Detection device for detecting an object and / or a person, as well as method, computer program, and storage medium

Abstract

A detection device 1 for detecting an object and / or a person 3, 3a, b, c in a monitoring region 2 monitored with a plurality of cameras 4, 4a, b, c, having a plurality of analysis modules 6, 6a, b, c,

Description

Prior art The invention relates to a detection device for detecting an object and / or a person in a monitoring region, wherein the monitoring region is monitored with a plurality of cameras. Based upon monitoring data of the cameras, characteristics for detecting the object and / or the person are determined. Methods of image-based detection of objects or persons are found in many fields of the art. For example, when monitoring airports, train stations, and / or other public areas, persons can be detected, recognized, and / or tracked based upon the recordings of cameras. For example, the video or image captures provided by the cameras are examined for characteristics that distinguish the person and / or object. For example, publication DE 10 2019 212 978 Al, which is most likely the closest prior art, describes a monitoring apparatus for person recognition in a monitoring region, wherein image sequences of the monitoring region are captured by the monitoring region with at least one camera. The monitoring images comprised by the image sequences are examined for the presence of at least one recognition characteristic, wherein the person is deemed to be recognized when the characteristic(s) have been found. The camera for capturing the monitoring images objectively comprises a fish eye. Disclosure of the invention A detection device having the features of claim 1 is proposed. Further, a method for detecting an object and / or a person in a monitoring region as well as a computer program and a storage medium having the computer program are proposed. Further advantages, effects, and embodiments result from the sub-claims, the description, and the attached figures. The invention relates to a detection device for detecting an object and / or for detecting a person. Detection is understood to mean, in particular, a recognition of an already known and / or stored object and / or person. Recognition is understood to mean, for example, the locating of the object and / or the person based upon distinguishing characteristics, wherein the distinguishing characteristics comprise, for example, shape, contrast, clothing, size, and / or other metadata. The detection of the object and / or the person is carried out in a monitoring region, wherein the monitoring region is monitored by means of a plurality of cameras, in particular in terms of imaging and / or video technology. The monitoring region can be an exterior region or an interior space. The monitoring region is video-monitored with a plurality of cameras. In particular, the monitoring region is monitored with at least one and preferably more than 20 or 100 cameras. The cameras are arranged in order to monitor all or a sub-region of the monitoring region, and in particular the cameras are arranged in order to monitor and / or capture the monitoring region from different perspectives, angles of view, and / or positions. Hie cameras are configured in order to provide monitoring data, in particular to provide it externally. The plurality of cameras can be part of the detection device, or, alternatively, the plurality of cameras is not an actual part of the detection device, but rather only data-linked and / or electronically linked to the detection device. The respective monitoring data of a camera includes monitoring images and / or monitoring videos for the monitoring region and / or part of the monitoring region. For example, the monitoring data includes monitoring images and / or monitoring videos of objects and / or persons in the monitoring region and / or the respective part of the monitoring region. The monitoring data is provided to the detection device. The detection device comprises a plurality of analysis modules. The analysis modules are data-linked to the cameras. In particular, each analysis module is data-linked to exactly one, at least one, or more cameras. Particularly preferably, precisely one camera is associated with each analysis module. The cameras are configured in order to provide the monitoring data to the respective analysis module. The analysis modules form separate modules and can be pooled into a common super-module, e.g. housing or computer. Separate modules can be understood, in particular, as separate software modules or electronic components. The analysis modules preferably operate without direct coupling between the analysis modules. The analysis modules preferably respectively form and / or comprise a neuronal network and / or an artificial intelligence system. In particular, the neuronal networks are configured as deep neuronal networks and / or as convolutional neuronal networks. In other words, the detection device comprises a plurality of neuronal networks. For example, the analysis modules and / or their neuronal networks are configured and / or trained to determine different characteristics based upon the respective monitoring data provided. The detection device comprises a shared memory module, specifically an expandable shared memory module and / or a plurality of shared memory modules. In particular, the shared memory module forms a memory, preferably an expandable memory. The individual analysis modules are linked to the shared memory module or to the plurality of shared memory modules. Preferably, all analysis modules are data-linked to the one shared memory module. For example, data can be transmitted and / or requested from the analysis module to the shared memory module by means of the data linkage. In particular, the shared memory module is configured as a commonly used memory to which the individual analysis modules have access at the same time. The analysis modules are respectively configured in order to determine characteristics for the monitoring data provided, in particular for the respective timestamp, which are also called object and / or person characteristics. In particular, memory data is queried, requested, and / or retrieved from the analysis modules of the shared memory module. The analysis modules, in particular, retrieve associated memory data that is associated with the respective monitoring data and / or is relevant. For example, the analysis modules comprise connections between memory data and associated monitoring data, for example via the assignment of the camera and / or perspective. The analysis module is thus subsequently provided with the retrieved memory data. The analysis modules are respectively configured in order to determine the characteristics, in particular object and / or person characteristics, based upon the monitoring data, in particular monitoring videos and / or monitoring images, as well as the associated memory data. In particular, the analysis modules are configured in order to determine the monitoring data and the characteristics by means of the neuronal network, in particular with the memory data. Memory data is, for example, characteristics, monitoring data, and / or metadata, specifically a previous moment in time and / or results of other analysis modules of the detection device. Optionally, it can be provided that the object and / or the person is determined, recognized, and / or detected based upon the determined characteristics. The characteristics are in particular distinguishing, specific, and / or unique characteristics for detecting and / or differentiating the objects and / or the persons. The invention is based upon the consideration of carrying out the characteristic determination through the use of a plurality of analysis modules, specifically the plurality of principally separate neuronal networks, so that a particularly lean, fast-processing, trainable, and expandable architecture is enabled. In particular, the individual analysis modules themselves are relatively lean and / or require little memory, because they essentially only need to store the data that is currently in question and / or to be processed. The actual storage backup and / or data management is carried out externally in one or more shared modules. In particular, it is conceivable that the different analysis modules are respectively linked to different cameras, hr particular, the different analysis modules of the detection device respectively evaluate and / or process different monitoring data, monitoring images, and / or monitoring videos. In particular, the analysis modules process and / or determine the characteristics for different perspectives, positions, and / or viewing angles on the monitoring region. In particular, different characteristics for different viewing angles on the monitoring region and / or parts of the monitoring region are thus obtained by the analysis modules. Particularly preferably, the cameras are configured in order to jointly monitor and / or provide at least one same object and / or at least one same person as monitoring data. It is particularly preferred that a plurality of objects and / or a plurality of persons are arranged in the monitoring region. In particular, the cameras are configured in order to provide the monitoring data to the analysis modules for this plurality of objects and / or persons. For example, the cameras provide monitoring images and / or monitoring videos of the plurality of objects and / or the plurality of persons, preferably from different perspectives, to the analysis modules. The analysis modules are configured in order to determine characteristics for the plurality of objects and / or the plurality of persons, in particular to determine the characteristics for the different objects and / or different persons separately and / or specifically. The determined characteristics of the different objects and / or persons are provided to the shared memory module. Tire shared memory module is configured in order to store the provided characteristics for the respective objects and / or respective persons. In particular, the shared memory module is configured in order to store the provided characteristics in an object-specific and / or person-specific manner. For example, in the monitoring region, two persons are arranged, who are provided by the cameras as monitoring data to the analysis modules. One camera captures front images as monitoring data, another camera captures rear views of the subjects, wherein the analysis modules, in this case for example a first or second analysis module, determine characteristics separately for the two subjects, a first analysis module based upon front views separately for the two subjects and a second analysis module for the rear views, also separate. In the shared memory module, a storage of the determined characteristics for the individual objects is preferred, for example “first person” comprising front characteristics and rear characteristics and “second person” comprising front and rear characteristics. Particularly preferably, the analysis modules are configured in order to provide the determined characteristics, in particular object- and / or person-specifically, to the shared memory module. The provided and / or already stored characteristics are themselves retrievable as memory data from the shared memory module. For example, a first analysis module already provides memory data based upon first monitoring data to the shared memory module, wherein a further analysis module can retrieve it as memory data and / or can use it in order to determine the characteristic. This is based upon the consideration that characteristics that are perspective-independent are determinable in this way, because, for example, an analysis module that evaluates front images for a person can retrieve memory data that describes the object from a rear side. Preferably, the detection device comprises exactly one or at least one gallery module. The gallery module is data-linked to the shared memory module. Data is exchangeable between the gallery- module and the shared memory module. The gallery module is configured in order to store characteristic data sets. In particular, the gallery module includes stored and / or storable characteristic data sets. Tire characteristic data sets are in particular configured for objects and / or persons, and specifically they are specific to objects and / or persons. In particular, the characteristic data sets are configured in order to comprise specific characteristics for detecting the object and / or the person. In particular, the shared memory module is configured in order to send and / or transmit to the gallery module the characteristics that have been determined, provided, and / or stored for the persons and / or objects in the monitoring region. This embodiment is based upon the consideration that the determined, provided, and / or stored characteristics of the shared memory module can be specifically and / or permanently secured and stored in the gallery' module, in particular after reaching an agreement. For example, the gallery module is configured in order to store the characteristics provided by the shared memory module as a new characteristic set when no characteristic data set already exists for that object and / or person. Further, for provided characteristics of an already existing characteristic data set, the gallery module is configured, for example, in order to adapt the already existing characteristics in the data set to the newly provided characteristics and / or to change them. An efficient use of memory in the gallery module is thus achieved, for example. Optionally, it is provided that the data linkage between shared memory module and gallery module is based upon and / or uses soft addressing, add vectors, and / or delete vectors. In particular, the data linkage between the shared memory module and the gallery module uses keys (keys). For example, the galleiy module comprises a memory that is writable as memory matrix M with N rows and W columns. For example, read and write operations for the memory matrix are described by read vectors r and write vectors w, which are based upon weights wr(read) and weights ww (write): r = Yt=iM[i,*]wr [i], wherein * stands for all j=l, ,W. For example, for writing to the memory, an erase vector e is applied, and then the write vector is added: v: M[i,j]«-M[i,j] (1 - ww[i]e[ / ]) + ww [t] v[ / ]. One embodiment of the invention provides that the analysis modules are configured in order to determine characteristics for at least one common object and / or common person, wherein the determined characteristics are provided to the shared memory module, preferably for each object and / or person. For example, in the monitoring region, there are two different objects or persons, wherein the analysis modules for each of the two objects and / or persons separately determines the characteristics, wherein the analysis modules transmit the characteristics for each of the two objects and / or persons to the shared memory module. The analysis module is configured in order to achieve agreement regarding the characteristics of the individual objects and / or persons. For example, for a person and / or an object of two analysis modules, different characteristic values of a particular characteristic are determined, wherein the shared memory module is configured in order to achieve agreement on these different values, for example by means of averaging and / or prompting the analysis modules to determine the characteristics again, for example by providing both values and / or a disagreement notification. Hie shared memory module is preferably configured in order to provide the characteristics associated with a respective object and / or person, as an object and / or person data set to the gallery module. The gallery module is configured in order to change, supplement, or delete already stored characteristic data sets and / or store new characteristic data sets based upon the provided object and / or person data sets. For example, the gallery module already comprises a characteristic data set for a determined object, and the shared memory module provides an object data set for that object, wherein, when the provided characteristic values match the stored characteristics, the gallery module leaves the existing characteristic data set or adjusts and / or changes it in case of deviations. It is particularly preferred that the characteristic data sets are configured to be object-specific and / or person-specific. In particular, the characteristic data sets comprise identification means, for example an ID, numbering, and / or designation. The characteristic data sets of the respective object and / or the respective person respectively comprise characteristics that are based upon different perspectives and / or views on the object and / or the person and / or different timestamps. In particular, the characteristic data sets comprise characteristics of the respective object and / or person that are perspective-independent and / or consider and / or connect different perspectives. A further subject matter of the invention is a method for detection, in particular recognition of an object and / or a person in the monitoring region. From the monitoring region, with a plurality of sensors, in particular cameras and / or radar sensors and / or LIDAR sensors, and / or motion sensors, monitoring data is captured, wherein the monitoring data respectively comprises video data and / or image data and / or sensor data. Each of the sensors or cameras is linked to one of the analysis modules, in particular the neuronal network. The analysis modules determine characteristics based upon the provided monitoring data. Hie analysis modules are data-linked to a shared memory module, wherein the analysis modules transmit the determined characteristics to the shared memory module. In particular, the analysis modules can request and / or obtain data from the shared memory module, wherein this data is preferably used when determining the characteristics. In particular, the shared memory module is linked to a gallery module, wherein characteristics determined for an object and / or a person are provided to the gallery module and secured there, in particular after achieving agreement regarding the individual characteristics. A further subject matter of the invention is a computer program to be carried out on a computer, wherein the computer program is configured and / or set up in order to carry out the method as described above while in operation. A further subject matter of the invention is a storage medium, in particular a preferably non-volatile, machine-readable storage medium. The computer program is stored on the storage medium. Further advantages, effects, and configurations result from the attached figures and their description. The figures show: Fig. 1 Detection device having a monitoring region; Fig. 2 schematic data flow in the detection device. Fig. 1 schematically shows an exemplary embodiment of a detection device 1 for monitoring a monitoring region 2. In the monitoring region 2, a plurality of persons 3a, b, c are arranged. Persons 3a, b, c differ in their appearance, for example, size, clothing, sex, and age. The monitoring region 2 is monitored with a plurality of cameras 4a, b, c, in this case three. The cameras 4a, b, c respectively capture the monitoring region 2 in a different perspective 5a, b, c. The cameras 4a, b, c respectively take monitoring images and / or monitoring videos and provide them as monitoring data to an analysis module 6a, b, c. Each analysis module 6a, b, c evaluates the monitoring data of the assigned camera 4a, b, c. Analysis modules 6a, b, c comprise and / or are configured as a neuronal network. The neuronal network is configured in order to determine characteristics based upon the monitoring images and / or monitoring videos, generally the monitoring data. The determination of the characteristics is preferably carried out using memory data, wherein the memory data comprises, for example, characteristics of an object and / or a person 3a, b, c determined at a past prior point in time and / or determined [by] another analysis module 4a, b, c. The detection device 1 comprises a shared memory module 7. The shared memory module 7 is data-linked to the analysis modules 6a, b, c. The data link is in particular a bi-directional link, such that data is transmissible and / or requestable in both directions. The analysis modules 6a, b, c are configured in order to provide the characteristics determined based upon the monitoring data to the shared memory module 7. In particular, analysis modules 6a, b, c can request the memory data from the shared memory module. The shared memory module 7 is configured in order to cache the provided characteristics, in particular in an object-specific and / or person-specific manner. Preferably, the analysis modules 6a, b, c and / or the shared memory module 7 are configured in order to establish connections between characteristics to be determined, already determined characteristics of other analysis modules, and / or memory data. If characteristics determined by different analysis modules 6a, b, c influence one another, the shared memory module and / or analysis modules 6a, b, c are configured in order to produce agreement regarding the characteristics. After achieving agreement, the characteristics, in particular separately for each object and / or person, are provided to a gallery module 8. The gallery module 8 stores the particular characteristics, also called “features”, specifically for each object and / or each person 3a, b, c. The characteristics are in particular perspective-independent. For example, the characteristics for the individual objects and / or persons are stored as characteristic vectors (feature vectors). The individual characteristic vectors and / or characteristic collections for the individual objects and / or persons are provided with an identification 9, for example an identification number. By addressing, the characteristic vector of each object or person is addressed, retrieved, deleted, or changed. Fig. 2 schematically shows an example of the data flow in the detection device from Fig. 1. The analysis modules 6 - here, for the sake of overview, two 6a, b - are arranged in a common computer module 10 together with the shared memory module 7. The monitoring data, in this case monitoring images 1 la, b, is provided to this computer module 10. Tire analysis modules 6a, b respectively adopt the monitoring images 1 la, b that originate from the camera 4a, b to which they are assigned. The analysis modules respectively comprise a neuronal network that examines the monitoring images Ila, b for the presence of characteristics. For this purpose, memory characteristics, for example characteristics determined at the previous time stamp, can be retrieved from the shared memory module 7. The characteristics determined by the analysis modules 6a, b for the given timestamp are provided to the shared memory module 7. The shared memory module 7 transmits the characteristics for the respective object or person to the gallery module 8, in which the characteristics are specifically stored for each object and / or person.

Claims

1. A detection device (1) for detecting an object and / or a person (3, 3a, 3b, 3c) in a monitoring region (2) monitored with a plurality of cameras (4, 4a, 4b, 4c), comprising:a plurality of analysis modules (6, 6a, 6b, 6c), wherein the analysis modules (6, 6a, 6b, 6c) are respectively data-linked to different cameras (4, 4a, 4b, 4c), wherein the analysis modules (6, 6a, 6b, 6c) are respectively provided with monitoring data of the linked cameras (4, 4a, 4b, 4c) so that tire plurality of analysis modules (6, 6a, 6b, 6c) receive monitoring data for the monitoring region (2) from different perspectives (5, 5a, 5b, 5c); anda shared memory' module (7), wherein the analysis modules (6, 6a, 6b, 6c) are data-linked to the shared memory module (7);wherein the analysis modules (6, 6a, 6b, 6c) are respectively configured in order to retrieve memory data associated with the monitoring data of the monitoring region from a different perspective (5, 5a, 5b, 5c), provided by a different analysis module, from the shared memory module (7), andwherein the analysis modules (6, 6a, 6b, 6c) are respectively configured in order to determine characteristics, object characteristics, and / or person characteristics based upon the monitoring data and the memory data.

2. The detection device (1) according to claim 1, characterized in that, for at least two objects and / or persons (3, 3a, b, c) in the monitoring region (2), monitoring data from the cameras (5, 5a, b, c) is provided, wherein the analysis modules (6, 6a, b, c) are configured in order to determine the characteristics for the at least two objects and / or persons (3, 3a, b, c) and provide said characteristics to the shared memory module (7), wherein the shared memory module (7) is configured in order to store the provided characteristics specifically for the respective objects and / or persons (3, 3a, b, c).13 03 253. The detection device (1) according to any one of the preceding claims, characterized in that the analysis modules (6, 6a, b, c) are configured in order to provide the determined characteristics, object characteristics, and / or person characteristics to the shared memory module (7), wherein provided characteristics, object characteristics, 5 and / or person characteristics are retrievable as memory data.

4. The detection device (1) according to any one of the preceding claims, characterized by a gallery module (8), wherein the gallery module (8) comprises characteristic data sets for stored and / or storable objects and / or persons (3, 3a, b, c), wherein the gallery 10 module (8) is data-linked to the shared memory module (7).

5. The detection device (1) according to claim 4, characterized in that the shared memory module (7) is configured in order to provide the characteristics provided and / or stored for the respective objects and / or persons (3, 3a, b, c) to the gallery 15 module (8) for storage.

6. The detection device (1) according to claim 4 or 5, characterized in that the data linkage of the shared memory module (7) and the gallery7 module (8) is based upon and / or supported by soft addressing, add vectors, and / or remove vectors.

207. The detection device (1) according to any7 one of the preceding claims, characterized in that the analysis modules (6, 6a, b, c) are configured in order to determine characteristics for at least one common object and / or person (3, 3a, b, c) in the monitoring region (2) and provide them to the shared memory module (7), wherein 25 the analysis modules (6, 6a, b, c) are configured in order to achieve agreementregarding characteristics associated with the respective object and / or the respective person (6, 6a, b, c), wherein the shared memory module (7) is configured in order to provide the characteristics associated with the object and / or person (3, 3a, b, c) as an object and / or person data set to the gallery module (8).

308. The detection device (1) according to any one of claims 4 to 7, characterized in that the gallery module (8) is configured in order to change, supplement, or delete stored characteristic data sets and / or store new characteristic data sets based upon the provided object and / or person data sets.3513 03 259. The detection device (1) according to any one of claims 4 to 8, characterized in that the characteristic data sets are configured to be object-specific and / or person-specific, wherein the characteristic data sets comprise characteristics based upon different perspectives (5, 5a, b, c) and / or timestamps.

510. The detection device (1) according to claim 9, characterized in that the characteristic data set comprises perspective-independent characteristics.

11. The detection device (1) according to any one of the preceding claims, characterized 10 in that the characteristics determined by the analysis modules (6, 6a, b, c) formperspective-independent characteristics.

12. A method for detecting an object and / or a person (6, 6a, b, c) in a monitoring region (2), in particular by means of the detection device according to any one of the15 preceding claims, wherein provided monitoring data is processed by means of aplurality of analysis modules (6, 6a, b, c), wherein the analysis modules (6, 6a, b, c) determine characteristics, wherein the analysis modules (6, 6a, b. c) preferably respectively comprise a neuronal network and determine the characteristics by means of the neuronal network, wherein the analysis modules (6, 6a, b, c) share a shared20 memory module (7), wherein the analysis modules (6, 6a, b, c) provide thecharacteristics to the shared memory module (7).

13. A computer program, wherein the computer program is configured and / or set up in order to perform, apply, and / or support the method according to claim 12 while in25 operation.

14. A machine-readable, in particular non-volatile machine-readable, storage medium, wherein the computer program according to claim 13 is stored on the storage medium.

Images