out alternately by means of the buffers B21 and 322 mounted in parallel and read by the computer UP. This therefore allows the transmission of each image to the uP calculator for processing which will be defined hereinafter. The data output bus D3 of the IIP computer is connected to a set of the same general design as the previous one but operating in the opposite direction.

 This other set also consists of shift registers RD3 and RD32 associated with buffers B31 and B32. The information from the computer UP via the data bus D3 is fed to the buffers B31, B32 and transferred to the shift registers RD31, RD32, the various operations being controlled by a clock CLK3. The output of each shift register RD31, KD32 is connected to a gate input ET31, ET32 whose other input is controlled by a enable signal arriving at VAL3 and for one of the gates ET31, ET32 via A logic inverter 13. The outputs of the AND31, ET32 gates are connected to the inputs of an OR3 gate. This arrangement makes it possible to empty the shift registers RD31, RD32 alternately to supply the digital correlators CN4 hereinafter described.

 A series of K digital correlators CN41, CN42> CN43,... CN4K with K = N <I, connected in series, receives on the input A of each correlator CN4 the signals coming from the shift registers RD1, each correlator is at The pixels with LM <M # J. Moreover, on the other input of the first correlator CN41 (input B), the signals coming from the shift registers RD31, RD32 are received via the gate OR3 and the gates ET31, ET32.

 These digital correlators CN4 are controlled from the same clock signals CLK1 and validation VAL12 as the set of registers with offsets HDI, via the gate ET13.

 The signals from each of the serial CN4 correlators are sent to a multiplexing unit MUL followed by an accumulator ACC which performs the sum of the correlation coefficients and searches for the maxima. This accumulator is connected to LAI locking devices associated with a POS position counter.

 The outputs of the accumulator ACC and LAT locking devices are connected to an assembly INT ensuring the interface with the data bus of the computer PP.

 In Figure 3 we have not detailed the scheme of this set of calculation and commdnde that can between any conventional type comprising at least one central computer associated with the necessary RAM and dead and at least one clock. The role of this set will be explained below.

The image signals (input and / or output on the data buses of the computer pP), the various validation signals (VAL1, VAL2 and VAL3 ', and clock (CLK1, CLK2, CLK3) constitute the basis of the information exchanged between this set of calculation and the various components of the device according to the invention shown in Figure 3.

 Referring to Figure 4, there is shown schematically the respective dimensions of the various images according to an exemplary embodiment.

The field of the matrix sensor is represented by the rectangle at I lines of
J pixels. The selected portion to integrate for correlation operations is represented by the rectangle has N rows of M pixels. The one selected for the window is the rectangle of K lines of L pixels.

 It has been explained, above, why it was sufficient to take into consideration a part of the image representing, preffX, ence, an elongated rectangle in the direction of circulation of the vehicle (the arrow represents the direction of circulation of the recorded images by the camera).

On the left side of FIG. 4, the image N x M retained is taken from the central part of the image I x J taken by the camera on the leading edge side. Similarly, the part K x L of the recorded N x M image retained for the window is taken at the leading edge center.

 The numbers M, N, 1, J, K, L, are chosen taking into account the commercial components and the technical requirements imposed by the maximum speed and the precision sought, as pointed out above. Similarly, the number of levels corresponding to each image is determined according to the needs but it was found that in most cases it could be limited to two levels 0/1 (black / white).

 Thus, the shift registers RD21, 22, 31, 32 and buffers B21, 22, 31, 32 may have 16-bit inputs / outputs, the CN4 digital correlations operating on 7 bits, etc.

The operation of the device of FIG. 3 is as follows:
As an example, the camera A. matrix I x J 256 x 256 takes 120 frames per second non-interlaced with output data line by line. The floor is preferably illuminated by a pulsed laser controlled at the desired rate. The interest of a pulsed laser consists of a reduction to blur due to the speed and a better immunity to the noises coming from the lighting outside the system because the whole works in a precise band of radiation (a filter bandpass is then added to the lens of the camera). The image is transmitted to a set of control, pretreatment, adaptation and management of the sensor of any conventional type then to
N shift registers RD11 A RDIN which can for example be each of type 1 x M = 1 x 256.

Simultaneously, the set of shift registers RD91, RD22 and buffers B21, B29 loaded by the data bus, the RAM RAM nemory associated with the computer with the data relating to the window, under the known VAL ?,
CLK2. The window K x L thus de fi ned is stored in random access memory.

 Since the image signals arrive sequentially line by line in the shift registers RD11 å ROIN, the image travels successively, at the rhythm imposed by VAL12 and CLK12, each of the N registers connected in series. Validation VAL11 pellet the arrival of a new image from the CAM camera or the recycling of the previous image e from RD1N, and this for the reasons indicated below.

 For its part, the computer PP processes the images received by the data bus from the set of shift registers RD21, RD2L buffers B21, B22, and retransmits them after processing by the data bus to the buffers B31 B32 and the buffers. shift registers RD31, RD32, to arrive at the set of CN correlators under the validation command VAL3.

These correlators CN41 to CN4K receive, on the one hand, shift registers RD11A RD1K, the image data arriving in the registers RD1, and on the other hand, the window data coming from the registers RD31, 32, which ensures correlation.

 These digital correlators CN4 actually receive the window cut in an image of order n so as to search in the image of order n + 1 the new coordinates of the contents of this window.

The data arriving sequentially, pixel by pixel and line by line, in the set of shift registers RD11 to RD1h ', the corr-la-
Thus CN41 to CN4K will receive image data scanning the n pixel-by-pixel and line-by-line image, the window being fixed. In fact, this amounts to moving relative to making the n-order window go through the entire n-order image from one end to the other of each
line and line by line from where a systematic exploration. In addition to the work of coordination of the operations carried out by the computer assisted by its clocks and its memories, which ensures a correlation in two dimensions, the central calculator1 while the scanning and correlation operations of the n-order image, will process the window to rotate a small angle AC and repeat the operation on the n-order image recycled to Rflil by VAL11. The operation is then restarted by rotating the window a little more in one direction and / or in the other to control the rotations on the right and on the left.

 The signals from the CN4l to CN4Y correlators (i.e. the binary values of the correlation coefficients) are passed through MUL multiplexers and ACC accumulators whose purpose is to add up the correlation coefficients, to keep in memory the maximum and compare each time the sum of coefficients to the maximum previously recorded to update it and, at the end of the cycle on an image, define the coordinates (two dimensions and angle of rotation) corresponding to the maximum correlation recorded. are provided trace at the interconnection between ACC accumulators, LAT lock devices associated with the position meters and the JNT interface adapter connected to the data bus of the computer.

 While a two-dimensional displacement correlation cycle is performed on an image of order n, the window taken from the n-order image by the registers RD21, 22 the buffers B21, 22, and stored in RAM by the wup calculator, is processed angularly and sent to the correlators CN4 via the bus, the buffers B31, 32, the registers RD31, 32 while the new n + I order image arriving sequentially in the RD11 registers at RDIN is correlated with the order window n.

 In the foregoing, the correlation has been carried out pixel by pixel as the sequential arrival of the image data in each shift register RDI1 to RADIN, line by line, as the data pass is transmitted. image of a shift register RD11, 12, 13, etc. to the following and angle angle as the angular offsets and the resumption of the relative scanning of the whole image by the window . It is obvious that one can envisage a different exploration, for example, pixel by pixel, angle by angle1 line by line, which would be possible with a device of the same type as that of the figure 3. The device used will have in all the cases include the elements necessary to ensure the stepwise rotation of the window, and its translation into signals that can be correlated in CN4 with the image signals from RD1.

 Correlation operations can also be simplified and shortened. Indeed, seeking a maximum of correlation, it can be known only after relative exploration of the entire image through the window when the maximum recorded can be updated several times during the same operating cycle1 that is to say, during the cycle of successive correlations of the image of order n by the n-1 order.

 However, if the duration of such a cycle is short in time with respect to the quantities involved in mechanics, it is likely that, except in case of accident, the speed variations from one cycle to the next are low and that the coordinates (two-dimensional x, y and angular @) of the nl-order entry in the n-order image differ little from those of n-order window n in the n + 1 order image It is therefore possible to reduce the field of investigation, that is to say the extent of sweeps (two-dimensional x, y and angular w) between images and thus successive scanning cycles. consider scans starting with the previous (bi-dinalional x, y and annular o) position of the window and the immediate neighborhood (x = # x, y4Ay,; x + 2bx; y.2 # y; (@ 2 # @; etc ..., for example).

 The results from the interface lliT and fed to the computer by the data bus allow starting from the successive values of the coordinates x, y, o, to calculate the displacements between successive images and thus the linear velocities V, V and angular # and linear, gamma, gamma, and angular accelerations. Accurate knowledge of movements and speeds allows the computer to control all devices users of these data: anti-lock, anti-slip, anti-skating (anti-skating), mapping, setting and / or speed limitation, fuel consumption, etc. ..

 Although the devices according to the invention are essential for their application to motor vehicles, they are obviously adaptable to other types of vehicles, since between images taken successively, the correlation is possible without interference. As has been noted above, on motor vehicles, the distance between the camera lens and the ground used as reference object varies only because of the unevenness of the road and the movement of the suspension, which remains relatively limited on a modern road; the choice of the depth of field of the camera solves in practice the problem that might arise from this fact.

 In cases where this distance can vary in an appropriate manner either because of particular suspensions or because it is not a terrestrial vehicle, the devices according to the invention can be used if the computer can have the measurement the distance between the lens and the reference element (infrared measurement, ultrasound ...) and thus make the necessary correction at the level of the optics of the camera (this is actually only a development by an automatic automatic device).

 On id figure 3, we use a single camera providing both the n-1 order window and the n order image. As was emphasized above, for vehicles with very high speeds for which the field of a single cambra would be insufficient to allow the correlation between successive images can be used several cameras, for example two feeding separately in information registers A. RDI image registration offset and RD2 window definition.

Claims (9)

 1.- Device for determining displacements and / or speeds and / or accelerations according to two dimensions and in rotation of a vehicle a. from images taken successively by at least one camera which may be a charge transfer device, mounted on said vehicle, said determination of displacements and / or speeds being performed by correlation between at least part of the signals of an image and at least a part of the signals of a next image 1 characterized by the fact that the camera image signals (CAM) are sent on the one hand for setting RAM memory of a calculator unit (uP) via shift registers (RD21, 22) and buffers (B21, 22) taking from the image signals those corresponding to a window and, secondly, a set of shift registers (RDI) connected in series, the computer having means for stepwise rotation of the window and transmission of corresponding signals which are sent via buffers (B31 32) and shift registers (RD31, 32) to a set of digital correlators series of lines (CN4) in series receiving, by decliners, the image signals transmitted by the line shift registers (RD1), the outputs of the digital correlators (CN4) being connected to the computer via device for determining the maximum of the sum of the correlation coefficients.  2.- Device according to claim 1, characterized in that the camera image signals (CAM) are I lines and J pixels, the set of shift registers (RD1) receiving these signals sequentially consisting of N line registers (RD11 to RDIN) of M pixels with N - I and M J.  3.- Device according to claim 2, characterized in that the image signals (CAM) are received by the shift registers (RD21, 22) defining the window signals at K rows of L pixels with K < NIJ and L # MF J.  4.- Device according to claim 3, characterized in that the set of digital correlators (CN4) consists of K line digital correlators (CN41 to CN4K).  5.- Device according to claim 4, characterized in that an input of each K digital correlators (CN41 to CN4K) is connected to the output of K shift registers (RD11 to RD1K).  6.- Device according to one of claims i to 5, characterized in that the computer includes means ensuring the rotation of the between and the emission of the corresponding signals and controls the correlation operations for various angular orientations of the window .  7.- Device according to one of claims 1 3 6, characterized in that the window signals taken from an image are correlated with those of a subsequent image transmitted sequentially to the shift registers lines (RD11 a RDiN) and digital correlators (CN41 to CN4K).  8.- Device according to one of claims 1 to 7, characterized in that images taken are those of the ground, illuminated by laser.  9.- Device according to claim 9, characterized in that the ground is illuminated stroboscopically.