DE102022202942A1 - Monitoring device, monitoring arrangements, method, computer program and storage medium - Google Patents

Abstract

Monitoring device 1 with a camera 2 for recording at least one surveillance image 3, a surveillance scene 4, the camera 2 having a discrimination module 6a and an image model determination module 6b, the discrimination module 6a being designed to detect objects 5a-e in the surveillance image 3 and for to determine the detected objects 5a-e object information data 7a-e, wherein the object information data 7a-e include an object class, an object type, an object ID 8a-e and / or an object geometry information 9a-e, wherein the image model determination module 6b is formed, based to determine an image model 11 on the surveillance image 3, the image model 11 comprising a scene model 12 of the surveillance scene 4 and the object information data 7a-e.

Description

State of the art

The invention relates to a surveillance device with a camera for recording at least one surveillance image of a surveillance scene.

Monitoring indoor and outdoor areas using cameras for surveillance purposes is widespread. While in the early days of video surveillance the camera recordings were viewed and evaluated entirely by surveillance personnel, computer-implemented image evaluation methods are now often used, in particular image evaluation based on neural networks. Methods and algorithms for image evaluation usually include a large number of method steps, for example starting with a pre-processing step, filtering and scaling in the individual frames, with segmentation methods, classification methods and/or tracking methods often being used for further processing. For evaluation, the recorded videos and/or images are sent from the camera to an evaluation device, with large amounts of data having to be transmitted given the image and video qualities.

Disclosure of the invention

A monitoring device with the features of claim 1 is proposed. Furthermore, a monitoring arrangement, a method, a computer program and a storage medium are proposed. Preferred and/or advantageous embodiments result from the subclaims, the description and the attached figures.

A monitoring device is proposed. The monitoring device is designed in particular to record, evaluate and/or process images and/or videos. In particular, images and video are used essentially synonymously below, so that an evaluation and/or an analysis of an image is understood as the evaluation and analysis of frames of a video. The monitoring device can form a hardware module, for example a surveillance camera, or can form a module for connection and/or integration into the surveillance camera. Alternatively, the monitoring device can form a software module, for example for execution, application and/or implementation in the surveillance camera.

The monitoring device comprises at least one camera, in particular a surveillance camera. In particular, the monitoring device itself forms a camera. The camera is designed and/or set up to record at least one surveillance image, referred to as an image for short. In particular, the camera is designed to record a video, whereby the video can be understood as a large number of surveillance images. The camera records a surveillance scene as the surveillance image. The surveillance scene can, for example, include or form an indoor area and/or an outdoor area. In particular, the surveillance scene includes static and moving objects, referred to as moving and dynamic objects. Furthermore, the surveillance scene can be divided into a background and a foreground, the background in particular comprising static objects. The surveillance image is in particular a high-resolution surveillance image, for example with a quality of at least 4K or HD.

The monitoring device, in particular the camera, comprises a first and a second computing unit. In particular, the first and second computing units form separate and/or separate units, in particular separate and/or separate hardware units, hardware components, computer and/or processor units. Preferably, the first and second computing units each include their own and/or independent system resources, such as processors and/or memory. The first and second computing units are connected for data via a feedback connection. The feedback connection is designed, for example, as a wired connection, as a conductor track or as a wireless connection. In particular, the feedback connection is designed as a bidirectional connection.

The first computing unit includes a discrimination module and an image model determination module. In particular, the discrimination module and the image model determination module can form a common module. The discrimination module and/or the image model determination module can form hardware or software modules. Particularly preferably, the discrimination module and/or the image model determination module are designed and/or set up to use a machine learning algorithm and/or include a neural network.

The discrimination module is designed to evaluate and/or analyze the surveillance image, the surveillance images and/or the video, in particular to provide methods of image evaluation, object recognition and/or object tracking turn around. The discrimination module is designed to detect, discriminate and/or classify objects in the surveillance image. For example, the discrimination module is designed to segment the surveillance image and/or to determine features of objects in the surveillance image, with the objects being detected, defined and/or localized based on the specific features.

The discrimination module is designed to determine object information data for the detected, discriminated and/or classified objects, in particular to determine separate and/or separate object information data for each of the detected objects. The object information data preferably includes and/or forms alphanumeric data. The object information data each includes and/or describes an object class of the associated detected object and/or an object type of the associated detected object, an object ID and/or object geometry information. The object class describes, for example, which class of objects the respective object belongs to; in particular, the classes can be defined in a user-defined manner. For example, static, moving and/or dynamic objects can be used as object classes; alternatively and/or additionally, human, face, vehicle, security-relevant event, firearm or piece of luggage can be used as object classes. The object ID in particular forms a unique identification option, for example number, designation or similar. For example, the detected objects are numbered consecutively. The object geometry information describes and/or includes at least one piece of information about the geometry of the object, for example a size, position, shape and/or a relationship to another of the detected objects.

The image model determination module is designed and/or set up to determine an image model. For this purpose, the object information data is provided to the image model determination module, and the surveillance image is also provided to it, in particular also the detected objects and/or determined object areas for the detected objects. The image model is determined in particular based on the surveillance image, in particular the video, and the object information data. The image forms and/or describes in particular a scene model, in short a model of the surveillance scene. Furthermore, the image model includes the object information data and/or a subset of the information comprised of the object information data. In particular, the scene model forms an abstraction of the surveillance scene, in particular having a smaller data size, generalized areas and/or sections, summaries and/or a representation of areas through symbols, data sets or other representatives. In particular, the discrimination module and/or the image model determination module determined object areas for the detected objects, for example in the form of bounding boxes. For example, the image model determination module is designed to reproduce the object information data in the scene model instead of the image content of the object areas, the image content being, for example, simplified, alienated, removed or blackened. In particular, the scene model can form a description of the surveillance scene using the object information data, for example the structure of the surveillance scene through the object information data and their relationships, for example based on the object geometry information. In particular, the objects and/or the object regions are included and/or represented in a simplified manner in the scene model, in particular in the form of the object information data. For example and/or the scene model describes that there are two objects in the surveillance scene, one of the Car object class and one of the Person object class, the position, size and orientation for each of the objects being determined and/or described by the object information data.

The second computing unit includes an event detection module. The event detection module is designed in particular as a software or hardware module. The image model and/or the object information data is provided to the second computing unit by the first computing unit and/or the image model determination module. The event detection module is designed to detect and/or determine one or more events based on the image model, in particular the evaluation of the image model. The events are, for example, events in the surveillance scene and/or events of the object and/or events between objects. For example, the event is the crossing of a physical and/or virtual boundary. For this purpose, the discrimination module, for example, differentiated, classified and/or discriminated against objects into humans and non-humans.

The second computing unit, in particular the event detection module, is designed and/or set up to provide, transmit and/or send feedback data to the first computing unit, the discrimination module and/or the image determination module. In particular, the feedback data is provided and/or transmitted via the feedback connection.

Feedback data from the second computing unit includes in particular static information about necessary discrimination information and/or dynamic feedback resulting from the situation. In particular, accident detection fundamentally requires information about the position of vehicles, pedestrians and, if necessary, animals. An accident can at least be estimated based on the position of the detected objects, in particular due to the overlapping of the objects. The feedback data includes the necessary object classes and additional information, in particular the position of the detected objects and their size.

For dynamic feedback data, static information about necessary discrimination information is assumed, which is additionally analyzed. In particular, facial recognition can be switched on when a person has been recognized in a certain area. Face recognition does not generally take place for the entire video stream, but is only activated when certain previous events have been recognized. The feedback data not only includes the required discrimination (the facial data), but also additional information for which object or for which area this discrimination must take place if the object is clearly identifiable.

The invention is based on the idea of processing images and videos into a scene model on the camera side, before their transmission and/or in the immediate vicinity of the camera, whereby the scene model describes and/or includes the essential and/or relevant information of the objects in the surveillance scene , so that for further use and/or evaluation it is not the high-resolution surveillance images and/or videos that need to be transmitted, but rather the scene model with a smaller size. In particular, the invention is based on the consideration that most methods for monitoring and/or image evaluation include several method steps, with the different steps and/or methods often having in common that the amount of data is significantly reduced in each step, the information content and/or the semantic meaning of the data increases significantly after each step and the process steps become more specific after each step. The invention uses the consideration that in particular the initial steps, here detection of objects and object information data, are essentially generic and/or similar for all subsequent method steps, so that they can be taken over together by the discrimination module and the further method steps on the resulting scene model can be applied.

It is particularly preferred that the discrimination module is designed to track the detected objects, in particular the moving and/or dynamic objects. In particular, tracking the object is understood to mean tracking the object, determining the change in position of the object with respect to images recorded at different times and/or determining a trajectory of the object. Tracking data is determined and/or determined for the detected objects by the discrimination module, wherein the tracking data is and/or is preferably provided to further modules and/or algorithms. In particular, the object information data includes the tracking data of the associated object. For example, the object information data includes, as tracking data, the change in the position of the object and/or the object area over time, in particular with regard to two successive surveillance images.

It is particularly preferred that the image model and/or the scene model forms and/or comprises a symbolic representation of the surveillance scene, in particular that the objects and/or object areas of the surveillance scene and/or the surveillance image in the image model and/or in the scene model as symbolic ones representation are included. For example, the symbolic representations can be designed as an alphanumeric data set comprising the object information data. Alternatively and/or additionally, the symbolic representations include, for example, simplified pictograms or object areas without image content, so that in the image model, for example, the object areas of the detected objects, in particular their image content, are removed and/or simplified. For example, the image model is designed as the surveillance image with simplified image content, for example in the form of pictograms. The image model and/or the scene model is thus designed as a simplified version of the surveillance image, with the detected objects being included in the form of the object information data, so that the correspondingly removed and/or simplified object areas can be further used and/or evaluated using the object information data. For example, object class car, object ID 2 is assigned to one of the object areas as object information data and a size, orientation and/or position within the surveillance scene and/or the surveillance image is assigned as object geometry information. Such an image model can be used for evaluation by applications and/or image evaluation methods, since the corresponding Object areas include the essential generic information and/or features even without the presence and transmission of the actual image content.

In particular, it is provided that the image model, the scene model and/or the object information data include and/or describe the detected objects as symbolic representatives. For example, the object areas or the image contents of the object areas are replaced by symbolic representatives, with the symbolic representatives forming, for example, simple pictograms, monochrome, regular and/or irregular areas. In particular, representatives and representations are understood to be synonymous. In particular, this embodiment is based on the consideration that the image contents of the object areas no longer have to be transmitted for data reduction when symbolic representatives are used, in particular when using symbolic representatives, for example simple pictograms, an intuitive interpretation by the monitoring personnel is still possible.

It is particularly preferred that the geometry information data for the respective object include and/or describe an object size, an object position, an object orientation and/or a Z order. The object size describes, for example, an extent and/or size of the object, preferably in image space, for example in pixels. The object size can, for example, include partial sizes such as width, height and object shape. The object position is in particular the position of the object, the object area and/or a center point and/or center of gravity of the object area within and/or in relation to the surveillance scene, the surveillance image, an image coordinate system or a reference coordinate system. The object orientation describes, for example, the orientation of the object with respect to a reference orientation, for example determined by the surveillance image. The object orientation includes, for example, a relative rotation of the length and width of the object with respect to the image coordinate system and/or reference coordinate system. The object orientation, the object size and/or the object position is specified in particular relative to a reference object or the further detected object. In particular, the geometry information data allows the arrangement of the objects shown in the surveillance image to be reproduced in a simplified but reconstructable manner.

Preferably, the second computing unit and/or the event detection module is designed to adapt, configure and/or optimize the first computing unit, the discrimination module and/or the image model determination module based on the feedback data and/or by means of the feedback data. For example, the second computing unit and/or the event detection module is designed to use the feedback data to adapt and/or configure the discrimination module for the detection, discrimination and/or classification of specific objects and/or those required for event determination, for example the quantity of objects to be detected, discriminated and /or classifying objects.

Particularly preferably, the event detection module is designed to execute and/or use at least one evaluation application, in particular to apply it to the image model to determine the event. The evaluation applications are in particular interchangeable, supplementable and/or individually obtainable, for example from an app store. The evaluation applications are in particular applications for the specific evaluation of the image, for example for facial recognition, person recognition, object tracking and/or for the recognition of certain events, such as a fire, a panic or a shooting. For example, the evaluation application is designed to determine entry into an area, for example as a virtual tripwire, whereby instead of evaluating the actual surveillance images, the evaluation application evaluates the image model, since the object information data includes and can be determined sufficient information and / or details to carry out the task is whether an object has entered a certain area and/or whether the object is an authorized or unauthorized person.

It is particularly preferred that the event detection module and/or the second computing unit is designed to provide, send and/or transmit the feedback data, in particular to the first computing unit, the discrimination module and/or the event detection module, when an evaluation application is started or initialized , installed and/or registered. Alternatively and/or additionally, the second computing unit and/or the event detection module transmits and/or sends the feedback data when the image model is received, in particular when it is received for the first time and/or the received image model changes more than a tolerated change.

In particular, it is provided that the monitoring device has a validation module. The event detection module is designed, for example, an event, in particular also called scene event. For example, to detect the scene event, the evaluation application is applied to the image model. The scene event is, for example, the occurrence of an event that the evaluation application should determine, for example entering the area or a fire. The validation module is designed and/or set up to apply the evaluation and/or the application of the evaluation application to the surveillance image belonging to the image model and to determine validation data in the event of the detection of the scene event. In other words, after detecting the scene event based on the image model, the validation module checks whether the scene event actually occurred by evaluating the surveillance images itself. The validation data describes and/or includes, for example, the results of the application of the evaluation application to the surveillance image, in particular a statement as to whether the scene event actually occurred or was incorrectly detected as such by the evaluation application based on the image model. This embodiment is based on the idea that only when an event is detected and/or the evaluation application is positively applied to the simplified data, here the image model, is the evaluation and/or transmission of the more data-intensive surveillance images initiated.

It is particularly preferred that the evaluation application has a basic module and a special module. For example, the basic module is a simplified version and/or abstraction of the special module and/or the evaluation by the special module. For example, the basic module is designed for simplified evaluation of images, in particular the image model, while the special module is designed and/or set up for more detailed evaluation. In particular, the special module is designed for use on the surveillance image instead of for use on the image model. It is provided, for example, that the validation module applies the special module to the surveillance image to determine the validation data, while the discrimination module applies the basic module to the image model to apply the evaluation application to the image model.

A further subject of the invention is a monitoring arrangement with the monitoring device and a remote module. The remote module is designed, for example, as a central computer or a cloud. In particular, the monitoring device and the remote module are connected in terms of data technology for wired or wireless data transmission. For this purpose, the monitoring device includes a transmission module for providing the image model to the remote module. The remote module comprises a remote application module, wherein the remote application module is preferably designed and/or set up like the application module of the monitoring device. The remote application module is designed and/or set up to execute and/or use evaluation applications, in particular evaluation applications as described for the application module of the monitoring device. The remote application module applies the evaluation application to the image model. This embodiment is based on the idea that the preprocessing of the surveillance image takes place on the camera side or on the side of the monitoring device and the provision and determination of the image model takes place. This image model, which requires less storage space, is then provided to the remote module for further use and/or evaluation, so that less data needs to be transmitted than for transmitting the surveillance image. Further evaluation then takes place on the remote module, or by means of the remote application module, in which the evaluation application is applied to the provided image model.

For example, in order to use the evaluation application, an increased need for computing and/or storage power is required, which in particular would not be available on the monitoring device or the camera, so that the computing and/or storage-intensive parts of the evaluation are carried out centrally.

In particular, the transmission module is designed and/or set up to provide and/or send the image model to the remote module in encrypted, photographic and/or compressed form. For example, this is based on a key exchange between the sending module and the remote module. In particular, it is provided that the remote application module comprises a remote validation module, wherein the remote validation module is designed and/or set up like the validation module of the monitoring device. The remote validation module is designed and/or set up to cause an evaluation of the monitoring image and/or the application of the evaluation application to the monitoring image based on the detection of a scene event by the evaluation application, for example to cause the transmission module to provide the monitoring image to the remote module. This is based on the consideration that in the event of the detection of a specific event, for example the recognition of a person based on the image model, the original data, here the surveillance image, is evaluated and/or initiated for verification purposes.

A further subject of the invention is a method in which at least one surveillance image showing and/or comprising a surveillance scene is recorded. Based on the surveillance image and/or in the surveillance image, objects are detected, with the object information data being determined and/or determined for the detected objects as previously described. Based on the surveillance images and/or for the surveillance image, an image model is ascertained and/or determined, wherein the image model comprises and/or forms a scene model of the surveillance scene and/or comprises the object information data.

A further subject of the invention is a computer program, the computer program comprising a program code. The computer program is designed and/or set up to execute, apply and/or implement the method steps when it is executed on a computer, the monitoring device and/or the monitoring arrangement.

A further subject of the invention is a storage medium, in particular a machine-readable storage medium, the computer program being stored on the storage medium.

• 1 schematische Darstellung eines Ausführungsbeispiels einer Überwachungsvorrichtung;
• 2 weiteres Ausführungsbeispiel umfassend ein Validierungsmodul;
• 3 Ausführungsbeispiel einer Überwachungsvorrichtung als Teil einer Überwachungsanordnung.

Preferred and/or advantageous embodiments result from the attached figures and their description. Show:

• 1 schematic representation of an exemplary embodiment of a monitoring device;
• 2 further exemplary embodiment comprising a validation module;
• 3 Embodiment of a monitoring device as part of a monitoring arrangement.

1 shows an exemplary embodiment of a monitoring device 1. The monitoring device 1 is designed to monitor an indoor or outdoor area, in particular for video- and / or image-based monitoring of the area, for example for tracking, recognizing people or taking on other monitoring tasks. For this purpose, the monitoring device 1 includes a camera 2. The camera 2 is designed, for example, as a pivotable and/or zoomable camera, in particular as a color camera. The camera 2 is arranged and designed to record and provide a surveillance image 3, in particular the surveillance image 3 is part of a video, the video being understood as a sequence of surveillance images 3. The surveillance image 3 shows a surveillance scene 4, the surveillance scene 4 being part of the surveillance area. The surveillance scene 4 shows and/or includes a plurality of objects 5a-e, which belong to different objects, in particular different object types or categories. In particular, the surveillance scene 4 and the surveillance image 3 include a background.

The monitoring device 1, in particular the camera 2, includes a first computing unit 6. The first computing unit 6 includes a discrimination module 6a and an image model determination module 6b. The surveillance image 3 is provided to the discrimination module 6a. The discrimination module 6a is designed to apply object recognition to the surveillance image 3 or the video. The discrimination module 6a detects, discriminates and/or classifies the objects 5a-e based on the surveillance image 3, for example based on a neural network and/or other algorithms and/or methods of object recognition. Furthermore, the discrimination module 6a is designed to classify, identify and/or track the objects 5a-e based on the surveillance image 3.

The discrimination module 6b is designed to determine object information data 7a-e for the detected objects 5a-e, surveillance images 3. The object information data 7a-e are preferably alphanumeric data and/or metadata of the surveillance image 3 and/or the objects 5a-e shown in the surveillance image 3. For example, the object information data 7a-e describe and/or form the results of the object recognition, for example outputs of a layer of the applied neural network.

The object information data 7a-e include an object class, the object class describing which class the associated object 5 belongs to, for example by assigning a class name. In the present example, the object classes are the Person class, Object 5b is the Dog class, Object 5c is the Plant class, Object 5d is the Car class and Object 5e is the Bucket class. Furthermore, the object information data 7a-e includes an object ID 8a-8e. The object ID is, for example, a unique number for the object 5a-e and/or for the data record of the object information data 7a-e. Furthermore, the object information data 7a-e each includes object geometry information 9a-e for the associated object 5a-e. The object geometry information 9a-e is determined and/or designed, for example, as a bounding box, further comprising information about the orientation, position and/or distances to other objects from the object geometry information 9a-e are. For example, the object information data 7a-e for the object 5a include and/or describe the size, shape, orientation and position of the bounding box for the object 5a, the information that it is a person as object classes and the object ID 8a- e carries and/or includes.

The image model determination module 6b is designed to determine an image model 11. The image model determination module 6b determines the image model 11 based on the surveillance image 3 and the object information data 7a-e. The image model 11 forms a model and/or an abstraction of the surveillance scene 4 of the surveillance image 3, in which the objects 5a-e are abstracted in the form of the object information data 7a-e. Instead of the image information, the image content and/or the individual pixels of the surveillance image 3, in particular the objects 5a-e, the objects 5a-e and/or their object areas are included and/or represented as metadata in the image model 11, the metadata comprising the object information data . In particular, the image model 11 forms and/or describes a scene model 12 in which the essential contents of the surveillance image 3 are represented and/or reproduced, in particular distances and/or size ratios of objects 5a-e, which are reproduced and preserved, objects 5a- e are described abstractly based on the object class and/or the object type and the objects 5a-e can be clearly identified using their IT.

The monitoring device 1, in particular the camera 2, comprises a second computing unit 13. The second computing unit 13 comprises an event detection module and is designed to execute and/or use at least one evaluation application. The evaluation application is, for example, an application for image-based monitoring and/or image evaluation. For example, the evaluation application is designed to determine whether a person or an object 5a-e enters a definable area based on the surveillance image 3 and/or the image model 11. For this purpose, the event detection module applies the evaluation application to the image model 11 and determines the variables to be determined, for example entering an area. Instead of applying the evaluation application to the monitoring image 3 or with parallel application and/or execution of evaluation applications by the application module to the monitoring image 3, the application module applies the evaluation application and/or the evaluation applications to the image model 11, so that the basic procedural steps are carried out via the Object detection, classification and/or tracking have already been carried out.

The first and second computing units 6, 13 are via a feedback connection 14 for transmitting data, in particular for transmitting the image model 11 and feedback data. The feedback data is provided and/or sent by the second computing unit 13, in particular the event detection module, to the first computing unit 6, in particular to the discrimination module 6a.

In particular, the second computing unit 13, in particular the event detection module, is designed to execute and/or use a plurality of evaluation applications for determining events, in particular for determining different events. For example, the event detection module uses a first evaluation application to determine whether a boundary and/or line has been exceeded as a first event and uses a second evaluation application to determine the behavior of a person as a second event. The first evaluation application requires people, animals and vehicles as detected and/or discriminated objects 5a-e, whereby the second evaluation application requires 5a-e people as detected and/or discriminated objects. For this purpose, the second computing unit 13 and/or the event detection module provides feedback data for the two evaluation applications.

2 shows an exemplary embodiment of the monitoring device 1, in particular based on and/or based on the exemplary embodiment 1 In contrast to this, the second computing unit 13 includes a validation module 15. The discrimination module 6 and the image model determination module 6b are as in 1 trained and designed to determine the image model 11. The event determination module determines an event by applying the evaluation application to the image model 11. The event is, for example, the detection of the occurrence and/or existence of a constellation of objects 5a-e. For example, the evaluation application is designed to detect the entry and/or entry into an area by an object 5a-e, the entry and/or entry forming an event. The occurrence of the event is provided and/or communicated to the validation module 15, whereby the validation module 15 triggers or is triggered to request and evaluate the monitoring image 3 when the event occurs. The validation module 15 then applies the evaluation application, in particular a special module included in the evaluation application, to the monitoring image 3 and determines validation data. By applying the evaluation application to the surveillance image 3, the more computationally intensive evaluation of the actual image data can only be carried out and/or applied when suspected that the event has occurred. The validation data includes, for example, the confirmation and/or the denial of the existence of the event and/or additional information and/or data about the scene event.

3 shows a monitoring arrangement 16 comprising the monitoring device 1 and a remote module 17. The monitoring device 1 is based in particular on the exemplary embodiment 1 and 2 , in contrast to this, the application module or the extended application module 14 is not part of the camera 2, but rather part of the remote module 17, for example a cloud.

The camera 2 is again designed to record the surveillance image 3 and includes the discrimination module 6a and the image model determination module 6b for determining the image model 11. The camera 2 includes a transmitting module 18 and the remote module 17 includes a receiving module 19. The transmitting module 18 and the receiving module 19 are connected to one another in terms of data technology, the data connection in particular allowing and/or carrying out an encrypted, typographical and/or compressed data transmission.

The transmitting module 18 is designed to provide the image model 11 to the remote module 17, in particular to the receiving module 19, in particular to provide it in encrypted and compressed form. The receiving module 19 provides the image module 11 to the second computing unit 13. The second computing unit 113 is in turn designed to carry out further evaluations and/or analyzes based on the image model 11. When an event is determined by the event detection module, the camera 2, in particular the transmitting module 18, does so Monitoring image 3 requested, the receiving module 19 providing the monitoring image 3 to the validation module 15 and the validation module 15 being designed to apply the evaluation application to the monitoring images 3 and to determine the validation data.

Claims (15)

Monitoring device (1) with a camera (2) for recording at least one surveillance image (3), a surveillance scene (4), wherein the camera (2) has a first computing unit (6) and a second computing unit (13), the first and second computing units (6, 13) being connected for data purposes via a feedback connection (14), wherein the first computing unit (6) has a discrimination module (6) and an image model determination module (6b), wherein the discrimination module (6a) is designed to detect, discriminate and/or classify objects (5a-e) in the surveillance image (3) and to determine object information data (7a-e) for the objects (5a-e), whereby the Object information data (7a-e) include an object class, an object type, an object ID (8a-e) and / or object geometry information (9a-e), wherein the image model determination module (6b) is designed to determine an image model (11) based on the surveillance image (3), the image model (11) comprising a scene model (12) of the surveillance scene (4) and the object information data (7a-e), wherein the second computing unit (13) has an event detection module and the second computing unit (13) is provided with the image model (11), wherein the event detection module is designed to evaluate the image model (11) and, based on the evaluation of the image model (11), events in the To detect surveillance scene (4), wherein the second computing unit (13) and/or the event detection module is designed to provide feedback data to the first computing unit (6) via the feedback connection (14). Monitoring device (1). Claim 1 , characterized in that the discrimination module (6a) is designed to track the detected objects (5a-e) and to provide tracking data, the object information data (7a-e) comprising the tracking data. Monitoring device (1). Claim 1 or 2 , characterized in that the image model (11) and/or the scene model (12) forms and/or comprises a symbolic representation of the surveillance scene (4). Monitoring device (1) according to one of the preceding claims, characterized in that the image model (11) and/or the object information data (7a-e) include the detected objects (5a-e) as symbolic representatives. Monitoring device (1) according to one of the preceding claims, characterized in that the second computing unit (13) and/or the event detection module is designed to adapt the first computing unit (6) and/or the discrimination module (6a) based on the feedback data and/or to configure, in particular for the discrimination, detection and / or classification of a selection of objects (5a-e). Monitoring device (1) according to one of the preceding claims, characterized in in that the event detection module is designed to execute an evaluation application for evaluating the image model (11) and/or for determining the events, wherein the second computing unit (13) and/or the event detection module is designed to receive the feedback data during registration, first execution and/or To provide initialization of the evaluation application and/or to cause the feedback data to be provided in response to the provision of the image model (11) by the first computing unit (6). Monitoring device (1). Claim 6 , characterized by a validation module (15), wherein the event detection module is designed to detect the event based on the application of the evaluation application to the image model (11), wherein the validation module (15) is designed at and / or based on the detection of Event to apply the evaluation application to the monitoring image (3) and to determine validation data. Monitoring device (1). Claim 6 or 7 , characterized in that the evaluation application comprises a basic module for application to the image model (11) and a special module for application to the surveillance image (3). Monitoring device (1) according to one of the preceding claims, characterized in that the feedback connection (14) is designed as a bidirectional connection for transmitting the image model (11) and the feedback data. Monitoring arrangement (16) with the monitoring device (1) according to one of the preceding claims and a remote module (17), wherein the monitoring device (1) comprises a transmission module (18) for providing the image model (11) to the remote module (17), wherein the Remote module (17) comprises a remote application module, wherein the remote application module is designed to apply an evaluation application to the image model (11). Monitoring arrangement (16). Claim 10 , characterized in that the transmission module (18) is designed to send the image model (11) in encrypted and/or compressed form to the remote module (17). Monitoring arrangement (16) according to one of the Claims 10 or 11 , characterized in that the remote application module comprises a remote validation module, wherein the remote validation module is designed to initiate an evaluation of the monitoring image (3) and/or the application of the evaluation application to the monitoring image (3) when an event is detected. Method, wherein at least one surveillance image (3) showing and/or comprising a surveillance scene (4) is recorded, wherein objects (5a-) are recorded by a first computing unit (6) based on the surveillance image (3) and/or in the surveillance image (3). e) are detected and object information data (7a-e) are determined and/or determined for the detected objects (5a-e), wherein from the first computing unit (6) based on the surveillance image (3) and/or for the surveillance image (3 ) an image model (11) is determined and/or determined, wherein the image model (11) comprises and/or forms a scene model (12) of the surveillance scene (4) and includes the object information data (7a-e), wherein the image model (11) a second computing unit (13) is provided and the second computing unit (13) evaluates the image model (11) and detects events in the surveillance scene (4) based on the evaluation of the image model (11), wherein the second computing unit (13) of the first computing unit (6) provides feedback data via the feedback connection (14). Computer program comprising a program code, wherein the computer program is designed and/or set up, when executed on a computer, the monitoring device (1) and/or the monitoring arrangement (16), according to the method and/or the method steps Claim 12 to carry out, apply and/or implement. Machine-readable storage medium, with the computer program on the storage medium Claim 13 is stored.

Images