Method for tracking at least one object and monitoring device

Abstract

Method for tracking at least one object (5) by means of at least one trackable imaging sensor (2) with which the at least one object (5) is detected, wherein: - several comparison patterns (8.1-8.6), each exhibiting at least one object feature (9.1-9.6) of the at least one object (5), are stored in chronological order in order to perform a recognition of the at least one object (5) in one or more subsequent images (4.1-4.6) captured by the at least one imaging sensor (2) based on at least one to be determined correspondence value of the at least one stored object feature (9.1-9.6) of the at least one object (5) by at least one of the comparison patterns (8.1-8.6) with newly captured object features (9.1-9.6), and wherein - the at least one imaging sensor (2) is tracked by the at least one object (5) and a movement profile (10.1, 10.2) of the at least one object (5) is stored, characterized in that - when recognizing the at least one object (5), if several comparison patterns (8.1-8.6) are stored, at least one older comparison pattern (8.1-8.6) is taken into account to determine the at least one match value at least to the same extent as newer comparison patterns (8.1-8.6) of the at least one object (5), wherein the oldest stored comparison pattern (8.1-8.6) is used to determine the at least one match value in which the at least one stored object feature (9.1-9.6) has at least a predetermined minimum match value with the newly recorded object features (9.1-9.6) of the at least one object (5) and / or that - when tracking the at least one imaging sensor (2), if the at least one object (5) is no longer recognized in a subsequently acquired image (4.1-4.6) of the at least one imaging sensor (2), an estimation of the further movement (11.1, 11.2) of the at least one object (5) is carried out, in which older stored movements (12.1, 12.2) of the at least one object (5) are taken into account at least to the same extent as newer stored movements (13.1, 13.2) of the at least one object (5), whereby it is determined whether the movement pattern (10.1, 10.2) of the at least one object (5) is diffuse or directed.

Description

[0001] The invention relates to a method for tracking at least one object by means of at least one trackable imaging sensor, with which the at least one object is detected. The invention also relates to a monitoring device with at least one imaging sensor.

[0002] Tracking, particularly video tracking, refers to the monitoring of objects using image processing through the automatic repositioning of an imaging sensor or camera system. The tracked object can then be displayed to a user or a higher-level system. This involves recognizing a manually or automatically defined object, determining its position, and moving the capturing camera so that the object is displayed as close as possible to the center of the image and remains there. The image sensor or image acquisition system can consist of one or more cameras in the visual and / or thermal imaging range, such as long-wave infrared (LWIR), medium-wave infrared (MWIR), very long-wave infrared (VLWIR), far-infrared (FIR), short-wave infrared (SWIR), and near-infrared (NIR), as well as in the ultraviolet (UV) range. To enable object recognition, object representations in the image and / or object features are analyzed.Object properties are determined at the beginning, stored as comparison patterns or so-called templates, and an attempt is made to find objects with this appearance or these characteristics in subsequent images. This can be achieved, for example, by storing a section of an image, which is then compared with corresponding sections in the following images, particularly using cross-correlation or similar methods, whereby the position with the maximum match is determined. This can also be done using other known object features, such as centroid, boundary, phase, edge analysis, or histograms (e.g., HoG / Histogram of Oriented Gradients).

[0003] However, all methods must take into account the fact that objects can change their appearance and thus their characteristics or properties over time. This occurs, for example, simply through the rotation of an object (e.g., a vehicle), which significantly alters its appearance in relation to the camera system. One known approach is to adapt the stored object properties to the changed appearance by storing the newly captured features of a previously recognized object as starting points for subsequent searches in later images. This essentially retrains the tracker. However, this becomes problematic when the object's properties change because other objects are nearby or at least partially obscure it.For example, if two different objects are moving towards each other, or if the object being tracked moves behind a stationary object such as a tree, building, or vehicle, the attributes of the tracked object become mixed with those of the encountering object when the objects meet. After the encounter, it can easily happen that the tracker continues to track the wrong object. This is a frequently observed negative effect in existing tracking systems.

[0004] As soon as a tracked object is no longer visible in the image, for example due to rapid, temporary occlusion, a decision must be made on how the automatic camera tracking should proceed. Known systems either remain stationary or continue moving at the last set speed and direction. However, this can lead to an incorrect direction of movement, especially if the tracked object was previously moving diffusely, potentially causing the camera to leave the object's field of view entirely.

[0005] DE 103 02 306 A1 concerns object tracking with a swiveling camera.

[0006] DE 10 2006 001 033 A1 discloses a method and a device for detecting objects in the driving environment for a motor vehicle.

[0007] DE 11 2009 000 485 T5 generally shows video surveillance and, in particular, the comparison of objects depicted in multiple images.

[0008] US Patent 2010 / 0157089 A1 discloses a computer-implemented method and a camera system for determining the current spatial representation for recognition in a current image of an image sequence. The method derives an expected spatial representation for recognition based on at least one previous image, generates a spatial representation for recognition, and extends the spatial representation to obtain an extended spatial representation based on the expected spatial representation. The method determines a similarity measure between the extended spatial representation and the expected spatial representation and then determines the current spatial representation for recognition based on this similarity measure.

[0009] DE 10 2008 005 993 A1 relates to a method for tracking objects using a camera, wherein a motion model for the object is provided for tracking the object. Based on a sequence of images captured by the camera, a motion pattern of the camera's own movement is estimated, whereby defined parameters such as frequency or direction of the camera's own movement are determined from the object's position in the captured image sequence to describe this motion pattern. This estimated motion pattern of the camera's own movement is then integrated into the object's motion model in order to advantageously increase the accuracy of object tracking and to avoid time-consuming and computationally intensive post-processing such as subsequent image stabilization.

[0010] US Patent 2004 / 0100563 A1 relates to a video tracking system and method comprising a video camera with selectively adjustable pan, tilt, and focal length. A processor receives video images captured by the camera. The processor is programmed to recognize target objects in the images and selectively adjust the camera to track the target object. The camera is set to variable speeds, selected as a function of a property, such as the speed, of the target object. The camera's focal length is selectively adjusted depending on the distance of the target object from the camera.The images captured by the camera are geometrically transformed to align images with different fields of view, to facilitate image analysis and thereby enable continuous camera adjustment for generating video images with relatively smooth transition movements.

[0011] DE 37 40 789 A1 describes a method for tracking a moving object over time in a sequence of images and a device for carrying out this method.

[0012] The present invention is based on the objective of improving a method of the type mentioned at the outset and avoiding the disadvantages of the prior art, in particular, on the one hand, it is intended to prevent the wrong object from being tracked after an encounter between different objects, especially with partial occlusion, and on the other hand, to provide improved tracking behavior of the imaging sensor in the event of loss of the object, for example by rapid temporary occlusion.

[0013] This problem is solved according to the invention by a method for tracking at least one object by means of at least one trackable imaging sensor, with which the at least one object is detected, wherein: - one or more comparison patterns, each exhibiting at least one object feature of the at least one object, are stored in chronological order in order to perform a recognition of the at least one object in one or more subsequent images captured by the at least one imaging sensor based on at least one to be determined correspondence value of the at least one stored object feature of the at least one object, or at least one of the comparison patterns with newly captured object features, wherein - the at least one imaging sensor is tracked to follow at least one object and a movement history or trajectory of the at least one object is stored, wherein - when recognizing at least one object, if several comparison patterns are stored, at least one older comparison pattern is taken into account to determine the at least one match value, at least to the same extent as newer comparison patterns of the at least one object, wherein the oldest stored comparison pattern is used to determine the at least one match value where the at least one stored object feature has at least a predefined minimum match value with the newly recorded object features of the at least one object and / or wherein - when tracking the at least one imaging sensor, if the at least one object is no longer recognized in a subsequently acquired image of the at least one imaging sensor, older stored movements of the at least one object are taken into account at least to the same extent as newer stored movements of the at least one object, whereby it is determined whether the movement of the at least one object is diffuse or directed.

[0014] The tracking method according to the invention significantly improves object tracking when the object is at least partially obscured or lost. This is achieved by considering the history of object features, object representation, and / or movement patterns. Older stored object features are taken into account at least to the same extent as newer stored object features. This advantageously minimizes the risk of the tracker erroneously switching to a new object. The object features of a changing object are relearned as before, while the previous object features are retained, for example, in a list. Alternatively or additionally, the history of object movements can be considered when controlling the tracking of the imaging sensor during temporary loss of object recognition.Object movements can be captured and / or stored as motion vectors. For example, a determined long-term direction of movement, or in the case of a wobbly motion, a standstill, can be selected for tracking the imaging sensor. An older comparison pattern is stored earlier than a newer one. Thus, older and newer comparison patterns are related to each other in terms of the temporal sequence of their storage.

[0015] Comparison patterns can include so-called object templates.

[0016] According to the invention, it can further be provided that: - if the at least one match value to be determined is greater than or equal to a predefinable first limit, a reliable recognition of the at least one object has taken place, whereby no new comparison pattern of the at least one object is stored, and / or - if the at least one match value to be determined is greater than or equal to a predefinable second limit and less than a predefinable first limit, a new comparison pattern is stored using newly recorded object characteristics of the at least one object, and / or - if the match value to be determined is less than a predefinable second limit, then at least one object was no longer recognized.

[0017] The degree of agreement can be determined, for example, by cross-correlation at the pixel level. The first, predefined limit value can be set at an agreement value of approximately 10% to approximately 100%, in particular approximately 60% to approximately 90%, preferably approximately 88%. The second, predefined limit value can be set at an agreement value of approximately 10% to approximately 100%, in particular approximately 60% to approximately 90%, preferably approximately 85%.

[0018] It is advantageous if, when recognizing the at least one object to determine the at least one match value, at least one older comparison pattern is used preferentially over at least one newer comparison pattern of the at least one object.

[0019] For example, a minimum agreement value can be specified, using the oldest stored comparison pattern in which at least one stored object characteristic exhibits at least the specified minimum agreement value with the newly recorded object characteristics. Starting with the oldest comparison pattern, the agreement value of each subsequent stored comparison pattern can be determined chronologically, and then it is checked whether this value is greater than or equal to the specified minimum agreement value. This approach reduces computational effort. The minimum agreement value could, for example, correspond to the second threshold value.

[0020] Alternatively, the agreement values ​​of the at least one stored object attribute of all stored comparison patterns with the newly recorded object attributes can first be determined. This agreement can then be weighted accordingly using a factor, with older comparison patterns receiving a higher weighting. Finally, the stored comparison pattern in which the at least one stored object attribute has the highest weighted agreement value with the newly recorded object attributes is selected.

[0021] This allows older object features to be considered further. When searching for the object's position in subsequent images, older object features are now prioritized over newer ones. A corresponding weighting can also be applied. Thus, if two objects were to appear separately after an encounter, even if the tracker were to mistakenly relearn the object's appearance before the encounter, the original object's appearance would be prioritized, ensuring the correct object is tracked.

[0022] It is advantageous if the temporal distribution of the storage of comparison patterns of at least one object is uniform or logarithmic.

[0023] Especially when tracking an object over a longer period with the associated frequent changes, the number of comparison patterns or object attributes to be processed can increase significantly, thus greatly increasing the computational effort. Therefore, a selection of the object attributes to be stored can be made. Additionally, the maximum number of stored object attributes or comparison patterns can be limited. The temporal distribution of the stored object attributes can be uniform or logarithmic.

[0024] Comparison patterns or object characteristics that are no longer recognized within a defined period can be removed or deleted. Comparison patterns or object characteristics that have been recognized and / or tracked with a predetermined frequency over a longer period can be prioritized for storage.

[0025] It is advantageous if the first comparison pattern or the first defined object features are permanently stored to prevent the loss of an undisturbed object feature due to frequent random occlusions or encounters.

[0026] Object features of the at least one object can include appearance properties and / or representations of the at least one object.

[0027] The tracking of the at least one imaging sensor can be carried out in such a way that, taking into account the movement of the at least one object, it is always displayed in the area of ​​the center of the subsequently acquired images. The scanning camera is moved in such a way that the detected or re-detected object is displayed as close as possible to the center of the image and remains there during further observation.

[0028] When tracking the at least one imaging sensor, an estimate or prediction of the object's movement can be made if the object is no longer recognized in a subsequent image. As soon as a tracked object is no longer found in the image, for example, due to rapid, temporary occlusion, a decision should be made regarding how the camera tracking should behave. Here, too, the object's history or movements can be advantageously considered, since every detected movement can exhibit a certain degree of inaccuracy, caused, for example, by fluctuating movement of the object or the camera's position.

[0029] It is possible to determine whether the motion of at least one object is diffuse or directed. Estimating the motion of a previously diffuse movement (e.g., compensating for a wobbly motion) would quickly lead to leaving the previously defined object area without considering the history or distinguishing between "diffuse" and "directed" motion.

[0030] When tracking the at least one imaging sensor, if the at least one object is no longer recognized in a subsequently acquired image, the current position of the at least one imaging sensor can remain unchanged if a diffuse motion profile of the at least one object is detected. When diffuse motion is detected, the camera position can remain unchanged because this is the best estimate for the further motion profile of the object.

[0031] It is advantageous if, during the tracking of the at least one imaging sensor, when the at least one object is no longer recognized in a subsequently acquired image, the further movement of the at least one object is estimated by means of a, in particular modified or normalized, vector addition of a number of previously acquired and / or estimated movements of the at least one object, if a directed movement pattern of the at least one object is detected.

[0032] This also takes the history into account in the case of a directed movement pattern, leading to an increase in estimation accuracy.

[0033] The length of the estimated further movement of the at least one object can be determined by averaging or normalizing the individual lengths of the preceding detected and / or estimated movements of the at least one object used in the vector addition.

[0034] It is advantageous if a search area in the captured images of the at least one imaging sensor is expanded or enlarged over time, in particular stepwise, if the at least one object is no longer recognized in the subsequent captured images after a defined period of time.

[0035] Since all estimates are subject to a certain degree of uncertainty, the search area can be enlarged or expanded over time. This is because the deviation of the motion estimate from the object's actual behavior leads to a greater discrepancy between the expected position and the actual position over time. Furthermore, the probability that the object will change its movement also increases with time. After the object is detected again and any incorrect or inaccurate camera position is corrected, the search area can then be reduced back to its original size. If the lost object is not found again after a configurable period (e.g., 15 seconds), the automatic camera tracking can stop and the original area sizes are restored. The user or the automatic object recognition system can then assign a new object to the tracker.

[0036] Claim 14 specifies a monitoring device comprising at least one imaging sensor and at least one control unit electrically connected to the at least one imaging sensor.

[0037] Advantageous embodiments and further developments of the invention are specified in the dependent claims. An exemplary embodiment of the invention is described below with reference to the drawing.

[0038] They show: Fig. 1 a schematic representation of a monitoring device according to the invention, which is set up to carry out a method according to the invention; Fig. 2a to 2f simplified representations of an object tracking to explain the method according to the invention; Fig. 3 a simplified representation of a time-logarithmically distributed storage of object characteristics or comparison patterns; Fig. 4 a simplified representation to illustrate the method according to the invention in the case of a diffuse motion pattern; Fig. 5 a simplified representation to illustrate the method according to the invention in the case of a directed motion; Fig. 6. A representation of a search area during object tracking; and Fig. 7. A representation of an expanded search area after an object loss.

[0039] Fig. Figure 1 shows a monitoring device 1 comprising an imaging sensor 2, for example, a camera or a thermal imaging device, and a control unit 3 electrically connected to the imaging sensor 2. The control unit 3 can be configured as a software-parameterized detection and control unit and can also be used for digitization. Alternatively, digitization can be performed directly within the imaging sensor 2. The monitoring device 1 also includes an evaluation and / or display unit 4. Particularly for large monitoring areas extending over several kilometers, multiple imaging sensors 2 can be used to record the image sequences. The imaging sensor 2 can be aligned with the areas to be detected, for example, by means of a pan and tilt head. Furthermore, the imaging sensor 2 can also be configured to perform other movements, especially linear ones.

[0040] In the Fig. Sections 2a to 2f explain in more detail the mode of operation of the method according to the invention. Fig. Figure 2a shows object tracking of a detected object 5, which is captured by the tracking imaging sensor 2 in an image 4.1, at a first time point t0. As shown from Fig. As can be seen in 2a, another object 6 is approaching object 5. The object area is indicated by a frame 7. In the Fig. In sections 2a to 2f, comparison patterns 8.1-8.6 with object features 9.1-9.6 are stored at times t0-t4. A pixel-level object representation within object area 7 can be used as an object feature 9.1-9.6. Furthermore, features such as height of gravity (HoG), center of gravity, boundary, phase, edge view, or similar characteristics are also suitable as object features 9.1-9.6.

[0041] As from the Fig. Figures 2b to 2d show which further time points t1, t2 and t3 are shown; the other object 6 first partially, then completely obscures the first object 5. In the Fig. In step 2d, object 5 is almost completely obscured by the other object 6, so that the object tracking detects a mixture of the two objects 5 and 6. When objects 5 and 6 meet, the object features 9.1 of the tracked object 5 are mixed with the features or properties of the encountering object 6. As can be seen from Fig. As can be seen in section 2e, it is therefore possible that the wrong object, namely the other object 6, is tracked further.

[0042] The inventive method for tracking the at least one object 5 by means of the at least one trackable imaging sensor 2, with which the at least one object 5 is detected, avoids this problem by continuing to consider older object features 9.1-9.6. Here, as before, the object features 9.2-9.6 of the changing object 5 are newly learned, but the previous object properties or object features 9.1-9.6 are retained, for example, in a list. Considering the object area 7 in Fig. 2a, this shows the learned object 5 at time t0. The object areas 7 in the Fig. 2b and Fig. 2c indicates relearning with object features 9.2 and 9.3 at times t1 and t2. At time t3, in Fig. 2d a relearning of the object features 9.4 to the approaching other object 6. Without considering the object history according to the invention, at time t4 in Fig. 2e the wrong object 6 is tracked further, with further relearning taking place. When searching for the object position in the subsequent images, older object features 9.1-9.6 are taken into account, at least to the extent that newer features 9.1-9.6, or even older features, are preferred over newer ones, according to the invention. Therefore, if, upon encountering the object according to the Fig. 2a to 2f, if both objects 5 and 6 are visible separately again, the tracker would, even in the case of an erroneous relearning, prioritize the old appearance of object 5 before the encounter and, as in Fig. 2f is shown at time t4, continue tracking the correct object 5.

[0043] According to the invention, a method for tracking the object 5 using the trackable imaging sensor 2 is proposed, with which the object 5 is detected, wherein: - one or more comparison patterns 8.1-8.6, each exhibiting at least one object feature 9.1-9.6 of the at least one object 5, are stored in chronological order in order to perform a recognition of the at least one object 5 in one or more subsequent images 4.1-4.6 captured by the at least one imaging sensor 2, based on at least one to be determined correspondence value of the at least one stored object feature 9.1-9.6 of the at least one object 5, and to perform at least one of the comparison patterns 8.1-8.6 with newly captured object features 9.1-9.6, wherein - the at least one imaging sensor 2 which tracks at least one object 5 and a motion profile 10.1,10.2 (see Fig. 4 and Fig. 5) of at least one object 5 is stored, wherein - when recognizing at least one object 5, if several comparison patterns 8.1-8.6 are stored, to determine the at least one match value, at least one older comparison pattern 8.1-8.6 is taken into account at least to the same extent as newer comparison patterns 8.1-8.6 of the at least one object 5 are considered and / or wherein - when tracking the at least one imaging sensor 2, if the at least one object 5 is no longer recognized in a subsequently acquired image 4.1-4.6 of the at least one imaging sensor 2, an estimate of the further movement 11.1,11.2 of the at least one object 5 is made, in which older stored movements 12.1,12.2 of the at least one object 5 are taken into account at least to the same extent as newer stored movements 13.1,13.2 of the at least one object 5.

[0044] For example, at least eight comparison patterns 8.1-8.6 with at least one object characteristic 9.1-9.6 of the at least one object 5 can be stored in chronological order.

[0045] If the match value to be determined is greater than or equal to a predefinable first limit, preferably about 88%, reliable recognition of the at least one object 5 has been achieved, whereby no new comparison pattern 8.1-8.6 of the at least one object 5 is stored, and / or - if the agreement value to be determined is greater than or equal to a predefinable second limit value, preferably about 85% and less than the predefinable first limit value, a new comparison pattern 8.1-8.6 is stored using newly acquired object features 9.1-9.6 of the at least one object 5, in particular for relearning, and / or - if the agreement value to be determined is smaller than the specified second limit value, at least one object 5 was no longer recognized.

[0046] The degree of agreement can be determined, for example, by cross-correlation at the pixel level.

[0047] The tracking of the imaging sensor 2 can involve a wide variety of movements, such as panning movements or linear movements.

[0048] When recognizing the at least one object 5 to determine the match value, at least one older comparison pattern 8.1 may be used preferentially over at least one newer comparison pattern 8.2-8.6 of the at least one object 5.

[0049] The oldest stored comparison pattern 8.1-8.6 is used to determine the match value, in which the at least one stored object characteristic 9.1-9.6 exhibits at least a predefined minimum match value with the newly recorded object characteristics 9.1-9.6 of the at least one object 5. The minimum match value can, for example, correspond to the second limit value.

[0050] Alternatively, the agreement values ​​of the at least one stored object characteristic 9.1-9.6 of all stored comparison patterns 8.1-8.6 with the newly recorded object characteristics 9.1-9.6 can first be determined. This agreement can then be weighted accordingly using a factor, with older comparison patterns 8.1-8.6 receiving a higher weighting. Finally, the stored comparison pattern 8.1-8.6 in which the at least one stored object characteristic 9.1-9.6 has the highest weighted agreement value with the newly recorded object characteristics 9.1-9.6 is selected.

[0051] The temporal distribution of the storage of comparison patterns 8.1-8.6 of the at least one object 5 can be uniform or logarithmic. In Fig. Figure 3 shows a simplified, logarithmic time distribution of comparison patterns 8.1-8.6, indicated by diamonds. Time t is plotted on the horizontal axis.

[0052] Comparison patterns 8.1-8.6 that are no longer recognized within a defined period can be deleted.

[0053] Comparison patterns 8.1-8.6, which were recognized and / or tracked with a predetermined frequency over a defined longer period, may be preferentially stored.

[0054] The first comparison pattern 8.1 at time t0 can be permanently stored.

[0055] The object features 9.1-9.6 of the at least one object 5 can have appearance properties and / or representations of the at least one object 5.

[0056] The tracking of the at least one imaging sensor 2 can be carried out in such a way that, taking into account the movement or the course of movement 10.1-10.2 of the at least one object 5, this is shown in the area of ​​the center of the subsequently acquired images 4.1-4.6.

[0057] It is determined whether the trajectory or motion profile 10.1,10.2 of the at least one object 5 is diffuse or directed. The in Fig. The movement pattern shown in section 4 (10.1) is diffuse, whereas the one in Fig. The movement pattern shown in Figure 5 is directed towards Figure 10.2.

[0058] When tracking the at least one imaging sensor 2, if the at least one object 5 is no longer recognized in a subsequently acquired image 4.1-4.6, the current position of the at least one imaging sensor 2 may remain unchanged if a diffuse motion pattern 10.1 of the at least one object 5 is detected, with older stored motions 12.1 and newer stored motions 13.1. Thus, the current position 11.1 of the at least one object 5 is presumed to be a further motion of the object 5, and a linear further motion 14 of the object 5 is not erroneously assumed.

[0059] The movements can, for example, be executed as motion vectors.

[0060] When tracking the at least one imaging sensor 2, if the at least one object 5 is no longer recognized in a subsequently acquired image 4.1-4.6, the further movement 11.2 of the at least one object 5 can be estimated by means of a vector addition, in particular a modified or normalized vector addition, of a number of previously acquired and / or estimated movement vectors or movements 12.2, 13.2 of the at least one object 5. Older stored movements 12.2 and newer stored movements 13.2 of the at least one object 5 can be weighted equally. The length of the estimated further movement 11.2 of the at least one object 5 can be determined by averaging the individual lengths of the previously acquired and / or estimated movements 12.2, 13.2 of the at least one object 5 used in the vector addition.This averaging is necessary because otherwise the length would result from adding all the previous movements 12.2, 13.2, which would be too large. The total length of the resulting vector could be divided by the number of vectors being summed. For example, if the last ten preceding movements 12.2, 13.2 are added, the total length of the resulting vector can be divided by 10. Alternatively, one-tenth of the length of each individual vector or movement 12.2, 13.2 could be added. However, the longer the required estimate, the less accurate the estimated further movement 11.2 becomes.

[0061] Since all estimates are subject to a certain degree of uncertainty, the search range 15 can also be used (see Fig. 6 and Fig. 7) will increase over time, as the deviation of the estimate from the actual behavior of object 5 leads to a greater deviation between the expected and the actual position of at least one object 5 over time. Furthermore, the probability that object 5 will change its movement also increases over time.

[0062] In Fig. Figure 6 shows an object area (7) and a search area (15). An automatic adjustment of the search area (15) is shown. Fig.7. While the object area 7 retains its unchanged size, the search area 15 is enlarged. The search area 15 in the captured images of the at least one imaging sensor 2 is gradually expanded over time if the at least one object 5 is no longer recognized in the subsequently captured images 4.1-4.6. The camera tracking is adjusted to a probable location of the object 5. After the object 5 is detected again and any incorrect camera position is corrected, the search area 15 is then reduced back to its original size. If the lost object 5 is not found again after an adjustable time (e.g., 15 seconds), the automatic camera tracking stops and the original area sizes are restored. The user or the automatic object recognition can then assign a new object 5 to the tracker. Reference symbol list 1 monitoring device 2 Imaging sensor 3 Control unit 4 Display / evaluation unit 4.1-4.6 Images of the recorded scene 5 tracked object 6 other objects 7 Object area 8.1-8.6 Comparison Patterns 9.1-9.6 Object characteristics 10.1, 10.2 Movement patterns 11.1, 11.2 further estimated movement 12.1, 12.2 older stored movements 13.1, 13.2 newer saved movements 14 rectilinear motion 15 Search area

Claims

Method for tracking at least one object (5) by means of at least one trackable imaging sensor (2) with which the at least one object (5) is detected, wherein: - several comparison patterns (8.1-8.6), each of which has at least one object feature (9.1-9.6) of the at least one object (5), are stored in temporal sequence in order to perform a recognition of the at least one object (5) in one or more subsequently detected images (4.1-4.6) by the at least one imaging sensor (2) based on at least one to be determined correspondence value of the at least one stored object feature (9.1-9.6) of the at least one object (5) with newly detected object features (9.1-9.6), and wherein - the at least one imaging sensor (2) is tracked and a motion profile (10.1, 10.2) of the at least one object (5), characterized in that – when recognizing the at least one object (5), if several comparison patterns (8.1-8.6) are stored, at least one older comparison pattern (8.1-8.6) is taken into account to determine the at least one match value at least to the same extent as newer comparison patterns (8.1-8.6) of the at least one object (5), wherein the oldest stored comparison pattern (8.1-8.6) is used to determine the at least one match value in which the at least one stored object feature (9.1-9.6) has at least one predetermined minimum match value with the newly acquired object features (9.1-9.6) of the at least one object (5) and / or that – when tracking the at least one imaging sensor (2), if the at least one object (5) is in a subsequently acquired image (4.1-4.6) of the at least one imaging sensor (2) is no longer recognized, an estimation of the further movement (11.1,11.2) of the at least one object (5) is carried out, in which older stored movements (12.1,12.2) of the at least one object (5) are taken into account at least to the same extent as newer stored movements (13.1,13.2) of the at least one object (5), whereby it is determined whether the movement pattern (10.1,10.2) of the at least one object (5) is diffuse or directed. The method according to claim 1, characterized in that: - if the at least one agreement value to be determined is greater than or equal to a predefinable first limit value, a reliable recognition of the at least one object (5) has been achieved, whereby no new comparison pattern (8.1-8.6) of the at least one object (5) is stored, and / or - if the at least one agreement value to be determined is greater than or equal to a predefinable second limit value and less than a predefinable first limit value, a new comparison pattern (8.1-8.6) is stored using newly acquired object features (9.1-9.6) of the at least one object (5), and / or - if the at least one agreement value to be determined is less than a predefinable second limit value, the at least one object (5) has no longer been recognized. Method according to claim 1 or 2, characterized in that, in the recognition of the at least one object (5) for determining the at least one match value, at least one older comparison pattern (8.1, 8.2) is preferably used compared to at least one newer comparison pattern (8.3-8.6) of the at least one object (5). Method according to one of claims 1 to 3, characterized in that the temporal distribution of the storage of comparison patterns (8.1-8.6) of the at least one object (5) is uniform or logarithmic. Method according to one of claims 1 to 4, characterized in that comparison patterns (8.1-8.6) that are no longer recognized within a defined period are deleted. Method according to one of claims 1 to 5, characterized in that comparison patterns (8.1-8.6) that have been recognized and / or tracked with a predetermined frequency over a defined longer period of time are preferably stored. Method according to one of claims 1 to 6, characterized in that the first comparison pattern (8.1) is permanently stored. Method according to one of claims 1 to 7, characterized in that the object features (9.1-9.6) of the at least one object (5) have appearance properties and / or representations of the at least one object (5). Method according to one of claims 1 to 8, characterized in that the tracking of the at least one imaging sensor (2) is carried out in such a way that, taking into account the movement (10.1, 10.2) of the at least one object (5), this is shown in the area of ​​the center of the subsequently acquired images (4.1-4.6). Method according to one of claims 1 to 9, characterized in that when tracking the at least one imaging sensor (2), if the at least one object (5) is no longer recognized in a subsequently acquired image (4.1-4.6), the current position of the at least one imaging sensor (2) remains unchanged when a diffuse motion profile (10.1) of the at least one object (5) is detected. Method according to one of claims 1 to 10, characterized in that, when tracking the at least one imaging sensor (2), if the at least one object (5) is no longer recognized in a subsequently acquired image (4.1-4.6), the further movement (11.2) of the at least one object (5) is estimated by means of a vector addition of a number of previously acquired and / or estimated movements (12.2, 13.2) of the at least one object (5) upon recognition of a directed movement pattern (10.2) of the at least one object (5). Method according to claim 11, characterized in that the length of the estimated further movement (11.2) of the at least one object (5) is determined by averaging the individual lengths of the preceding detected and / or estimated movements (12.2, 13.2) of the at least one object (5) used in the vector addition. Method according to one of claims 1 to 12, characterized in that a search area (15) in the captured images (4.1-4.6) of the at least one imaging sensor (2) is gradually expanded over time if the at least one object (5) is no longer recognized in the subsequently captured images (4.1-4.6). Monitoring device (1) with at least one imaging sensor (2) and with at least one control unit (3) electrically connected to the at least one imaging sensor (2), which is configured to carry out a method according to one of claims 1 to 13.

Images