Method for directing an agent unit toward a target object

Description

[0001] The invention relates to a method for aligning an active agent unit to a target object, in which a target image of the target object is created, characteristic image features of the target object are determined from the target image and the active agent unit is aligned to the target object using the characteristic image features.

[0002] US 2011 / 063446 A1 discloses a system with multiple cameras for detecting, tracking, and recognizing characteristics of faces, license plates, etc. A wide-angle camera provides a video feed. The captured images allow the detection and tracking of objects of interest. Based on the location of the objects, a mirror device for aligning the field of view of a telephoto camera is aligned in turn with each detected and tracked object. Image processing algorithms such as SIFT can be used to detect an object in an image. US 2010 / 092033 A1 deals with a method for georeferencing a target between two subsystems of a targeting system.A target image is captured at the location of a transmitter subsystem, target descriptors are generated for a selected portion of the target image, and information about the target location and target descriptors is sent to a receiver subsystem. Based on this, the optical axis of a camera in the receiver subsystem is aligned, and a second portion of a target image captured by this camera is identified, which is correlated with the first portion of the target image from the transmitter subsystem. Publication DE 10 2010 020 798 A1 deals with preventing damage to friendly vehicles when fired upon with submunitions ejected from munitions such as artillery shells.

[0003] The paper "Feature-based tracking algorithms for imaging infrared anti-ship missiles" at GRAY GREER J ET AL: ,TECHNOLOGIES FOR OPTICAL COUNTERMEASURES VIII, SPIE, 1000 20TH ST. BELLINGHAM WA 98225-6705 USA, Vol. 8187, No. 1, October 6, 2011, deals with feature-based tracking algorithms, such as SIFT, for anti-ship missiles.

[0004] To disrupt or destroy elements of an aircraft, it is known to direct a high-energy laser at the aircraft, for example to disrupt an optical system or destroy components. To do this, the laser must track the movement of the aircraft in order to deposit the energy required to disrupt or destroy it at the desired location. To achieve a high energy input per area on the aircraft, the laser beam must be focused and directed at the aircraft. It thus irradiates only a small area on the aircraft, which it must also remain aligned with while tracking the movement in order to continuously deliver sufficient energy to this area.

[0005] It is an object of the invention to provide a method for aligning an active agent unit to a target object, with which the active agent unit can be precisely aligned to the target object.

[0006] This object is achieved by a method according to the features of patent claim 1, in which target key points in the target image are determined as characteristic image features, and in particular, at least one multidimensional target descriptor is determined for each target key point. The effective means unit can then be aligned to the target using the positions of the target key points. By determining the target key points in the target image of the target object, a position of the target object in the target image can be determined very precisely. Accordingly, the effective means unit can be aligned very precisely to a desired point on the target object.

[0007] The positions of the target keypoints may be directions in which the target keypoints are located relative to a reference direction, for example to a current orientation of the agent unit.

[0008] The active agent unit can be a laser unit, a gun, a missile with an active agent part, or the like, and expediently comprises an optical system through which the target image is recorded. It also expediently comprises an active agent transmitter through which the energy to be deposited on the target object is emitted. The active agent transmitter comprises an optical system through which, for example, a laser beam is directed at the target object. The recording optical system for recording the target image and the transmitting optical system can be the same optical system, so that the recording beams and the emitted beams are guided through the same optical elements. It is also possible to separate the two optical systems from one another, particularly in their movement relative to one another, so that, for example, the transmitting optical system can be guided very precisely to a movement of the target object, while the recording optical system can be moved independently of this.Furthermore, the active agent unit advantageously comprises a control unit for controlling elements of the active agent unit, for example an alignment drive for aligning the active agent transmitter to the target object and expediently for determining the characteristic image features, i.e. the key points in the target image and in particular the multidimensional target descriptors.

[0009] The target keypoints are determined from the target image using a feature detector. A feature detector is capable of finding points in an image that characterize the image. A good feature detector is capable of finding the same target keypoints in re-acquired images of the target object, even when the captured target object is scaled, rotated, geometrically transformed, or subjected to lighting changes.

[0010] A number of suitable feature detectors are known, such as the SIFT feature detector (Scale-Invariant Feature Transform), the SURF feature detector (Speeded Up Robust Features), the ORB feature detector (Oriented Fast and Rotated Brief), and other methods. The SIFT method and the SIFT feature detector are described, for example, in Lowe, David G.: Object recognition from local scale-invariant features; Computer vision, 1999; The proceedings of the seventh IEEE international conference on Vol. 2 leee, 1999, pp. 1150-1157, and in Lowe, David G.: Distinctive image features from scale-invariant keypoints; International journal of computer vision 60 (2004), No. 2, pp. 91-110. The SURF method with the SURF feature detector is described, for example, in Bay, Herbert; Tuytelaars, Tinne; Van Gool, Luc: Surf: Speeded up robust features; Computer Vision-ECCV 2006. Springer, 2006, pp. 404-417.The ORB method with the ORB feature detector is described, for example, in Rublee, Ethan; Rabaud, Vincent; Konolige, Kurt; and Bradski, Gary: ORB: an efficient alternative to SIFT or SURF. In: Computer Vision (ICCV), 2011 IEEE International Conference on IEEE, 2011, pp. 2564-2571. Other well-known methods for feature discovery and automatic feature comparison include Laplacian of Gaussian (LoG) and Normalized Cross-Correlation Function (NCCF). However, other methods that appear suitable for automatically comparing key points or descriptors can also be used.

[0011] Before determining the keypoints, the target image from which the keypoints are to be obtained can be smoothed with a Gaussian filter to remove image noise. After determining the keypoints, they can be examined for robustness in a further process step, whereby ultimately only those keypoints that meet specified stability requirements are selected. Robust keypoints, for example, are those that are insensitive to perspective distortion and / or rotation. Suitable for this purpose are keypoints whose image properties deviate from their background. Rotational stability can be achieved if a keypoint is assigned one or more orientations based on the orientations of the surrounding image gradients.Future calculations or comparisons can now be performed on the basis of image data that have been transformed relative to the assigned orientations, and in particular scaling and positioning, in order to ensure invariance for these transformations.

[0012] The descriptors are then determined from the keypoints thus found. For this purpose, the image gradients surrounding the keypoint are determined, which can be summarized in a multidimensional feature vector. This feature vector is also advantageously robust against changes in illumination, distortion, rotation, and the like. In addition to its image features at the keypoint, a descriptor also advantageously contains the coordinates of its keypoint, so that the keypoint is included in the descriptor.

[0013] A subsequent comparison of the target descriptors from the target image with previously determined reference descriptors is carried out using a suitable algorithm, for example the Random Sample Consensus Algorithm (RANSAC) or the FLANN matcher. Generally speaking, the distance function, which determines the vector distance, i.e., the directional difference between two vectors, can be used to compare descriptors. However, this can lead to ambiguous results in images in which certain objects are repeated. In practice, a distance quotient can therefore be used, whereby the best match, i.e., the smallest distance, is divided by the second-best match. Very similar results can then be assigned a lower probability that the found match is correct. The SIFT method can use kd trees or other search trees.The Hamming distance of the descriptor bit strings can be used for the letter feature detector from the ORB method. To optimize the nearest neighbor search, locality-sensitive hedging can also be used, so that similar bit strings are hegged in such a way that they are highly likely to end up in the same baskets.

[0014] The SIFT feature detector is described in more detail below as an example. Blurred copies of the image to be analyzed are created with varying degrees of blurriness. This is done by repeatedly applying a Gaussian filter to the image with increasing strength, or by repeatedly applying the same Gaussian filter to the last filtered image. Images adjacent to each other on the blur scale are then subtracted from each other, a process known as the Difference of Gaussian (DoG). The blurred copies of the image with varying degrees of blurriness and the DoG images together form an octave. Since the Gaussian filter significantly changes high-contrast regions in the image, subtracting the images results in the location of locations where the images have changed significantly – i.e., high-contrast regions such as edges and corners.The keypoints can now be found as suggested in the above-mentioned publication by Lowe from 2004. Keypoints that lie within a surface and whose gradients are therefore small are eliminated. If the keypoint lies on an edge, the gradient perpendicular to the edge is large, and the gradient parallel to the edge is small. Such keypoints are also removed. A keypoint lying on a corner contains two orthogonal and large gradients. Such keypoints are particularly robust and should be retained.

[0015] Since only sharp contours are found in the source image, but not significantly softer contours, a next higher octave is now formed - or even before the key points are calculated - by, for example, only using every second pixel of the blurriest image or by pixelating the source image. In a next higher octave or image octave, the image on which the descriptors are based has a lower resolution than the image of the lower octave. The lower resolution can be created, for example, by combining a group of pixels, in particular 2 pixels or nxn pixels, for example 2 x 2 pixels, into one pixel, e.g. by addition, averaging or removing just one pixel from the group. The key points are now determined in this second octave.In addition to their x and y coordinates, the keypoints also include a scale factor, i.e., the octave from which they originate. In this way, a large number of keypoints are assigned to the image of the target object. The descriptors assigned to the keypoints or containing the keypoints therefore conveniently also include the octave, i.e., the scale factor.

[0016] In order to align the weapon unit to the target object, the target object must be recognized as such. There are two ways to do this. The target object, an identical object, or a similar object was recorded at an earlier point in time, and a reference image of the target object was created. The recording unit can be the weapon unit or a different unit. Reference keypoints and multidimensional reference descriptors were determined from the reference image. If the target object is to be recognized as such, the target descriptors from the target image can be compared with the reference descriptors from the reference image. If a comparison or correlation is above a predetermined threshold, the target object is considered to be recognized. In this way, automatic object recognition can be carried out.

[0017] The reference image will be stored in a database. This database will contain multiple images depicting the target object, or an identical or similar object, from multiple perspectives. Images of other objects that are dissimilar to the target object can also be stored in the database. Generally, it is sufficient to store the reference descriptors instead of the images themselves, so that in this context, a saved image can only be understood as the descriptors determined from them and, if necessary, key points, if these are not already part of the descriptors. If images of several dissimilar objects are present in the database, the target object can be automatically identified by comparing the images or descriptors.

[0018] Another option for identifying the target object as such is to use a previous target image as a reference image. For example, an operator takes a target image in which the target object is visible. The image can be captured using the active agent unit. The target object is then marked in this target image by the operator or automatically. This can be done by placing a target area around the target object in the target image. Reference descriptors are then determined and saved from this target image or target area. To reduce computing time, it is advantageous to select a target area of ​​the target image in which the target object is at least partially depicted, with the reference descriptors being extracted exclusively from the selected area. Image areas not required for the correlation are ignored, so that the correlation can be accelerated.

[0019] To ensure reliable comparability of the target descriptors with the reference descriptors, it is advisable for the target descriptors to be obtained using the same method as the reference descriptors. When comparing descriptors, a correlation or comparison of two descriptors is considered successful if the similarity of the descriptors exceeds a predetermined comparison value. In the target image, the keypoint that led to a successful correlation with the keypoint of the reference image can now be marked as the correlation point. The correlation point is identical to its keypoint in the target image and is assigned a keypoint in the reference image, for which data can be transferred from its reference descriptor.

[0020] Individual correlation points in the target image may also be incorrect, so that similar keypoints in the target image and reference image coincidentally occur, leading to a successful point correlation. Therefore, the target object should ideally only be classified as recognized once a sufficient number of correlation points have been found in the target image. The number can be predetermined, for example, in a predetermined relationship with image parameters of the target image. Only once the target image has been successfully correlated with the reference image is the target object considered recognized.

[0021] In this respect, it is advantageous for aligning the agent unit to the target object if the target descriptors are compared with reference descriptors obtained from a reference image and the target object is classified as detected if the comparison result is above a threshold value.

[0022] In a further advantageous embodiment of the invention, a sequence of target images is created, from which the target key points are determined. The target key points can then be tracked in the sequence of target images, and the missile unit can be adjusted according to the movement of the target key points, for example, through an image area or through a landscape. While the missile unit is aligned with the target object, new target images are expediently continuously recorded and correlated with the reference image(s). The target images are recorded, for example, at a fixed time interval, such as at regular intervals, or depending on the flight or movement parameters of the target object.Each of the recorded target images is correlated with the reference image or the target images to be used for correlation are selected from the totality of the recorded target images according to a predetermined algorithm, for example every nth image, where n is an integer.

[0023] Particularly when depositing laser energy in the target object, it is advantageous to specifically irradiate a weak structure that can be influenced by the laser energy. In this case, it may happen that a target structure, referred to below as the target point, is poor in characteristic image features, so that it cannot be easily detected in the target image. The target point can therefore be free of a key point. In order to nevertheless be able to deposit sufficient energy at a desired target point on the target object, the invention provides that the target point on the target object is determined in its position relative to target key points and that the active agent unit is aligned to the target point.

[0024] The target point can be set, for example, by an operator or automatically. In this case, the target point is expediently set relative to key points in the reference image, in particular relative to their position. Such reference key points can be specially marked. If a successful correlation is achieved with a sufficient number of specially marked reference key points, the position of the target point can also be determined in the target image, since each reference key point is also assigned a target key point. The active agent unit can then be precisely aligned to the target point so that, for example, a beneficial effect can be achieved there. The position of the target point can be defined as coordinates of the point, a spatial position in the recorded target image and / or a direction of the point, absolute or relative, for example to an alignment of an observing or recording optics.

[0025] The target point is defined relative to reference keypoints of a reference image. The target keypoints in the target image are assigned to the reference keypoints based on a comparison of target descriptors with reference descriptors, and the target point is defined relative to the target keypoints. The reference keypoints used, and in particular the target keypoints, are marked as connected to or assigned to the target point.

[0026] The missile unit may include an optical system that is aligned with the target object. The optical system may be a seeker optical system of a missile that is steered toward the target using the optical system's alignment. The missile's flight movement can be tracked by the optical system, and the missile can track the target object. The optical system is a laser optical system, and a laser beam is directed at the target object by an optical system transmitter. Laser energy can be selectively deposited on the target object, thereby altering the target object.

[0027] Particularly with fast-moving target objects, the computing time required to compare the descriptors can result in the alignment of the agent unit being too slow, and the target object having already moved some distance in the direction of movement by the time alignment has been completed. The agent energy is then deposited imprecisely or incorrectly. To avoid this, it is advantageous to calculate a future position or direction of a target point on the target object from the target object's previous movement. This can be achieved, for example, through interpolation, which infers a future movement speed and direction from a past movement speed and direction.The movement of the target object, i.e. its speed and direction of movement, can be determined from a movement of the target object in a frame of several target images and / or a frame reading from an optics tracking the target object.

[0028] Conveniently, a future position of the target point is determined using a Kalman filter. The Kalman filter can be used, for example, as a tracking filter to estimate the motion parameters of the observed object, in this case the target object.

[0029] Conveniently, future positions of the target keypoints detected in target images are determined using a Kalman filter. Advantageously, a future position of the target point is determined from these future positions. By determining the multiple positions of the target keypoints, this computation method is less error-prone than directly extrapolating the target point position using the Kalman filter.

[0030] If the target object is a passing vehicle, the perspective of the passing target object rotates. It is therefore possible that the current target image shows the target object from a different perspective than the reference image. Even if the comparison of descriptors is rotation-invariant to a certain extent, perspective changes of more than 45° can lead to misidentifications, meaning that the target object is no longer correctly recognized. To avoid this, a further advantageous embodiment of the invention proposes that reference descriptors be present that were obtained from multiple reference images of the target object taken from different directions.For example, the reference descriptors were obtained from a reference image of the target object from the front, a reference image of the target object viewed obliquely from the front, from the side, obliquely from the rear, and from the rear—i.e., from five reference images showing the target object from different perspectives. When changing perspectives, the descriptor comparison will always be successful with reference descriptors from at least one reference image. It is advisable to use the reference key points, in particular exclusively the reference key points, of the reference image that provides the best comparison result to align the weapon unit.

[0031] The more reference images or their reference descriptors are available, the longer the computing time for the descriptor comparison. To keep computing time low, it is proposed that the multiple reference images be assigned an image sequence and that only a maximum of two reference images or the reference descriptors derived from them be used for the descriptor comparison. The two reference images are located next to each other in the image sequence. It is advisable to use exclusively the reference image with the previously best correlation result and the next reference image or its descriptors in the image sequence.

[0032] Furthermore, the invention is directed to an active agent unit according to the features of claim 9 with an optics, an active agent transmitter, an alignment drive and a control unit which is prepared to determine characteristic image features of the target object from a target image of the target object and to control the alignment drive using the characteristic image features so that the active agent transmitter is aligned with the target object.

[0033] According to the invention, the control unit is configured to determine target key points in the target image as characteristic image features and to determine at least one multidimensional target descriptor for each target key point. Furthermore, the control unit is configured to align the active agent transmitter to the target object using the positions of the target key points, to determine a target point on the target object relative to the target key points, and to align the active agent transmitter to the target point.

[0034] They show: FIG 1 shows an unmanned aerial vehicle being targeted by a weapon unit, FIG 2 shows a target image of the target object created by the weapon unit, which is correlated with reference images of the target object, and FIG 3 shows a schematic representation of five reference images with images of the target object from different perspectives.

[0035] FIG 1 shows a target object 2 in the form of an unmanned aerial vehicle flying obliquely over a vehicle 4 with an active agent unit 6. The vehicle 4 can be a watercraft or a land vehicle. The active agent unit 6 is equipped with an active agent transmitter 8, which can be pivoted two-dimensionally with the aid of an alignment drive 10. A camera 12 can also be pivoted two-dimensionally via the alignment drive 10 and, in particular, synchronously with the active agent transmitter 8. The active agent transmitter 8 is a laser unit for emitting a laser beam. Both the active agent transmitter 8 and the camera 12 each have an optics 14 and 16, respectively, wherein the optics 14 are designed to focus the laser beam on a target point 18 and the optics 16 serve as camera optics for recording an image of the target object 2.

[0036] A control unit 20 of the active agent unit 6 serves to control the alignment of the active agent unit 6 or the active agent transmitter 8 and the camera 12, to take images of the target object 2 with the aid of the camera 12 and to control the emission of a laser beam with the aid of the active agent transmitter 8.

[0037] As the target object 2 flies past the vehicle 4, the target object 2 is to be destroyed by the missile unit 6 at the target point 18 or impaired there in such a way that further flight is prevented. The target point is a mechanically unstable point of the target object 18, at which the effect of laser radiation can cause mechanical destruction. To align the missile unit 6 to the target point 18, the target object 2 is initially recorded by the camera 12, and a target image 22 is created, which is FIG 2 is shown as an example.

[0038] FIG 2 shows a target image 22, which is compared with several reference images 24. The reference images 24 can also have been recorded by the effective means unit 6 or by another recording unit. In the former case, the reference image 24 is recorded by the effective means unit 6, and an operator marks an image area in this image in which the target object 2 is depicted. He then specifies the target point 18, to which the laser is to be aligned later. The target point 18 can be specified, for example, by clicking on an image point on a display device on which the reference image 24 is displayed.

[0039] In the latter case, the reference images 24 were previously acquired and are stored in a database. Before or during the approach of the target object 2, the reference images 24 or characteristic image features extracted from the reference images are fed to the missile unit 6 and stored there.

[0040] Regardless of the method used to create the reference images 24, characteristic image features are extracted from them in the form of reference keypoints and reference descriptors. These are obtained from the reference images 24 using a SIFT feature detector or a SURF feature detector. Accordingly, the reference descriptors are available as multidimensional feature vectors, as obtained using the two aforementioned methods. For image comparison, it is sufficient if only the reference descriptors are present instead of the reference images 24, whereby these descriptors can contain the reference keypoints as data. The target image 22 is processed in the same way, and target descriptors are extracted that contain target keypoints in the target image 22.

[0041] In order to track the passing target object 2, the target descriptors of the target image 22 are compared with the reference descriptors from the reference images 24. If the result of a comparison of a target descriptor with a reference descriptor is above a threshold value, the comparison is successful and the respective target key point in the target image 22 and / or the respective reference key point in the reference image 24 are marked as correlation points. FIG 2 Correlation points associated with each other in images 22 and 24 are connected by straight lines. These lines merely serve to visualize key points associated with each other. If the correlation of the images is successful, i.e., if a sufficient number of correlation points were found in the target image 22, or if more correlation points than a threshold value were found, with the threshold value being predetermined, then the correlation is successful and the target object 2 can be classified as recognized.

[0042] Some or all of the correlation keypoints have a positional relationship to target point 18. They are marked accordingly. Due to its lack of image features, target point 18 itself is not characterized by a keypoint, i.e., it is keypoint-free. By assigning target keypoints to reference keypoints, this marking and the corresponding position information can also be assigned to the relevant target keypoints in target image 22. Using the resulting position information, target point 18 can be precisely located in target image 22, even if it is not recognizable from within.

[0043] The orientation of the camera 12 is known, either in space or relative to the active agent transmitter 8. Accordingly, the direction of the target point 18 is also known, and the active agent transmitter 8 can be aligned to the target point 18 on the target object 2. The target point 18 is irradiated with the laser beam to achieve the desired result.

[0044] As described above as an example for the SIFT method, image octaves O1, O2, O3, ..., hereinafter referred to as Oi, are formed to form the reference image 24. The reference image 24 forms the first image octave O1, and lower-resolution images of the target object 2 form higher image octaves O2, O3, ... of the reference image 24. The reference key points in the higher octaves mark pixels with softer image structures.

[0045] In FIG 2 Three image octaves O1, O2, O3 are shown. However, it is possible and advantageous to create more image octaves and to extract the reference descriptors from each of them. The same procedure can be used with the target image 22, so that image octaves and associated target descriptors are also created for this. However, for the sake of clarity, this is not shown in FIG 2 not shown.

[0046] The reference descriptors D1i are extracted from the first image octave O1, the reference descriptors D2i from the second image octave O2, and so on. All target descriptors can now be compared with all reference descriptors Dii to achieve a successful correlation.

[0047] During the flyby of the target object 2 past the active agent unit 6, the perspective of the view of the active agent unit 6 on the target object 2 changes continuously. In this case, the image of the target object 2 in the target image 22 is rotated to the image of the target object 2 in the reference image 24. In the Figuren 1 und 2 In the embodiment shown, target object 2 can be seen obliquely from the front and slightly from below. Once target object 2 has flown past, it will be seen from behind. A correlation of the image of the target object from behind with an image of target object 2 from the front will generally not lead to a successful correlation.

[0048] In order to be able to track the target object 2 for a long time, the database contains reference images 24 or reference descriptors Dii extracted from them for several views of the target object 2. This is exemplified in FIG 3 shown.

[0049] FIG 3 schematically shows five reference images 24 with different views A, B, C, D and E of the target object 2. View A is from the front, view B is diagonally from the front, view C is from the side, view D is diagonally from the back and view E is exactly from the back. All views are taken from slightly below, so that the FIG 1 The view shown is approximately view B. In the FIG 2 View B shown is the one shown in FIG 3 shown view B. The octaves B 2 , B 3 , ... do not have to be represented as images, but can be formed as data from which the reference descriptors Dii are formed. In FIG 2 The octaves are shown as pictures only for the sake of clarity. Accordingly, picture C 1 is FIG 2 the representation C from FIG 3 .

[0050] It is now possible to correlate the target descriptors with all reference descriptors Dii from all representations AE. However, this involves considerable computational effort and takes a certain amount of time. To reduce the computation time, the following procedure can be used.

[0051] The first target image 22 can be correlated with all images, or all target descriptors can be correlated with the reference descriptors of all images AE. It is also possible to compare the target descriptors with reference descriptors of several dissimilar objects in order to automatically determine which target object 2 is currently being addressed. In this way, a target object identification can initially be performed.

[0052] In the illustrated embodiment, the successful correlation will only be achieved with one or two of the images AE, so that it is clear from which perspective the target object 2 is currently seen. This first correlation phase is in FIG 3 marked with I. For example, it is assumed that the correlation with reference descriptors from the reference image 24 of view B was successful.

[0053] The views AE are given an order as in FIG 3 The correlation of a subsequent target image 22, which is intended for correlation, is now no longer carried out with all images of all representations AE, but only with a group of reference images 24. This group contains the reference image 24 in which the last successful correlation took place. Furthermore, it is expedient if the group contains the subsequent image, in this embodiment the reference image 24 of perspective C. This is shown in FIG 3 indicated by the correlation phase II.

[0054] During the flyby, the perspective of the view of the target object will now change, so that the correlation will be most successful after a while with reference descriptors of the target image 24 of view C. Now, the group of used reference images 24 can be changed, so that, for example, the correlation image, i.e., the reference image 24 with the last successful correlation, is used again, and the next following image is used for future correlations, i.e., correlations with future target images 22. This correlation phase in FIG 3 marked III. Finally, target object 2 is visible from the rear, and correlation phase IV begins. If the successful correlation is in reference image 24 of view E, it is sufficient to correlate only with this reference image 24.

[0055] It is conceivable that the target object 2 does not fly in a straight line, but moves in arcs, circles, or other geometries. In this case, it makes sense to populate the groups differently. For this purpose, the active means unit 6 comprises an input device on which an operator can specify a movement mode of the target object 2. The group of reference images 24 to be correlated is selected according to an operator input. It is also possible for the control unit 20 to independently recognize a movement pattern of the target object 2 and populate the groups according to the movement pattern.

[0056] Under certain circumstances, the target object 2 moves relatively quickly, so that despite a short computing time, the movement of the target point 18 within the computing time is so large that the alignment of the active agent unit 6 with respect to the target object 2 or the target point 18 undesirably lags behind. To avoid this, a future position of the target object 2 or the target point 18 can be calculated from previous successful correlations.

[0057] For this purpose, the movement of correlation keypoints in the target image 22 is calculated using a Kalman filter to one or more future positions of the corresponding correlation keypoints in the target image 22. From a set of future positions at a future point in time, the future position of the target point 18 at this point in time is determined based on the position information of the correlation keypoints relative to the target point 18. Accordingly, the weapon unit 6 or the weapon transmitter 8 is aligned in the calculated direction of the target point 18 at this point in time. List of reference symbols

[0058] 2Target object 4Vehicle 6Aid unit 8Aid transmitter 10Alignment drive 12Camera 14Optics 16Optics 18Aim point 20Control unit 22Aim point 24Reference image A-EPerspectives DReference descriptors OImage octaves

Claims

1. Method for aligning an effector emitter (8) of an effector unit (6) with a target object (2), by means of which effector emitter energy to be deposited at the target object (2) can be emitted, wherein a target image (22) of the target object (2) is created, characteristic image features of the target object (2) are established from the target image (22) and using the characteristic image features the effector emitter (8) of the effector unit (6) is aligned with the target object (2) by means of an alignment drive (10) controlled by a control unit (20) of the effector unit (6), wherein - the effector emitter (8) comprises optics (14) by means of which a laser beam is directed onto the target object (2), - target key points are established in the target image (22) as characteristic image features and at least one multi-dimensional target descriptor is established for each target key point, and the effector emitter (8) is aligned with the target object (2) using the positions of the target key points, - a target point (18) on the target object (2) is established relative to target key points and the effector emitter (8) is aligned with the target point (18), wherein the target point (18) is set relative to reference key points of a reference image (24) of the target object (2), target key points in the target image (22) are associated with the reference key points on the basis of a comparison between target descriptors and reference descriptors, and the target point (18) is set relative to the target key points, and - reference descriptors which were obtained from a plurality of reference images (24) of the target object (2) recorded from different directions are available, and the reference key points of the reference image (24) supplying the best comparison result are used for aligning the effector emitter (8) of the effector unit (6).

2. Method according to Claim 1, wherein the target descriptors are compared with reference descriptors obtained from a reference image (24), the target object (2) is classified as identified in the case of a comparison result above a threshold, and the effector unit (6) is aligned with the target object (2).

3. Method according to Claim 1 or 2, wherein a sequence of target images (22) is created, from which the target key points are established in each case, the target key points are tracked in the sequence of the target images (22) and the effector unit (6) is repositioned in accordance with a movement of the target key points through a landscape.

4. Method according to any of the preceding claims, wherein a future position of a target point (18) at the target object (2) is calculated from the movement of the target object (2) that has already taken place.

5. Method according to Claim 4, wherein the future position of the target point (18) is determined with the aid of a Kalman filter.

6. Method according to Claim 4 or 5, wherein future positions of the target key points identified in target images (22) are determined with the aid of a Kalman filter, and a future position of the target point (18) is determined from these future positions.

7. Method according to any of the preceding claims, wherein an image sequence (A-E) is assigned to the plurality of reference images (24), and the reference descriptors of at most two reference images (24) are used for comparison between target descriptors and reference descriptors.

8. Method according to Claim 7, wherein only the reference image (24) with the best correlation result in the correlation that took place immediately therebefore and the next reference image (24) in the image sequence (A-E) are used for the descriptor comparison.

9. Effector unit (6) having optics (16), an effector emitter (8), by means of which energy to be deposited at a target object (2) can be emitted and which comprises optics (14) by means of which a laser beam can be directed onto the target object (2), an alignment drive (10) and a control unit (20), which is prepared to establish characteristic image features of the target object (2) from a target image (22) of a target object (2) and control the alignment drive (10) using the characteristic image features in such a way that the effector emitter (8) is aligned with the target object (2), wherein the control unit (20) is prepared - to establish target key points in the target image (22) as characteristic image features and at least one multi-dimensional target descriptor for each target key point, and to align the effector emitter (8) with the target object (2) using the positions of the target key points, - to establish a target point (18) on the target object (2) relative to target key points and to align the effector emitter (8) of the effector unit (6) with the target point (18), wherein the target point (18) is set relative to reference key points of a reference image (24) of the target object (2), target key points in the target image (22) are associated with the reference key points on the basis of a comparison between target descriptors and reference descriptors, and the target point (18) is set relative to the target key points, and - from a plurality of reference images (24) of the target object (2) recorded from different directions, from which reference descriptors were obtained and are available, to use the reference key points of the reference image (24) supplying the best comparison result for aligning the effector emitter (8) of the effector unit (6).