Method for learning a model for person re-identification and method for identifying a person by the learned model

Abstract

The invention relates to a method 600 for learning a model for person re-identification using images from a camera 100, wherein the method comprises at least a step of reading in a detection signal 135 representing a person detected in a monitoring area of ​​at least a camera 100 of a camera network. The method further comprises at least a step of collecting a plurality of image signals 140 from the camera 100 using the read detection signal 135, the collected image signals 140 representing captured image segments in each image of the camera 100. Finally, the method comprises at least a step of adapting a model for person re-identification using the collected image signals 140 in order to re-identify the detected person in an image of the camera 100 or at least one second camera of the camera network.

Description

Technical Field

[0001] The invention relates to a method for learning for re-identifying persons using images from a camera, a method for identifying persons by means of a second camera of a camera network using a learned model for re-identifying persons, and a device using the method. The subject matter of the invention is also a computer program. Background Art

[0002] Monitoring systems for observing one or more monitoring areas are known, wherein at least one or more monitoring cameras are directed toward the one and / or multiple monitoring areas. In this case, significant objects (e.g., persons or vehicles) located in the monitoring area are tracked by the monitoring cameras. Image sequences and / or video sequences captured by the monitoring cameras are transmitted to a central unit (e.g., a monitoring center) and are analyzed and / or automatically analyzed there by monitoring personnel.

[0003] DE 10 2009 055 127 A1 describes a video monitoring system for detecting and tracking a significant object, wherein the object is detected by means of a detection device according to a predetermined model description and is tracked by a tracking device over a certain period of time, wherein the tracking device detects a plurality of tracking parameters and is coupled back to the detection device so that in the case of repeated detections the tracking parameters are provided to the detection device and are taken into account for the detection of the object. Summary of the invention

[0004] Against this background, a method for learning a model for re-identifying a person using images from a camera, a method for identifying a person using a second camera of a camera network using the learned model for re-identifying a person, a device using the method, and finally a corresponding computer program are proposed with the aid of the solution presented here. Advantageous expansions and improvements of the proposed device can be achieved by means of the measures listed in the description.

[0005] Using a plurality of recorded image sections of a camera, a model for person re-identification is learned and adapted in such a way that a person detected in an image section can also be re-identified and tracked in other cameras of the camera network.

[0006] A method is proposed for learning a model for person re-identification using images from a camera, wherein the method comprises the following steps:

[0007] reading a detection signal representing a person detected within a monitoring area of ​​at least a camera of the camera network;

[0008] collecting a plurality of image signals from the camera using the read detection signal, wherein the collected image signals represent a recorded image segment in each case one image of the camera;

[0009] Using the acquired image signals, a model for person re-identification is adapted in order to re-identify the detected person in the image of a camera or at least a second camera of the camera network.

[0010] The model may be a simplified representation of a process or a complex relationship. Preferably, the model is constructed as an artificial neural network, wherein the artificial neural network is arranged and connected in a predetermined network architecture, wherein the connection is used for information exchange. In this arrangement, the intensity of the information flow between two neurons can be changed by weighting. Preferably, the weighting of the connection of the neural network is changed in the adaptation of the model. The camera may be a photographic device that can record static or moving images on photographic film or electronically on a magnetic video tape or on a digital storage medium, or transmit the static or dynamic images via an interface. The camera network may be a networked security system with a plurality of statically arranged cameras, which are used to monitor a monitoring area. The monitoring area may be a certain usually public area or space (for example in an airport or a railway station), which is monitored by the camera network in order to lead to a continuous improvement of the security situation through the preventive and protective effects of video monitoring. The image segment may be a partial area of ​​a scene captured on a photographic image.

[0011] The advantage of the method proposed here is that, by determining at least one adjustment parameter for active camera adjustment, a model for person re-identification can be learned and trained in such a way that the amount of available data about persons detected in the monitoring area is artificially increased. Based on this artificially increased amount of data, more robust and lighting-independent features for person re-identification can be generated. As a result, for example, a detected person can be tracked seamlessly through the entire camera network without losing the identity of the person or confusing the person with another tracked person. In addition, the use of intelligent video and / or image content analysis simplifies the monitoring of critical facilities (such as railway stations, airports or public places), because intelligent algorithms can automatically detect and track people in the scene over time. Therefore, if a person, for example, stays in a critical or sensitive area, or if another deviation from normal behavior is detected, the monitoring system can automatically issue an alarm to security personnel.

[0012] According to one embodiment, specific person features of the detected person can be extracted from the collected image signals by means of an extraction step, wherein in particular an identifier for identifying the person is assigned to the specific person feature, wherein in an adaptation step the specific person feature is output to a camera of the camera network or to a second camera. Based on this identifier-linked assignment information in the specific person feature, a specific person can be uniquely and seamlessly tracked and identified by the entire camera network for monitoring a critical area.

[0013] According to one embodiment, by means of the determining step, preferably using the collected image segments, a control parameter can be determined for adjusting the recording characteristics of the camera or the second camera, wherein, in particular in the collecting step, the determined control parameter can be used for recording at least one further image signal. In this case, the control parameter is primarily used to adapt the recorded or to-be-recorded image segment to the currently measured exposure characteristics of the recorded or to-be-recorded scene, and can even be adapted according to the exposure characteristics expected in the future. By predictive adjustment of the camera or the second camera by means of the control parameter, various requirements in rapidly changing scenes are met. Preferably, the camera or the second camera is adjusted by means of the determined control parameter. In this case, in particular, the control parameter is changed stepwise and / or randomly within a predetermined adjustment range so that the image segment is recorded.

[0014] According to one specific embodiment, in the determining step, brightness and / or contrast and / or color representation can be identified as control parameters. By using or adapting the image segments recorded or to be recorded with respect to brightness, contrast and color representation, an increase in the data variability can be achieved, which leads to an improvement of the model for person re-identification. Preferably, in addition or as an alternative to the varied control parameters, the infrared cut filter of the camera can be disabled, and the image segment can be recorded with the infrared cut filter disabled, and the image segment is used for adapting the model for person re-identification.

[0015] According to one embodiment, in the step of collecting, image segments of the image signal can be collected along the movement trajectory of the detected person. The image segment captured along the movement trajectory is selected in such a way that specific features of the person can be extracted from the image segment, by means of which the detected person can also be uniquely assigned to another camera in the camera network.

[0016] A method is proposed for identifying a person by means of a camera of a camera network or a second camera using a learned model for person re-identification, wherein the method comprises the following steps:

[0017] corresponding to the steps of the method for learning a model for person re-identification using images from a camera;

[0018] detecting at least one second image signal, wherein the second image signal represents a second image segment of an image of the camera or of a second camera;

[0019] When using the model for person re-identification, the detected person is identified by the camera of the camera network or the second camera in the read-in second image signal.

[0020] According to one embodiment, in a step corresponding to a method for learning a model for re-identifying a person using images from a camera, a specific person feature of a person detected in the collected image signal can be output to the camera or a second camera, wherein in the step of identification, a person who has been detected by the first camera is identified by the specific person feature in the second image signal read by the camera itself or the second camera. Based on the allocation information associated with the identifier in the specific person feature, the determined person can be uniquely and seamlessly tracked and identified by the entire camera network for monitoring the critical area.

[0021] According to one specific embodiment, in a step corresponding to the method for learning a model for person re-identification using images from a camera, the determined control parameters can be output to the camera or the second camera, wherein in the step of detecting, the determined control parameters are used for the second image segment detected by the camera or the second camera. By predictive control of the camera using the control parameters, various requirements in rapidly changing situations can be reliably met, and a person already detected by the first camera can also be uniquely identified by the second camera of the camera network.

[0022] According to one embodiment, in a method for learning a model for person re-identification using images from a camera and / or in a method for recognizing a person by means of a second camera of a camera network using a learned model for person re-identification, a first camera and a second camera of the camera network can be connected to a computing unit, wherein the step of collecting and / or the step of recognizing are particularly performed on the computing unit. In this case, a plurality of image segments captured by the first camera and / or the second camera are collected on the computing unit in order to learn a model for person re-identification based on person features extracted from the image segments.

[0023] One or more of the methods presented here can be implemented, for example, in software or in hardware or in a hybrid form of software and hardware (for example in an apparatus or a control unit).

[0024] The solution proposed here also provides an apparatus, which is designed to execute, control or implement the steps of the variant of the method proposed here in a corresponding device. The object based on the invention can also be achieved quickly and efficiently by the embodiment variant of the invention in the form of an apparatus.

[0025] For this purpose, the device can have at least one computing unit for processing signals or data, at least one storage unit for storing signals or data, at least one interface to a sensor or actuator for reading a sensor signal from a sensor or for outputting a data signal or a control signal to an actuator and / or at least one communication interface for reading or outputting data embedded in a communication protocol. The computing unit can be, for example, a signal processor, a microcontroller or the like, wherein the storage unit can be, for example, a flash memory, an EPROM or a magnetic storage unit. The communication interface can be designed for wireless and / or wired reading or output of data, wherein a communication interface that can read or output wired data can, for example, electrically or optically read the data from a corresponding data transmission line or output the data to a corresponding data transmission line.

[0026] Here, a device can be understood as an electrical device that processes sensor signals and outputs control signals and / or data signals according to the sensor signals. The device can have an interface that can be constructed in hardware form and / or software form. In the hardware form, the interface can be, for example, part of a so-called system-specific integrated circuit, which contains various functions of the device. However, it is also possible that the interface is an independent integrated circuit or is at least partially composed of discrete components. In the software form, the interface can be a software module, which, for example, exists on a microcontroller with other software modules.

[0027] Also advantageous is a computer program product or a computer program with a program code, which can be stored on a machine-readable carrier or storage medium, such as a semiconductor memory, a hard disk memory or an optical memory and is used to execute, implement and / or control the steps of the method according to any of the embodiments described above, in particular when the program product or the program is executed on a computer or a device. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] An exemplary embodiment of the solution proposed here is shown in the drawings and is explained in more detail in the following description. The drawings show:

[0029] Figure 1 A schematic diagram of a first camera having a device for learning a model for person re-identification using images from the camera according to an exemplary embodiment is shown;

[0030] Figure 2 A schematic diagram of a second camera having a device for identifying a person by means of a learned model for person re-identification is shown according to an exemplary embodiment;

[0031] Figure 3 A diagram is shown for explaining the significance of active camera control with the aid of control parameters in the case of person re-identification according to an exemplary embodiment;

[0032] Figure 4 A diagram is shown for explaining the significance of active camera control with the aid of control parameters in the case of person re-identification according to an exemplary embodiment;

[0033] Figure 5 A simplified system structure for using a method for learning a model for person re-identification and a method for identifying a person by the learned model for person re-identification according to one embodiment is shown;

[0034] Figure 6 A flow chart of an exemplary embodiment of a method for learning a model for person re-identification using images from a camera according to an exemplary embodiment is shown;

[0035] Figure 7 A flow chart of an exemplary embodiment of a method for identifying a person using a learned model for person re-identification according to an exemplary embodiment is shown.

[0036] In the following description of advantageous exemplary embodiments of the invention, identical or similar reference numerals are used for elements which are shown in the various figures and act similarly, wherein a repeated description of these elements is omitted. DETAILED DESCRIPTION

[0037] Figure 1 A schematic diagram of a first camera with a device 105 according to an exemplary embodiment is shown for learning a model for person re-identification using images from the camera 100. The device 105 for learning a model for person re-identification has a reading device 110, a collecting device 115, a determining device 120, an extraction device 125 and an adaptation device 130.

[0038] The reading device 110 is designed to read a detection signal 135, wherein the detection signal 135 represents a person detected in the monitoring area of ​​at least the camera 100 of the camera network. In addition, the collecting device 115 is designed to read a plurality of image signals 140 from the camera 100 and to collect the plurality of image signals using the detection signal 135 that has been read by the reading device 110, wherein the collected image signal 140 represents a recorded image segment in each image of the camera 100. In addition, the collecting device 115 is designed to collect image segments of the image signal 140 along the movement trajectory of the detected person. According to an exemplary embodiment, the image signal 140 can also be collected on an externally arranged computing unit to which the camera 100 is connected. The determining device 120 is designed to determine a control parameter 145 using the collected image signal 140 or the image segment represented by the image signal 140, which control parameter is used to adjust the recording characteristics of the camera 100. In this case, brightness and / or contrast and / or color representation can be identified as the control parameter 145. In this case, the collection device 115 is then also configured to record at least one further image signal 140 using the identified control parameters 145. The extraction device 125 is configured to extract specific person features 150 of the detected person from the collected image signals 140. In addition, an identifier for identifying the person is assigned to the specific person features 150. Finally, the adaptation device 130 is configured to adapt the model for person re-identification using the collected image signals 140 in order to re-identify the detected person in at least a second camera of the camera network. The adaptation device 130 is also configured to output the specific person features 150 and the control parameters 145 to the second camera of the camera network in advance.

[0039] Figure 2 A second camera 200 is shown with a device 205 for recognizing a person by means of a learned model for person re-identification according to an exemplary embodiment. The device 205 for recognizing a person by means of a learned model for person re-identification via the second camera 200 of the camera network has a detection device 210 and a recognition device 215 .

[0040] The detection device 210 is designed to detect at least one second image signal 220, wherein the second image signal 220 represents a second image segment in the image of the second camera 200. The detection device 210 is also designed to detect the determined control parameters 145 and to use them for the detected second image segment, wherein, when using the collected image segment, the control parameters 145 are determined by Figure 1The adapter of the device in the apparatus outputs the determined adjustment parameter 145 to the second camera 200, and the adjustment parameter is used to adjust the shooting characteristics of the second camera 200. Figure 1 The device in the embodiment is used to learn a model for person re-identification. Finally, the detection device 210 is configured to detect the extracted specific person feature 150. The recognition device 215 is configured to recognize the detected person on the read second image signal 220 by the second camera 200 of the camera network using the model for person re-identification. Here, the recognition device 215 is particularly configured to recognize the detected person based on the specific person feature 150 in the detected second image signal 220, wherein the specific person feature 150 is determined by the detection device 210. Figure 1 The adapter device of the device in the embodiment outputs a specific person feature 150 to the second camera, wherein the specific person feature is used to identify the detected person from the collected image signal, Figure 1 The device in is used to learn a model for person re-identification. According to one embodiment, the detected person can also be identified on an externally arranged computing unit based on specific person features in the detected second image signal 220, to which the second camera 200 is connected.

[0041] Figure 3 A diagram for explaining the significance of active camera control with control parameters in the case of person re-identification according to an exemplary embodiment is shown. Here, the diagram has four recorded image segments 305, 310, 315, 320, which were recorded by a camera at four different times without active camera control.

[0042] In the image segments 305, 310, 315, 320, a person 325 detected by the camera can be seen, which person is tracked along a movement trajectory 330 over a certain period of time. Now, in order to learn a model that can be used for person re-identification, a plurality of image segments 305, 310, 315, 320 of the detected person are selected and collected along the movement trajectory 330 of the person 325 in order to extract or learn specific features of the person 325 in these collected image segments 305, 310, 315, 320. The enlarged left image 340 shows the overall scene of the superposition of the four recorded and collected image segments 305, 310, 315, 320. The four image segments 305, 310, 315, 320 were each recorded at four different times, wherein the image segment 305 was first recorded at the time t-3, the image segment 310 was then recorded at the time t-2, and then the image segment 315 was recorded at the time t-1. Image segment 320 recorded at time t shows the most current image segment.

[0043] In addition, Figure 3345 for activating camera control is shown, according to one exemplary embodiment, the x-axis of which represents a brightness parameter and the y-axis of which represents the time profile. A graph 350 of the function 345 here represents the deactivation of the camera control, wherein it is clear that the graph 350 is shown as a straight line over the time profile, since the camera control remains unchanged for all four image segments 305, 310, 315, 320.

[0044] like Figure 3 As shown, image segments 305, 310, 315, 320 are identical. Therefore, in a camera without active camera control, only a limited number of image segments of person 325 can be used, since the variability in the data is low, especially in scenes with constant lighting. Accordingly, three image segments 310, 315, 320 are deleted, since these additional image segments 310, 315, 320 have no added value at times t-2, t-1 and t. Therefore, only image segment 305 at time t-3 can be used. Conventional constant camera control is based on the assumption that the situation does not change significantly from image to image. Therefore, it is assumed that the camera is set for Figure 3 The control parameters calculated for the scene shown in the image remain valid for the following images. However, this assumption loses its validity in applications where rapidly changing scenes occur (e.g. when tracking a fleeing person). As a result, lags occur in conventional camera control, so that the image sensor of the camera can no longer correctly describe the changing lighting characteristics.

[0045] In addition, in the point-by-point interference event (punktuelle ) is problematic in situations where a group of people are walking in front of the camera. Therefore, there is a need for active camera adjustment which allows a satisfactory transition of the captured scene in an image segment even in a rapidly changing scene.

[0046] Another problem is that large differences in viewpoint and lighting between different camera views can severely change the appearance of person 325, which makes it difficult to re-identify person 325 in another camera of the camera network. Due to different lighting properties and / or different camera views and / or occlusions of person 325 and / or the only occasionally rotationally symmetric appearance of person 325 and due to the limited available training data, a great challenge in order to generate a robust model is the high variability in the appearance of persons.

[0047] Figure 4A diagram for explaining the significance of active camera control with control parameters in the case of person re-identification according to an exemplary embodiment is shown. The diagram has four recorded image segments 305, 310, 315, 320, which were recorded by a camera with active camera control at four different times.

[0048] In the image segments 305, 310, 315, 320, a person 325 detected by the camera can be seen, which person is followed along a movement path 330 over a certain period of time. The enlarged left image 340 shows the overall scene of the superposition of the four recorded and collected image segments 305, 310, 315, 320. The four image segments 305, 310, 315, 320 were each recorded at four different times, wherein the image segment 305 was first recorded at the time t-3, the image segment 310 was then recorded at the time t-2, and the image segment 315 was then recorded at the time t-1. The image segment 320 recorded at the time t shows the most current image segment.

[0049] In addition, Figure 4 , a function 345 for activating camera control is shown, wherein the x-axis of the function represents a brightness parameter and the y-axis of the function represents the time course according to one exemplary embodiment. A graph 350 of the function 345 here represents the activation of the camera control, wherein it is evident that the camera control is actively varied over the time course for each of the four image segments 305, 310, 315, 320.

[0050] If a person 325 enters the field of view of a camera of the camera network for the first time, a plurality of image segments 305, 310, 315, 320 of the person 325 are stored and collected in a computing unit or in a collection device of a device for learning a model for person re-identification, wherein specific person features of the person 325 are extracted from these collected image segments 305, 310, 315, 320. In addition, a unique and unambiguous identifier is assigned to the specific person feature, wherein the identifier can be a number and / or a letter combination. Based on these person data, the model for person re-identification is learned and adapted in order to re-identify the same person 325 in another camera of the camera network and to associate the person with the originally predefined specific person features and the identifier.

[0051] By means of active camera control, in each recorded image segment 305, 310, 315, 320, the recording characteristics of the camera are changed using determined control parameters so that new data are generated at times t-2, t-1 and t for learning a model for re-identifying a person. The control parameters can be parameters for adapting the brightness and / or contrast and / or color representation to the image segments 305, 310, 315, 320. It can be easily seen that Figure 4 The color of the people's clothes is slightly changed because the color representation is affected by the camera adjustment.

[0052] By actively controlling the camera setting, the data variability along the movement trajectory 330 of the person 325 can also be significantly increased. By artificially augmenting the data, a more robust model for person re-identification can be trained with the aid of a neural network, or the number of image segments 305, 310, 315, 320 for the person 325 in the computing unit can be significantly increased. Actively controlling the camera setting has a direct effect on the representation of the camera image. Due to the interruption of the active camera setting, meaningful data augmentation can no longer be achieved, which significantly reduces the accuracy of the person re-identification.

[0053] The camera can now also be adjusted in such a way that it switches to a special night mode. In this mode, although the image is switched to grayscale, the infrared cutoff filter is disabled. This has the advantage that wavelengths that are not visible to the person are also captured in the image. These images then have appearance-based features of the person 325 that cannot be perceived by humans but represent additional information for the re-identification of the person and thus increase the information content.

[0054] Figure 5 A simplified system structure of a method for learning a model for person re-identification and a method for identifying a person by means of a learned model for person re-identification according to an embodiment is shown. The system structure has a first camera and a second camera 200, which are arranged in a camera network 505 and connected to each other, and an externally arranged computing unit 510, to which the first camera and the second camera 200 are connected.

[0055] The first camera and the second camera 200 are both directed toward the first monitoring area 515 and the second monitoring area 520 to be monitored. In the first monitoring area 515 and the second monitoring area 520, a significant person 325 is detected using the detection signal 135, and the significant person is tracked by the first camera and the second camera 200 according to the movement trajectory 330 of the person 325. The first camera captures a plurality of image signals 140, wherein the image signals 140 represent a plurality of image segments in the image 340 of the first camera. Here, the image segments are captured along the tracked movement trajectory 330 of the person 325. The captured image segments are transmitted to an externally arranged computing unit 510, where the image segments are stored and collected. Using the collected image segments, control parameters 145 for adjusting the recording characteristics of the camera 100 are determined. Here, brightness and / or contrast and / or color representation are determined as control parameters 145. In the following, the determined control parameters 145 are used to capture at least one further image signal 140. In addition to the determination of the control parameters 145, specific person features 150 of the person 325 are extracted using the collected image segments, and a unique and unambiguous identifier is assigned to the person 325. Based on the person data, a model for person re-identification is now learned and adapted so that as soon as the person 325 is in the monitoring area of ​​the second camera 200, the same person 325 is also re-identified in the second camera 200 of the camera network 505 and the person is associated with the originally predefined specific person features 150 and the identifier. For this purpose, the control parameters 145 and the specific person features 150 of the detected person 325 are output to the second camera 200 of the camera network 505. The second camera 200 detects a second image signal 220, which represents a second image segment in the image 540 of the second camera 200. The determined control parameters 145 are used for the second image segment, and the specific person features 150 of the detected person 325 are identified on the read-out second image signal 220. Using the model for person re-identification, detected person 325 is therefore uniquely identified by second camera 200 .

[0056] It is also conceivable that the second image signal 220 is recorded by the first camera (for example at a later time) and used to identify persons in the area monitored by the first camera (at a corresponding later time). Therefore, it is not necessary to adapt the model only for the case where a person is identified in the image captured by the second camera 200 when using the data of the camera 100, but the model can also be adapted again for the recognition of persons from the image section of the image captured by the camera 100.

[0057] Figure 6 A flow chart of an exemplary embodiment of a method 600 for learning a model for person re-identification using images from a camera according to an exemplary embodiment is shown. Figure 1 The method 600 is used in the context of a device proposed in the invention for learning a model for person re-identification using images from a camera.

[0058] The method 600 first has a step 605, in which a detection signal is read out, which represents a person detected in at least the monitoring area of ​​a camera of the camera network. Next, the method 600 has a step 610, in which a plurality of image signals are collected from the camera using the read out detection signal, wherein the collected image signals represent the captured image segments in each of the images of the camera. Here, the image segments of the image signals are collected along the movement trajectory of the detected person. According to an exemplary embodiment, step 610 can also be performed on an externally arranged computing unit. In step 615 of the method 600, control parameters for adjusting the recording characteristics of the camera are determined using the collected image segments. Here, brightness and / or contrast and / or color representation can be identified as control parameters. Next, step 610 of the method 600 can be repeated in particular so as to use the determined control parameters for recording at least one further image signal. In a further step 620 of the method 600, specific person features of the detected person are extracted from the collected image signals, wherein in particular an identifier for identifying the person is assigned to the specific person features. Finally, method 600 has a step 625, in which a model for person re-identification is adapted using the collected image signals in order to re-identify the detected person in the image of the camera or at least a second camera of the camera network, wherein specific person features are also output to the camera or the second camera of the camera network in step 625.

[0059] Figure 7 A flowchart of an embodiment of a method for identifying a person by using a learned model for person re-identification according to an embodiment is shown. Figure 2 The method 700 is used in the context of the device proposed in the invention for identifying a person by a learned model for person re-identification. Before the method 700 for identifying a person by a learned model for person re-identification can be implemented, it is necessary to perform Figure 6 Steps in a method for learning a model for person re-identification.

[0060] The method 700 first has a step 705, in which at least one second image signal is detected, wherein the second image signal represents a second image segment in the image of the camera or the second camera. In this case, the control parameters determined in the above-mentioned method for learning a model for person re-identification are output to the camera or the second camera, so that in step 705 of the method 700, the determined control parameters can also be used for the detected second image segment. Finally, the method 700 has a step 710, in which the detected person is recognized on the read second image signal by the camera or the second camera of the camera network using the model for person re-identification. In this case, the specific person features of the person detected on the collected image data in the above-mentioned method for learning a model for person re-identification are output to the camera or the second camera, so that in step 710 of the method 700, the specific person features of the detected person can also be recognized on the read second image signal.

[0061] According to one exemplary embodiment, step 710 may also be performed on an externally arranged computing unit.

[0062] If an exemplary embodiment includes an “and / or” connection between a first feature and a second feature, this may be interpreted such that the exemplary embodiment has both the first feature and the second feature according to one exemplary embodiment and either only the first feature or only the second feature according to another exemplary embodiment.

Claims

1. A method (600) for learning a model for person re-identification using images from a camera (100), wherein: The method (600) comprises the following steps: reading (605) a detection signal (135), the detection signal representing a person (325) detected in a monitoring area (515) of at least the camera (100) of the camera network (505); collecting (610) a plurality of image signals (140) from the camera (100) using the read detection signal (135), wherein the collected image signals (140) represent captured image segments (305, 310, 315, 320) in each of the images (340) of the camera (100); and adapting (625) a model for person re-identification using the collected image signals (140) so as to re-identify the person in the monitoring area (515) of the camera (100) or at least one second camera ( 200), determining (615) adjustment parameters (145) for actively adjusting the recording characteristics of the camera (100), wherein the adjustment parameters (145) are randomly changed within a predetermined adjustment range, wherein, in the step of collecting (610), at least one further image signal (140) is recorded using the randomly changed adjustment parameters (145), so that the amount of data for training a model for re-identifying the person is artificially increased, wherein the detected person (325) is tracked by the camera (100), wherein, in the step of collecting (610), image segments (305, 310, 315, 320) of the image signal (140) are collected along a tracked movement trajectory (330) of the detected person (325).

2. The method (600) according to claim 1, comprising the step of extracting (620) specific person features (150) of the detected person (325) from the collected image signal (140), wherein: In the step of adapting (625), the specific person feature (150) is output to the camera (100) or the second camera (200) of the camera network (505).

3. The method (600) according to claim 1 or 2, comprising the step of determining (615) a control parameter (145) for adjusting the recording properties of the second camera (200). 4 . The method ( 600 ) according to claim 3 , in which, in the step of determining ( 615 ), brightness and / or contrast and / or color representation are identified as control parameters ( 145 ).

5. The method (600) of claim 2, wherein: An identifier for identifying the person (325) is assigned to the specific person characteristic (150).

6. The method (600) according to claim 3, wherein: In the step of determining (615), control parameters (145) for adjusting the recording properties of the second camera (200) are determined (615) using the collected image segments (305, 310, 315, 320).

7. The method (600) of claim 1, wherein: In the step of determining (615), control parameters (145) for adjusting the recording properties of the camera (100) are determined (615) using the collected image segments (305, 310, 315, 320).

8. A method (700) for identifying a person by means of a camera of a camera network (505) or a second camera (200) using a learned model for person re-identification, wherein: The method (700) comprises the following steps: corresponding to the steps of the method (600) according to any one of claims 1 to 7; detecting (705) at least one second image signal (220), wherein the second image signal (220) represents a second image segment in an image (540) of a camera (100) or a second camera (200); and identifying (710) a detected person (325) on the read second image signal (220) by the camera (100) or the second camera (200) of the camera network (505) using a model for person re-identification.

9. The method (700) according to claim 8, in which a specific person feature (150) of a person (325) detected in the collected image signal (140) is output to the camera or the second camera (200), wherein: In the step of identifying (710) the detected person (325), the specific person feature (150) is identified on the read second image signal (220).

10. The method (700) according to claim 8, in which the determined control parameters are output to the camera or the second camera (200), wherein: In the step of detecting ( 705 ), the determined control parameters ( 145 ) are used for the detected second image segment.

11. The method (600) according to claim 1 or 2 or the method (700) according to any one of claims 8 to 10, wherein: The cameras of the camera network (505) and the second camera (200) are connected to a computing unit (510).

12. The method (700) according to claim 11, wherein: The step of collecting (610) and / or the step of identifying (710) are performed on the computing unit (510).

13. A device (105; 205) for learning a model for person re-identification and / or identifying a person using a learned model for person re-identification, the device being configured to implement and / or control the steps of the method (600; 700) according to any one of claims 1 to 12 in corresponding units. 14 . A computer program product, which is configured to carry out and / or control the method ( 600 ; 700 ) according to claim 1 .

15. A machine-readable storage medium having stored thereon a computer program product according to claim 14.

16. The storage medium according to claim 15, wherein: The storage medium is a non-volatile machine-readable storage medium.

Images