DE102009001870A1 - Image processing device for monitoring system for processing of monitoring images of monitoring scene, has foreground image or actual moved foreground objects, and flow module, which is formed on basis of monitoring picture - Google Patents

Abstract

The image processing device (1) has a foreground image or actual moved foreground objects, and a flow module (5), which is formed on basis of a monitoring picture to determine an optical flow field. A background module (6) is provided, which is formed on basis of the foreground image and the optical flow field to determine or update a scene background image. Independent claims are also included for the following: (1) a method for determining scene background image during processing of monitoring images; and (2) a computer program with program code unit.

Description

State of the art

The The invention relates to an image processing apparatus for processing surveillance images of a surveillance scene, where moving foreground objects are present in the surveillance scene or in a scene background, with one Differential image module, which is formed from at least one the surveillance images by comparison with a comparison image determine a foreground image, the foreground image possible and / or actual moving foreground objects can, and with a flow module, which is based on of the at least one monitoring image, an optical flow field to determine. The invention also relates to a monitoring system, a method for determining a background image and a corresponding Computer program.

Monitoring systems, especially video surveillance systems are used, around large areas, such as squares, Stations but also buildings, such as libraries, To supervise prisons etc. Such monitoring systems include a plurality of surveillance cameras in a conventional manner, observe the relevant areas in the areas. The image data streams The surveillance cameras are often in a surveillance center merged and there optionally automated or by monitoring personnel controlled.

at automated monitoring is commonplace Image processing algorithms for the automated evaluation of the Use image data streams. A possible approach it is moving objects of static or essentially static Separate scene background, to track over the time and to trigger alarms for relevant movements. Indeed It presents a technical challenge, this scene background reliable to determine.

For example, from the document DE102007024868A1 an image processing device, a monitoring system, a method for generating a scene reference image and a computer program known, wherein the monitoring system comprises a learning device that learns the scene reference image by evaluating a medium-term observation and / or long-term observation of the surveillance scene.

From other applications of image processing, the so-called optical flow is known, wherein the displacement of a pixel from a first image to a pixel in a second, subsequent image is represented by a flux vector. This optical flow also in the Aron Lepisk's Master's Thesis, 2005, entitled "The Use of Optical Flow Within Background Substraction" by the School of Engineering Physics, Royal Institute of Technology discussed

Disclosure of the invention

According to the Invention is an image processing apparatus having the features of claim 1, a monitoring system having the features of claim 11, a method for determining a background image with the features of claim 12 and a computer program with the features of claim 13 proposed. Preferred or advantageous Embodiments of the invention will become apparent from the subclaims, the following description and the accompanying figures.

The Image processing apparatus according to the invention is suitable for processing surveillance images and / or formed as a surveillance scene of a surveillance area be included or are. The surveillance area can be formed arbitrarily and for example a place, a street, show a room, etc. The surveillance images are preferred as Part of a surveillance sequence, that is, a series of images, preferably with equi-temporal distance formed. In the surveillance scene, moving, in the context of the invention referred to as foreground objects in front of or in a scene background - also as a scene background image to denote - stop.

The Terms foreground objects or scene background are preferred functionally understood, with the foreground objects also in spatial may appear later areas or the scene background includes static objects from a foreground. A possible definition becomes the scene background by static and optionally supplementary quasi static Objects and / or areas that are foreground objects are moving through Objects formed in the surveillance scene.

The Image processing apparatus comprises a differential image module, which is formed from at least one of the monitoring images by comparison with a comparison image, a foreground image determine. The comparison image can be used as a calculated comparison image, a modeled comparison image and / or another monitoring image be educated. The comparison image can also consist of several monitoring images, formed for example by averaging or temporal filtering be.

The Foreground image includes possible and / or actual moving foreground objects, in particular is the foreground image limited to these foreground objects. actual Foreground objects are defined by the fact that they are real in the surveillance scene exist, whereas possible moving foreground objects z. B. be so-called ghost objects can, therefore objects, which as artifacts, erroneous Areas or false objects have been created by the image processing.

moreover The image processing device comprises a flow module, which is formed, based on, in particular, taking into account of the at least one monitoring image, an optical flow field to determine. Preferably, this forms at least one monitoring image a start or an end image for the flow vectors of the optical flow field.

Under an optical flow field (English: optical flow) is preferably a vector field denotes the direction of movement and / or speed and / or acceleration for each pixel of an image sequence indicates. The optical flow or the optical flow field can be used as the motion vectors projected onto the image plane of visible Objects are understood. In the simplest case, a flow vector designates the translation vector of a pixel in image coordinates between its Position in a first image and its position in a second Image. To determine the optical flux field are a variety known by methods, for example block matching or differential Methods.

in the According to the invention, it is proposed that the image processing device a background module, which is formed on the basis foreground image and optical flow field a scene background image to determine and / or update.

By the fusion of the input information of the foreground picture and the optical flow field, it is possible to use the difference image with the actual or possible moving Analyze foreground objects more closely. It is for example possible to make a statement for the case when complete foreground objects or parts of them do not move have that these are not actual moving foreground objects but must be assigned to the background. Such areas can then easily from the foreground image taken in the background, especially by doing this be copied into the scene background.

Special Advantages, the image processing device z. At system startup, Camera position changes and / or initialization phases the image processing device provide, especially at close populated scenes as the scene background quickly detects and subsequently a fast robust segmentation of the moving Foreground objects can be achieved.

at a possible embodiment of the invention is the comparison image from one or more other surveillance images and / or formed as a or the scene background image or based on this. In the simplest case, the comparison image another, in particular a previous monitoring image be. This embodiment is especially at initialization the image processing apparatus preferred. In another embodiment can the comparison image from several other monitoring images, for example by averaging, cumulation, smoothing, filtering or the like. In yet other embodiments the comparison image is designed as a scene background image, which is updated by means of the image processing device and / or is improved. The scene background image can be a real picture or a z. B. be modeled by CAD techniques image.

at a preferred embodiment of the invention is that Background module formed, objects and / or object areas in the foreground image with the flow vectors from the optical flow field to merge at the corresponding positions. At this Embodiment metadata are formed, wherein the objects and / or the object areas as possible and / or actual moving foreground objects with the associated flow vectors of the optical flow field are linked by data technology.

at a possible development of the invention forms the Foreground image a mask for the optical flow field, so that in a resulting data field only entries of river vectors are registered at the positions which the Pixels of actual and / or possible correspond to moving foreground objects in the foreground image.

The background module can be designed, for example as a set of rules, to be optional in that objects and / or object areas to which flow vectors with a length greater than a limit length are assigned are verified as actual foreground objects. This design is based on the consideration that objects and / or object areas in the foreground image which show sufficiently safe or meaningful flow vectors having a starting point in the one image and an end point in the other image are highly likely to be real or actual Foreground object are formed.

at another or additional possible training, in particular, as another rule, become objects and / or object areas in the foreground image, which have flow vectors with a length less than one or the aforementioned limit length assigned are recognized as background areas. The basis of this education lying consideration is that these objects or object areas mistakenly associated with the foreground image. there for example, they can be so-called ghost objects, which are formed by an actually moving one Foreground object has left a position and this position was recognized as a difference compared to the comparison image, however, this position is actually at this position a background area. Another interpretation is it is that an actual moving foreground object too has become a static object, for example an object like a cabinet, which is positioned in the surveillance scene becomes a static object and from then on a Background area forms.

at another possible supplement or training, which may also be configured as a rule, is the background module designed so that objects and / or object areas, which no Flow vectors are assigned, also as background areas be recognized. A consideration of this training is that background areas are recognized, which initially obscured were and then exposed. After these objects or object areas when calculating the optical flow field either no start or have no destination point, can these objects or Object areas no flow vectors are assigned. By the missing assignment of flow vectors become these objects or object areas recognized as belonging to the background area.

at a possible development of the invention are the objects and / or object areas formed as segmented objects. at the segmentation becomes objects and / or object areas, from a related and / or largely related Area and / or more coherent area segmented optical flow field properties into a common object.

Preferably If the background module is designed, the previously formulated embodiments, in particular the rules, or a selection thereof for all or at least a majority of the flow vectors of one of the segmented Apply objects to the segmented object as a whole in the frame an overall score as an actual foreground object and / or to verify as a background area or as a background area to recognize. With this training, the segmented Objects are analyzed in an overall view, so that themselves the significance of the analysis result further increased.

at A possible development of the invention is the background module trained to analyze the segmented objects and to a First condition with respect to the direction of the flow vectors and on a second condition with respect to a mean vector length to investigate. In the event that to the segmented Object associated flow vectors sufficiently rectified are, for example, at least 60%, preferably at least 70% and in particular at least 80% of the flow vectors rectified with an angular deviation of less than 30 °, preferably less than 20 ° and in particular less than 10 ° and also that the mean vector length of the flow vectors and / or the rectified flow vectors larger can be the segmented object as the or a limit length be reliably verified as a foreground object. Through the measures of segmentation and investigation global properties of the segmented objects can be the recognition accuracy be further increased. In addition or instead of Condition of the mean vector length can z. B. also the Distribution of the vector lengths of the individual flow vectors to be examined.

A Further embodiment of the invention provides for the case that one of the segmented objects is a mixed and / or statistical distributed alignment of the flow vectors before, this object as Identify background area.

at a possible development of the invention is only one Updated partial area of the scene background image. This embodiment can be used in particular advantageous if a static Object is moved and only the one of the previously static Object obscured background area in the scene background image introduced, in particular to be copied.

In an advantageous embodiment of the invention, the image processing device is designed such that objects and / or object regions from the foreground image which have been identified as being associated with the scene background image and / or as background regions are transferred into the scene background image, in particular copied. It can be provided that these new background areas replace the previous corresponding areas of the scene background image all at once or - for example, by an averaging or filtering - the scene backdrop complement Zen.

One Another object of the invention relates to a monitoring system with the features of claim 11, which according to the invention is an image processing apparatus, as described above or according to one of the preceding Claims. The monitoring system can have a or have multiple surveillance cameras, which on the or more monitoring areas is directed or are and which monitoring images to the image processing device deliver. In particular, the monitoring system is as a Video surveillance system for buildings, Places, etc. trained.

One Another object of the invention relates to a method for determination a background image when processing surveillance images with the features of claim 12, wherein in one step at least one of the surveillance images by comparison with a Comparison image is a foreground image is determined, which possible and / or actually moving foreground objects and with another step, based on the at least one monitoring image an optical flow field, for example by comparison with a preceding, in particular immediately preceding monitoring image the same surveillance scene is determined. According to the invention based on or by fusion of the foreground image and the optical Flow field determines a scene background image and / or updated. For the implementation of the method, a monitoring system and / or an image processing apparatus according to any one of the preceding Claims or as previously described, used.

One The last object of the invention relates to a computer program with the features of claim 13, which the method according to the Claim 12 executes when it is on a computer and / or a monitoring system and / or an image processing device in each case according to one of the preceding claims or as it has been previously described.

Further Features, advantages and effects of the invention will become apparent the following description of a preferred embodiment the invention and the accompanying figures. Showing:

1 a schematic block diagram of an image processing apparatus as a first embodiment of the invention;

2 in the same representation as 1 another embodiment of the invention;

3 an illustration of the generation of a difference image for one of the image processing apparatus of the preceding figures;

4 an illustration of the formation of an optical flow field for one of the image processing devices of the preceding figures;

5 an illustration of the accidental generation of a ghost object in the difference formation in connection with the image processing apparatus of the preceding figures;

6 a flowchart illustrating an embodiment of the inventive method, which can be performed on one of the preceding image processing devices;

7 a further illustration of one or the method according to the invention.

The 1 shows a schematic block diagram of an image processing device 1 as a first embodiment of the invention.

The image processing device 1 is part of a surveillance system 2 , which in addition one or more surveillance cameras 3 and which is suitable for monitoring areas and / or trained and / or arranged. The surveillance cameras 3 For example, they are directed to surveillance areas that are of interest for surveillance. The image processing device 1 is in the 1 in a starting phase or in an initialization phase, for example, after changing the camera position shown.

Starting from the surveillance cameras 3 image data streams become the image processing device 1 directed. In the image processing apparatus 1 the image data streams are split, alternatively, the image processing described below can also be done sequentially: In a difference module 4 becomes for a surveillance picture 8a (following figures) one of the surveillance cameras 3 a comparison with a comparison image performed, wherein the comparison image, for example, a previous monitoring image 8b (following figures) of the same surveillance camera 3 and / or an averaging of several surveillance images of the same surveillance camera 3 can be. The comparison result is a foreground image 10 (following figures), which areas in the monitoring image 8a shows which to the comparison image 8b are formed differently. In a first assumption, these regions are referred to as possible and / or actual, moving foreground objects.

The same surveillance picture 8a is connected to a flow module 5 which is based on this monitoring image 8a and a previous monitoring image 8b the same surveillance camera 3 an optical flow field 15 (following figures) determined. The optical flow field 15 shows flow vectors 16 (Subsequent figures), whose direction by a translation direction, for example, a pixel or other image area of the image position in the one monitoring image 8b to the image position in the other monitor image 8a is determined. The length or amount of the flow vector 16 is given by the translation length, by the translation speed and / or the translation acceleration. The river vector 16 may also be multi-dimensional and comprise one, some or all of the three components of motion mentioned above.

In order to keep the computational effort low, it can be provided that, unlike those in the 1 shown information flows only for the regions in the foreground image 10 , which were found as a difference to the comparison image, the optical flow field 15 be calculated. In this way, regions that have the current surveillance image 8a not different from the comparison image, from the processing step in the flow module 4 be spared.

In a background module 6 become the foreground picture 10 and the optical flow field 15 fused together, based on the fused data, a scene background image 9 (subsequent figures) and, for example, in a background memory 7 is filed.

• • Bereiche in dem Vordergrundbild, denen Flussvektoren 16 in dem optischen Flussfeld 15 zugeordnet sind, die eine vordefinierte Grenzlänge hinsichtlich der Translationslänge, -geschwindigkeit und/oder -beschleunigung überschreiten, werden als tatsächliche Vordergrundobjekte verifiziert.
• • Bereiche in dem Vordergrundbild, bei denen die genannte Länge der Flussvektoren 16 kleiner als die Grenzlänge ist, werden als Hintergrundbereiche identifiziert.
• • Bereiche, bei denen keine Flussvektoren 16 zu definieren waren, weil der Start- oder der Endpunkt der korrespondierenden Bildbereiche nicht auffindbar waren, werden ebenfalls als Hintergrundbereiche identifiziert.

Determining the areas of the scene background image 9 is based on one, some or all of the following selection rules:

• • Regions in the foreground image that have flow vectors 16 in the optical flow field 15 are assigned that exceed a predefined limit length in terms of translation length, velocity and / or acceleration are verified as actual foreground objects.
• • Areas in the foreground image where the said length of the flow vectors 16 is less than the limit length, are identified as background areas.
• • Areas where there are no flow vectors 16 were to be defined because the start or end point of the corresponding image areas could not be found are also identified as background areas.

By doing so, even areas in the surveillance image can 8a which in the difference module 4 were still recognized as possible moving foreground objects, to be properly identified as background areas.

The 2 shows the image processing apparatus 1 after the initialization phase, wherein in the difference module 4 one of the surveillance images with the scene background image 9 is compared, which z. B. during initialization according to the 1 was created. Alternatively, in the background memory 7 a CAD-generated image can also be saved as the initial image. After initialization according to the 1 can the scene background image 9 according to the structure of 2 be improved by continuous updating, with an earlier scene background image 9 by information from the current monitoring image 8b is updated iteratively.

The 3 schematically shows the formation of a difference image, wherein a monitoring image 8th with a scene background image 9 is compared to the foreground image 10 to build. When in the 3 As shown, the scene background shows a street 11 as well as several trees 12 , The surveillance picture 8th on the other hand shows passers-by as foreground objects 13 as well as a moving car 14 , In the generation of the foreground image 10 become the image areas of the surveillance image 8th in the foreground picture 10 copied differently to the scene background image 9 are formed.

The 4 shows an illustration for generating the optical flow field, wherein two monitoring images 8a , b again with the road 11 , several trees 12 and now a right-to-left car 14 are shown at two different times T1 and T2. An optical flow field 15 shows by way of example a plurality of flow vectors 16 showing the translation of the pixels of the vehicle 14 from the surveillance picture 8b to the surveillance picture 8a represent. To simplify the illustration, only some of the flow vectors 16 and also the position of the vehicle 14 at the time T1 and at the time T2 again shown in dashed lines.

The 5 Illustrates the formation of ghost areas in a schematic representation 17 in the foreground picture 10 , In the exemplary illustration, the surveillance images became 8a . 8b of the 4 used, wherein the surveillance image 8a as the current monitoring image and the surveillance image 8b is used as comparison image. In a comparison of the surveillance images 8a , b remain two areas, one - represented in the figure dark area - an actual foreground object, the - represented slightly lighter - area, but only the road 11 and thus a ghost area 17 forms. A distinction as to which of the areas is an actual and which of the areas is an incorrectly arranged one Foreground object is based on the foreground image 10 not possible. But here is the optical flow field 15 a clear indication, since the optical flow field 15 at the position of the car 14 parallel or nearly parallel flow vectors 16 with a sufficient length, at the position of the ghost area 17 however, shows no flow vectors. (As the position of a flow vector 16 the target point is used in the present presentation.) For the background module 6 is the ghost area 17 thus easily identifiable as a background area and can be stored in the background memory 7 to form the scene background image 9 or copied to its update.

• • Die Übernahme der als Hintergrundbereich identifizierten Bildbereiche in das Szenenhintergrundbild 9 in unbewegten Bereichen könnte erst nach einer festgelegten, minimalen Zeitspanne erfolgen, während dessen der optische Fluss an der jeweiligen Bildposition näherungsweise 0 sein muss.
• • Da die Bildverarbeitungsvorrichtung 1 vor allem der schnellen Berechnung eines ersten Szenenhintergrundbildes 9 dienen kann, könnte die Bildverarbeitungsvorrichtung 1 lediglich für eine feste Zeitspanne nach Start des Überwachungssystems 2 bzw. nach einer globalen Änderung im Überwachungsbild (zum Beispiel Kamerabewegung oder Beleuchtungsänderung) eingesetzt werden.
• • Um das Szenenhintergrundbild 9 nur in Bereichen schnell zu korrigieren, z. B. wenn im Betrieb des Überwachungssystems eine Bewegung startet (zum Beispiel: geparktes Auto fährt los), ist folgender Ansatz möglich: Bei Neudetektion eines Objekts wird ein erstes Vordergrundbild 10 gespeichert. Nur innerhalb der vorhandenen Bereiche des Vordergrundbilds 10 wird das beschriebene Verfahren solange angewandt, bis diese Bereiche keinen Vordergrund mehr enthalten. Dieser Ansatz führt auch bei Objekten, die den Bildbereich von außerhalb betreten, zu einem korrekten Ergebnis.
• • Optional werden pausierende, bereits verfolgte Objekte mit dem folgenden Vorgehen nicht in das Szenenhintergrundbild 9 integriert: Sogenannte fehlerhafte Bereiche sind dadurch gekennzeichnet, dass es sich um als Vordergrund detektierte Bildpunkte handelt, die einen Startpunkt eines Flussvektors 16 des optischen Flusses, aber keinen Endpunkt (oder andersherum) aufweisen. Werden nur solche fehlerhaften Bereiche (z. B. Geisterbereich 17) in das Szenenhintergrundbild 9 integriert, bleiben tatsächliche Objekte außen vor, da deren Bildpunkte auch Endpunkte von Flussvektoren 16 aufweisen, die jedoch z. B. bei pausierenden Objekten keine Bewegung zeigen und damit die Länge 0 haben.

To prevent even short-term pausing foreground objects in the scene background image 9 be included, several, optionally complementary approaches are conceivable:

• • The inclusion of the image areas identified as background areas in the scene background image 9 in non-moving areas could be done only after a fixed, minimum period of time during which the optical flux at the respective image position must be approximately 0.
• • Since the image processing device 1 especially the fast calculation of a first scene background image 9 could serve, the image processing device 1 only for a fixed period of time after the start of the monitoring system 2 or after a global change in the monitoring image (for example camera movement or illumination change).
• • To the scene background image 9 only in areas quickly correct, z. For example, if movement starts during operation of the monitoring system (for example: parked car starts), the following approach is possible: When an object is re-detected, it becomes a first foreground image 10 saved. Only within the existing areas of the foreground image 10 the method described is applied until these areas no longer contain a foreground. This approach also results in a correct result for objects entering the image area from outside.
• • Optionally, pausing, already traced objects are not included in the scene background image with the following procedure 9 integrated: so-called defective areas are characterized in that they are detected as foreground pixels, which is a starting point of a flow vector 16 of the optical flow, but have no end point (or vice versa). Are only such defective areas (eg 17 ) in the scene background image 9 integrated, actual objects remain outside, since their pixels also endpoints of flow vectors 16 have, however, z. For example, pausing objects show no movement and thus have the length 0.

The described device or the method described with any type of image (for example, gray scale, color, infrared etc.) use.

The 6 shows a schematic representation of the initial or iterative building or updating the background image 9 , A first surveillance picture 8b is used to initialize the scene background image 9 used. A more recent surveillance picture 8a is used for a foreground image 10 and second, the optical flow vectors 16 for the areas in the foreground image 10 to determine. The merged data is in the intermediate image 18 shown, with the car 14 a movement to the left and the passerby 13 makes a movement to the right. These areas belong both to the foreground picture 10 and also have flow vectors 16 which are greater than a predetermined limit length or other conditions with respect to the optical flow field 15 fulfill. These areas are in the surveillance image 8b verified as actual foreground images. The remaining areas of the surveillance image 8a are identified as the scene background and the scene background image 9 will be overwritten with these areas. Particularly striking in this approach is the quick update of the scene background image 9 at the passerby 13 to perceive which in the updated scene background image 9 already after an update step only partially forms a scene background.

The 7 shows the difference in learning the scene background image 9 in the event that only the foreground picture 10 (upper line) or the foreground picture 10 and the optical flow field 15 (bottom line) to form the scene background image 9 be used. In the foreground picture 10 in the upper line are as possible or actual foreground objects the car 14 and one of the passers-by 13 each shown twice, namely both at their start and at their final position. The following picture 19 shows the masked background, with the bright areas representing which areas of the comparison image 10 represent, are omitted from the background. Becomes the scene background image 9 only with the picture 19 updated, no significant areas are transferred from the foreground to the scene background. By contrast, by the common evaluation of the foreground image 10 and the optical flow field 15 between a ghost area 17 and actual foreground objects, so that when updating the scene background image 9 significantly more areas, such as those in the surveillance image 8b from the car 14 or by the passerby 13 hidden areas, into the scene background image 9 be transferred.

QUOTES INCLUDE IN THE DESCRIPTION

This list The documents listed by the applicant have been automated generated and is solely for better information recorded by the reader. The list is not part of the German Patent or utility model application. The DPMA takes over no liability for any errors or omissions.

Cited patent literature

• DE 102007024868 A1 [0004]

Cited non-patent literature

• - Master's Thesis by Aron Lepisk, 2005, entitled "The Use of Optical Flow Within Background Substraction" by the School of Engineering Physics, Royal Institute of Technology [0005]

Claims (13)

Image processing device ( 1 ) for processing surveillance images ( 8th . 8a , b) a surveillance scene, wherein foreground objects moved in the surveillance scene ( 13 . 14 ) in front of or in a scene background, with a differential image module ( 4 ), which is formed from at least one of the monitoring images ( 8a ) by comparison with a comparison image ( 8b . 9 ) a foreground picture ( 10 ), wherein the foreground image ( 10 ) may comprise possible and / or actual moving foreground objects, and with a flow module ( 5 ), which is formed on the basis of at least one monitoring image ( 8a ) an optical flow field ( 15 ), characterized by a background module ( 6 ), which is formed on the basis of the foreground image ( 10 ) and the optical flow field ( 15 ) a scene background image ( 9 ) to determine and / or to update. Image processing device ( 1 ) according to claim 1, characterized in that the comparison image as another ( 8b ) or several other surveillance images and / or as one or the scene background image ( 9 ) is formed and / or based on this or this. Image processing device ( 1 ) according to claim 1 or 2, characterized in that the background module ( 6 ) is formed, objects and / or object areas in the foreground image ( 10 ) with flow vectors ( 16 ) from the optical flow field ( 15 ) to merge at the corresponding positions. Image processing device ( 1 ) according to claim 3, characterized in that the objects and / or object areas to which flow vectors ( 16 ) having a length greater than a limit length are verified as actual foreground objects. Image processing device ( 1 ) according to claim 3 or 4, characterized in that objects and / or object areas, which flow vectors ( 16 ) associated with a length less than a limit length or the limit length are recognized as background areas. Image processing device ( 1 ) according to any one of claims 3 to 5, characterized in that objects and / or object areas, which are not subject to flow vectors ( 16 ) could be assigned as background areas. Image processing device ( 1 ) according to one of the preceding claims, characterized in that the objects or object areas are formed as segmented objects. Image processing device ( 1 ) according to claim 7, characterized in that the background module ( 6 ) is configured to verify one of the segmented objects, the rectified flow vectors, and / or an average vector length of the associated flow vectors greater than the limit length (to be the foreground object. Image processing device ( 1 ) according to claim 7 or 8, characterized in that the background module ( 6 ) is adapted to identify one of the segmented objects, which has a mixed and / or statistically distributed orientation of the associated flow vectors, as the background area. Image processing device ( 1 ) according to one of the preceding claims, characterized in that the background module ( 6 ) is adapted to update only a portion of the scene background image. Monitoring system ( 2 ) with an image processing device ( 1 ) according to any one of the preceding claims. Method for determining a scene background image during the processing of surveillance images ( 8th . 8a , b), wherein in at least one of the monitoring images ( 8a ) by comparison with a comparison image ( 8b . 9 ) a foreground picture ( 10 ), which comprises possible and / or actual moving foreground objects and with a further step, wherein on the basis of the at least one monitoring image ( 8a ) an optical flow field ( 15 ), for example by comparison with a preceding, in particular immediately preceding monitoring image ( 8b ) of the same surveillance scene, and where by merging the foreground image ( 10 ) and the optical flow field ( 15 ) a scene background image ( 9 ) is determined and / or updated. Computer program with program code means for performing all the steps of the method according to claim 12, when the program is stored on a computer and / or on a device ( 1 . 2 ) is carried out according to one of the preceding claims.