WO2011039184A2

WO2011039184A2 - Method and device for managing object view data in an object database

Info

Publication number: WO2011039184A2
Application number: PCT/EP2010/064356
Authority: WO
Inventors: Andreas Ernst; Tobias Ruf; Christian KÜBLBECK
Original assignee: Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V.
Priority date: 2009-10-02
Filing date: 2010-09-28
Publication date: 2011-04-07
Also published as: WO2011039184A3; EP2483838A2; DE102009048118A1; DE102009048118B4

Abstract

A method for managing object view data in an object database comprises detecting a view of an object in an image and comparing view data of the detected view to object view data of object views of objects in an object database on the basis of a comparison criterion. The method further comprises storing the view data of the detected view together with time information in the object database as new object view data when the comparison shows that the detected view does not correspond to any object view in the object database, or changing time information of existing object view data of an object view in the object database when the comparison shows that the detected view corresponds to the object view. The method further comprises managing the object view data in the object database on the basis of the time information of the object view data.

Description

Method and apparatus for managing object view data in an object database

description

Embodiments of the invention relate to image recognition systems, and more particularly to a method and apparatus for managing object view data in an object database. For automatic detection of objects (especially faces) in real-time image sequences, various approaches are already known (eg, "Paul Viola and Michael Jones: Rapid Object Detection using a Boosted Cascade of Simple Features.") In Proceedings IEEE Conf. On Computer Vision and Pattern Recognition , 2001 "," Christian Küblbeck and Andreas Ernst Face Vision Computing, 24 (6): 564-572, June 2006 ").

Likewise, the automatic recognition of faces (in some cases also objects) based on single images or image sequences is already used in practice. A distinction is made between methods based on 2D images and methods based on 3D techniques. More detailed information is available e.g. in "Harry Wechsler. Reliable Face Recognition Methods: System Design, Implementation and Evaluation. (International Series on Biometrics). 2006, Springer-Verlag New York, Inc., "Anil K. Jain and Stan Z. Li's Handbook of Face Recognition, Springer-Verlag New York, Inc. 2005. Sowecu, NJ, USA" and "W. Zhao, R. Chellappa, A. Rosenfeld, P.J. Phillips, Face Recognition: A Literature Survey, ACM Computing Surveys, 2003, p. 399-458 "to find.

In 2D procedures, one or more images of each face are stored in a system. If a face is to be recognized (identified), it will be compared with all faces stored in the system. In order to increase performance, the information required for comparison is often extracted once when the faces are deposited and also stored. There are different methods for extracting these feature vectors and for comparing the faces using these feature vectors. In contrast to 2D methods, 3D models are used to create 3D models of faces. Here, for example, strip light projection or laser scanner are used. Similar to 2D methods, the 3D model is already used for recognition in the System stored data in different ways compared and possibly identified.

Both approaches have advantages and disadvantages. For example, 2D techniques can be used with commercially available cameras, whereas 3D techniques require more sophisticated equipment to capture a 3D model. In contrast, 2D methods are generally vulnerable if the view of the face to be identified is not nearly frontal or is not stored in the system in that view. The object of the present invention is to provide an improved concept for managing object view data in an object database, which makes it possible to recognize objects in pictures more quickly.

This object is achieved by a method according to claim 1 and an apparatus according to claim 11.

An embodiment according to the invention provides a method for managing object view data in an object database. The method includes capturing a view of an object in an image and comparing view data of the acquired view with object view data of object views of objects in an object database based on a comparison criterion. Furthermore, the method includes storing the acquired view's view data together with time information in the object database as new object view data when the comparison reveals that the acquired view does not correspond to an object view in the object database or changing time information from existing object view data of an object view the object database if the comparison shows that the captured view corresponds to the object view. In addition, the method includes managing the object view data in the object database based on the time information of the object view data. Embodiments according to the invention are based on the core idea that object view data is provided with time information. This time information is updated when a stored object view is recognized by comparing in an image. As a result, on the one hand, it is possible to sort the object view data in the object database based on the time information and thus to implement, for example, by first comparing object view data of object views with the view data of the acquired view that have only recently been recognized in an image. As a result, the average time required to compare the view data until a first match is found based on the match criterion can be significantly reduced, as in successive pictures in the picture sequences, the same objects with the same or similar views usually appear.

On the other hand, object view data may also be removed from the object database based on the time information of the object view data when a time criterion is met. For example, object view data may be removed that has not been recognized in an image for more than a maximum amount of time. As a result, the amount of data in the object database can be adapted to the respective application and to the respective requirements. Is e.g. As data processing speed is of crucial importance, the amount of data in the object database can be kept low by removing object view data after a short time. This makes it necessary to compare the captured view with fewer object views in the object database. If, on the other hand, it is more important to recognize objects or views of objects even after a long time, then a correspondingly longer period of time can be selected for the removal of object view data. As a result, the inventive concept can be adapted flexibly to the respective application. Some embodiments according to the invention relate to an apparatus for managing object view data in an object database. The apparatus includes an object detector, a view data comparison device and an object database memory. The object detector is designed to capture a view of an object in an image. The view data comparison device is configured to compare view data of the acquired view with object view data of object views of objects in an object database based on a comparison criterion. Furthermore, the object database memory is configured to store the acquired view view data together with time information in the object database as new object view data when the comparison indicates that the acquired view does not correspond to an object view in the object database and time information from existing object view data To change the object view in the object database if the comparison shows that the captured view corresponds to the object view. In addition, the object database memory is configured to manage the object view data in the object database based on the time information of the object view data.

Embodiments according to the invention are explained below with reference to the accompanying figures. Show it: 1 is a flowchart of a method for managing object view data in an object database;

FIG. 2 is a flow chart of a method for managing object view data in an object database; FIG. and

3 is a schematic representation of an apparatus for managing object view data in an object database. Hereinafter, the same reference numerals are used in part for objects and functional units having the same or similar functional properties. Furthermore, optional features of the various embodiments may be combined with each other or interchangeable. 1 shows a flow diagram of a method 100 for managing object view data in an object database according to an embodiment according to the invention. Here, a view of an object in an image is captured 110 and the view data of the captured view is compared 120 with object view data of object views of objects in an object database based on a comparison criterion 120. If the comparison shows that the acquired view does not correspond to an object view in the object database, so the view data of the acquired view is stored together with time information in the object database as new object view data 130. If the comparison reveals that the acquired view corresponds to an object view, time information is changed from existing object view data of the corresponding object view in the object database 140. Further, the Object view data in the object database is maintained 150 based on the time information of the object view data.

By storing view data with time information and by changing time information from existing object view data, it is possible to record when an object view or an object was last detected. Thus, for example, the processing speed of images can be significantly increased because the object view data in the object database can be sorted based on the time information of the object view data, for example, and object view data of recently viewed object views can be compared first. The processing speed can be increased since successive images of a picture sequence usually contain similar or identical views of objects. Furthermore, object view data may be removed from the object database if these views have not been captured in any image for some time. As a result, the amount of data in the object database can be kept manageable and thereby also the processing speed, which can be achieved with the inventive concept can be significantly increased.

An image may include one or more objects, where an object may be, for example, a face, a human, a vehicle, or other movable or immovable object. The image may be part of an image sequence or image sequence, where an object may be included in multiple images of the image sequence. The views of an object in different images may differ, such as with moving objects, e.g. People or vehicles that may be the case.

The view data may, for example, directly comprise the image section of the view of the object or be derived therefrom or extracted parameters, also called feature vectors. This view data in the form of e.g. Feature vectors of the acquired view can be combined with object view data, object views of objects in the object database, e.g. can also be present in the form of feature vectors.

The compare criterion determines how similar the view data and object view data must be to identify or classify a captured view as an already saved object view. If more than one match occurs in the view data of the acquired view with the object view data in the object database based on the comparison criterion, then that object view data belongs to an object and can be combined and assigned to a single object. If the detected view of the object is recognized, the time information of the already stored object view is updated. Otherwise, the view data of the acquired view is stored together with time information as new object view data. The time information may be, for example, the time at which the image was generated, the time at which the view of the object was detected, or the time at which the new object view data was stored or the time information of the existing object view data was changed. Alternatively, for example, a dwell permanently set to zero or a maximum dwell time. Upon expiration of the maximum dwell time, or when the dwell time has reached a maximum predefined value, the object view data of the object view of an object or all of the stored object views of an object may be removed from the object database.

In other words, the management may include removing object view data of an object view from the object database if the associated time information satisfies a time criterion. The time criterion, as already mentioned, e.g. the expiration of a maximum residence time or exceeding a maximum residence time. By removing object views of objects that have not appeared in any image for a long time, the amount of data in the object database can be kept small. As a result, the time required to compare the view data from captured views with stored object views can be significantly reduced as compared to methods that do not dynamically update the object database.

Alternatively or additionally, the managing may include sorting the object view data of the object views in the object database based on the associated time information of the object data to increase the processing speed of the described method. Sorting allows object view data from object views that were recently captured in an image to be compared to the newly acquired view first. The processing speed can be increased because objects in a current image were also likely to be in the previous image. If the image is part of an image sequence, it can be determined in a further step, for example, whether the detected object in the current image corresponds to a known object from a previous image of the image sequence. In other words, it can be determined, for example, by comparing the size and / or positions of an object in an image with the size and / or positions of an object in a previous image, whether this is the same object. If it is the same object, at least one object view for the object must already be stored in the object database. Therefore, it may be sufficient to compare the captured view only with object view data already stored for that object. If none of the object views already stored for this object corresponds to the newly acquired view, the newly acquired view with a time information can be added to the object views of the already stored object. In addition, the newly acquired view can also be compared with all other stored object view data. to determine if the stored object was previously captured, and stored with an object view that was not recognized as corresponding in a previous comparison because the object views are too different. Thus, the newly acquired view can establish the relationship between the object view data stored for different but actually the same objects. As a result, all of these object view data can be aggregated and associated with a single object.

Each object view of an object may have individual time information. Alternatively, each object in the object database can also be assigned only time information that then applies to all object views of the object. If the time information of object view data of an object view of an object is changed, e.g. the individual time information of the object view, the individual time information of all object views of the object or the time information of the object assigned to the object view have been changed.

In some embodiments according to the invention, the object view data of each object view in the object database is associated with an object of the objects stored in the object database. The object database can be designed so that at least one object view is assigned to each object. Objects due to e.g. A merging of object views of objects identified as identically identifies no object view data can be removed from the object database. The object database may also be referred to as a gallery or dynamic gallery, and an object may also be referred to as an entry in the database or simply as an entry. Furthermore, a time information is also called timestamp in the following.

By evaluating the object database, it can be determined, for example, how often an object appears in a picture sequence, how long an object is present in a picture sequence or how many different objects occur in a picture sequence.

2 shows a flow chart of a method 200 for managing object view data in an object database according to an embodiment according to the invention.

The process is divided into several sub-processes, which can be processed cyclically and depending on the current state of the process. The following are first generally the sub-procedures described in more detail. Thereafter, the nodes of the figure will be explained again. Letters in parentheses refer to the corresponding nodes of the figure. To start the procedure, for example, the first image of the sequence is necessary, see (A). The automatic detection (detection) of objects is carried out. If no objects were detected (B), the cleanup of gallery (D) described above may follow. This process will be explained later in more detail. This is followed by the automatic detection of objects (C, B) for the next image in the sequence.

If objects have been detected (B), the first object is selected (E) and checked whether an assignment to an object in the previous image is possible (G). This process will also be explained later in more detail. If an assignment was not possible, then it is an object that has been newly detected in this picture, and the sequence goes to node (L). Here the current view of the object is compared with all gallery entries. The comparison with each individual entry (object in the object database) can be made by comparing the current view in turn with the views of the entry. As soon as the first comparison with a view was successful, there is a match with the gallery entry (object in the object database). Likewise, other gallery entries may match the current (captured) view of the object.

If no match was found with at least one gallery entry, a new entry for the object in the gallery is created (saved) and the current view of the object with the associated time stamp is inserted in this entry (J). If there are no more objects after that (I, D), the gallery is cleared and the process continues at node (D), otherwise the process continues for the next object (I, H) at node (G).

If in (L) a match was found with at least one gallery entry (the object could be recognized), then the matching gallery entries are merged into an entry and assigned to this object (M). The timestamps of the views that resulted in a successful comparison can then be updated (N). After sorting the list of views of all gallery entries (K), the process continues again at node (I). If in (G) the selected object could be assigned to an object from the previous image, an entry for this object already exists in the gallery. Now, the current view of the object can be compared (F) in turn with all views of the associated gallery entry. As soon as the first match has been found, it updates the timestamp of the matching view (N) and the process moves on to node (K).

If there is no match at node (F), it is obviously a new view of the object. To check if there are more entries in the gallery that correspond to this object, but were not recognizable with the previous views, a comparison of the current view with all other entries of the gallery (O) can follow. If at least one match has been found, all matching entries can be merged (P) with the current entry of the object and the process continues at node (N). If no match was found in another entry, the previously unknown current view of the object is inserted in the corresponding entry of the gallery (Q) and the process continues at node (K).

The individual nodes of FIG. 2 are again considered separately in more detail below:

Node (A): e.g. Starting point of the procedure with the first image of the sequence.

Node (B): Detection of the objects in the image, e.g. with already known methods. Depending on whether or not objects are found, node (E) or (D) follows.

Node (C): Select or wait for the next image of the sequence.

Node (D): clean up the gallery. For example, all views (object views) are first removed from all entries in the gallery whose timestamp has expired. This may be the case if the timestamp of a view with respect to the current time exceeds a fixed maximum time span. Thereafter, all gallery entries that no longer contain a view can be removed. Furthermore, the maximum number of views of each entry can be limited to N. If an entry exceeds the maximum number of views, all but N views are deleted with the most recent time stamp. In addition, it is optionally possible to display views that were newly added in the previous image but did not match in the current image to delete. As a result, only views in the gallery are kept, which led at least once to a match.

Node (E): If at least one object was found in the image, the first object is first selected. All other objects are then selected in (H).

Node (F): If an entry belonging to the object already exists in the gallery (object was already detected in the previous image), then the current view of the object is compared in sequence with all views already stored in the entry. Once the first match has been found, the comparison with other views can be aborted (for performance reasons). Since the list of views is sorted by the time stamp, cf. (K), in many cases within the image sequence, one single comparison per object is sufficient with the first view in the gallery (in successive images the views change only slightly). If no match was found when comparing with the stored views, then obviously the view of the object has changed.

The comparison of the views may be based on simple difference value generation, but may also use more complex image comparison methods (e.g., use of illumination independent features to compare and / or use histograms for image comparison). Furthermore, it is possible here to use a method known from the prior art, in particular to methods used for faces for identification.

Node (G): This node checks whether the object has already been detected in the previous image, i. E. an assignment to an object in the previous picture was possible. The assignment is possible if e.g. Size and position has at least one predetermined similarity value (the overlap area of the detections relative to the sum of the area of both detections exceeds a specified threshold value). In addition to this geometric assignment, a comparison of the image sections of the detections can be used as an assignment criterion (the assumption is that the appearance of an object differs only slightly in successive images of a sequence). For comparison, the same methods described in (F) may be used.

Node (H): Selects the next object from the objects detected in the current image.

Node (I): Checks if another detected object is present and leads to node (D) or (H) depending on it. ICnoten (J): For the newly detected object, no match was found with an entry in the gallery, either because the object is not yet available in the gallery or because the current view is not yet in any entries belonging to the object the gallery is saved as a view. For this reason, a new entry (object) is created in the gallery and the view with time stamp is inserted in the entry.

Node (K): This node sorts the list of views of all gallery entries by the timestamp. Thereby, e.g. Views that led to finding matches are moved to the top of the list and contribute to faster execution in (F).

Node (L): If an object has been redetected according to (G), the current view of the object is compared with all entries in the gallery to recognize the object if possible, if it already exists in the gallery with a similar view. The comparison may correspond to the method described in (F). In addition, however, an acceleration can still take place, e.g. through hierarchical search and / or first rough comparison (fast but less accurate method) and then more complex fine comparison (slower but more accurate method). Even if the object has already been found in earlier images and is present in the gallery, it may happen that an assignment is not immediately possible, e.g. because the current view is too different from the already saved views. This may (temporarily) contain multiple entries of the same object in the gallery, which in later pictures of the sequence are reassembled by (M) or (P), if then a comparison matches.

Node (M): If (L) results in a match of the current view with more than one gallery entry, these entries can be combined and assigned to the object. If it matches a single gallery entry, it will be assigned to the object.

Node (N): All views of gallery entries that matched (L), (F), or (O) can be timestamped. This shifts them to the top of the view lists in (K), which speeds up the search in subsequent images at (F) (the view of an object generally changes only slightly in successive images). Furthermore, the current timestamps mean that these views in (D) remain unaffected. Node (O): If (F) did not match the already saved views, then normally the view of the object has changed (could also be caused by occlusion or lighting changes). The new view of the object can be compared with all the gallery entries to find entries that also represent this object, but were unrecognizable with the previous views. The comparison can be made in the same way as described in (L).

Node (P): If at least one match was found in (O), all matching entries of the gallery can be merged.

Node (Q): If the new view (O) did not match the other entries in the gallery, the new view can be added to the associated gallery entry with a time stamp.

In order to clarify found matches in the procedure, unique IDs (key figures, keys, identification) can be used. For this purpose, e.g. each object in the image assigned a unique Id as long as the object is detected without interruption in the image sequence, see. Fig. 2 transition (G) to (F). If an object disappears and is detected again after some time, a new id is assigned to the object, cf. Fig. 2 transition (G) to (L).

Matches can be represented by enumeration or chaining of ids. Either by specifying for each object all Ids already assigned to the object (enumeration) or by specifying only the Ids that the object had at previous detections (concatenation can have more than one Id, see Fig. 2 Transition (O) to (P) and transition (L) to (M)).

Some embodiments according to the invention relate to a method for automatically recognizing objects and image sequences from different views.

The method of automatically recognizing objects in image sequences from different views may be based on the use of a dynamic (time-varying gallery) where each entry of the gallery may hold a list of different views of the object and each view may be time-stamped , Another aspect of the described concept is to update the gallery by removing views that no longer contributed to a set period of time for recognition and adding new views as soon as an object no longer matches the previous views.

Another aspect is adding new entries to the gallery and removing entries that no longer contain views.

Furthermore, an initiation of the comparison of an object with all entries of the gallery can be made by the appearance of a new view (this can significantly reduce the number of comparisons required or significantly increase the speed).

Also, one aspect of the described concept is the use of a timestamp for each view, which is updated whenever a view provides a match.

In addition, the timestamp can be used for sorting the views and / or evaluating the value of the views or as an aging criterion (reduces the number of comparisons required since current views are always at the top of the list). Another aspect is the use of a save duration for each view instead of a timestamp set to a maximum duration if the view was found in the image, otherwise it is degraded. When the storage period expires, the view is removed. Just as an aspect of the described concept is a merging of multiple entries of the gallery to find matches.

Another aspect is the assignment of detections in successive images within an image sequence to an object based on position and size and / or comparison of the image sections.

Also, one aspect is the use of Ids to uniquely identify objects as long as they have been detected in the image sequence without interruption. In addition, use of Id enumerations or concatenations to represent matches may be made. The inventive concept describes z .B. a method for fast, automatic (temporary) recognition of objects from different views in image sequences. Thus, for example, the purpose of the invention is to recognize an object (eg, a face) once detected within an image sequence (ie, associate it with the earlier detections) if, after it has disappeared, it repeats itself several times in the image sequence is detected. For this purpose, the method uses a dynamic (time-varying) gallery. This consists eg of a list of entries. An object is represented by one (or more) of these entries. Each entry in turn has a list of one or more views of the object (object views). A view consists of the image section of the object (and / or a feature vector pre-extracted to increase its performance) and a timestamp indicating when this view of the object was last seen in the video sequence (image sequence).

Some embodiments according to the invention relate to the recognition of objects, in particular faces, by skillfully storing views of the objects in a gallery (object database).

The inventive concept finds application, for example, in systems and methods for the detection and recognition of objects (e.g., faces, objects) or in systems for advertising effectiveness analysis (counting viewers of advertising). In this case, the method according to the invention makes it possible, for example, to recognize when persons are viewing the advertisement or the screen several times within a fixed period of time.

FIG. 3 shows a schematic representation of an apparatus for managing object view data in an object database according to an embodiment according to the invention. The apparatus 300 includes an object detector 310, a view data comparator 320, and an object database memory 330. The object detector 310 is connected to the view data comparison device 320, and the view data comparison device 320 is connected to the object database memory 330. The object detector 310 acquires a view of an object in an image 302 and provides view data of the acquired view of the view data comparison device 320 for comparing the view data of the acquired view with object view data of object views of objects in an object database based on a comparison criterion. supply. The object database memory 330 stores the acquired view view data together with time information in the object database as new object view data when the comparison shows that the acquired view does not correspond to an object view in the object database. Otherwise, the object database store 330 alters time information of existing object view data of an object view in the object database when the comparison reveals that the acquired view corresponds to the object view. Further, the object database memory 330 manages the object view data in the object database based on the time information of the object view data. The object detector 310, the view data comparison device 320 and the object database memory 330 may be stand-alone hardware units or part of a processor or computer, or computer programs or computer program products for execution on a computer or microprocessor. Some embodiments according to the invention relate to an image recognition system having an apparatus for managing object view data in an object database and a camera. The camera is designed to generate an image sequence, the image with the acquired view being part of the image sequence. Alternatively, the image sequence can also be recorded by more than one camera.

Although some aspects have been described in the context of a device, it will be understood that these aspects also constitute a description of the corresponding method, so that a block or a component of a device is also to be understood as a corresponding method step or as a feature of a method step. Similarly, aspects described in connection with or as a method step also represent a description of a corresponding block or detail or feature of a corresponding device.

Depending on particular implementation requirements, embodiments of the invention may be implemented in hardware or in software. The implementation may be performed using a digital storage medium, such as a floppy disk, a DVD, a Blu-ray Disc, a CD, a ROM, a PROM, an EPROM, an EEPROM or FLASH memory, a hard disk, or other magnetic disk or optical memory are stored on the electronically readable control signals that can cooperate with a programmable computer system or cooperate such that the respective method is performed. Therefore, the digital storage medium can be computer readable. Some embodiments according to the invention thus comprise a data carrier which stores electronically readable control Having signals capable of interacting with a programmable computer system such that one of the methods described herein is performed.

In general, embodiments of the present invention may be implemented as a computer program product having a program code, wherein the program code is operable to perform one of the methods when the computer program product runs on a computer. The program code can also be stored, for example, on a machine-readable carrier. Other embodiments include the computer program for performing any of the methods described herein, wherein the computer program is stored on a machine-readable medium.

In other words, an embodiment of the method according to the invention is thus a computer program which has a program code for performing one of the methods described herein when the computer program runs on a computer. A further embodiment of the inventive method is thus a data carrier (or a digital storage medium or a computer-readable medium) on which the computer program is recorded for carrying out one of the methods described herein.

A further embodiment of the method according to the invention is thus a data stream or a sequence of signals, which represent the computer program for performing one of the methods described herein. The data stream or the sequence of signals may be configured, for example, to be transferred via a data communication connection, for example via the Internet.

Another embodiment includes a processing device, such as a computer or a programmable logic device, that is configured or adapted to perform one of the methods described herein.

Another embodiment includes a computer on which the computer program is installed to perform one of the methods described herein.

In some embodiments, a programmable logic device (eg, a field programmable gate array, an FPGA) may be used to perform some or all of the functionality of the methods described herein. at In some embodiments, a field programmable gate array may cooperate with a microprocessor to perform any of the methods described herein. In general, in some embodiments, the methods are performed by any hardware device. This may be a universal hardware such as a computer processor (CPU) or hardware specific to the process, such as an ASIC.

The embodiments described above are merely illustrative of the principles of the present invention. It will be understood that modifications and variations of the arrangements and details described herein will be apparent to others of ordinary skill in the art. Therefore, it is intended that the invention be limited only by the scope of the appended claims and not by the specific details presented in the description and explanation of the embodiments herein.

Claims

claims

A method (100; 200) for managing object view data in an object database, comprising the steps of:

Detecting (110, B) a view of an object in an image;

Comparing (120, F, L, O) acquired view view data with object view data of object views of objects in an object database based on a comparison criterion;

Storing (130, J, Q) the acquired view view data together with time information in the object database as new object view data when the comparison reveals that the acquired view does not correspond to an object view in the object database, or changing (140, N) time information from existing object view data of an object view in the object database if the comparison indicates that the acquired view corresponds to the object view; and

Managing (150, D, K) the object view data in the object database based on the time information of the object view data.

The method of claim 1, wherein the managing (50) comprises removing (D) object view data of an object view from the object database if the associated time information satisfies a time criterion.

The method of claim 1 or 2, wherein the managing (150) comprises sorting (K) the object views in the object database based on the associated time information of the object views.

4. The method of claim 1, wherein each object view in the object database is associated with an object of the objects stored in the object database, wherein each object is associated with at least one object view.

5. The method according to any one of claims 1 to 4, with the following further step: Merging (M, P) identical objects and their object views in the object database into a single object, whereby identical objects are recognized, if the comparison reveals that the acquired view corresponds to more than one object view and the corresponding object views are assigned to different stored objects, wherein all object views of the objects recognized as the same in the joining (M, P) are assigned to the single object.

Method according to one of claims 1 to 5, wherein the image is part of an image sequence, with the following further step:

Determining (G) whether the detected object in the image (102) corresponds to a known object from a previous image of the image sequence.

A method according to claim 6, wherein the view data of the acquired view is compared with object view data of a known object (F) if the detected object corresponds to the known object from a previous image.

The method of claim 7, wherein the view data of the acquired view is compared with the object view data of the remaining objects in the object database (O) if the comparison reveals that the acquired view does not correspond to any of the object views of the known object.

A method according to claim 7 or 8, wherein the new object view data is associated with the known object (Q) upon storage when the comparison reveals that the acquired view does not correspond to an object view in the object database.

Method according to claim 7 or 8, wherein object view data of a further object of the object database are assigned to the known object (P) if the comparison reveals that the acquired view corresponds to an object view of the further object.

Apparatus (300) for managing object view data in an object database, comprising: an object finder (310) adapted to capture a view of an object in an image (302); a view data comparison device (320) configured to compare view data of the acquired view with object view data of object views of objects in an object database based on a comparison criterion; and an object database store (330) arranged to store the acquired view view data together with time information in the object database as new object view data if the comparison reveals that the acquired view does not correspond to an object view in the object database, and to change time information of existing object view data of an object view in the object database when the comparison shows that the acquired view corresponds to the object view, and wherein the object database memory (330) is adapted to receive the object view data in the object database based on the time information of the object view data manage.

12. An image recognition system comprising: an apparatus for managing object view data in an object database according to claim 11; and a camera configured to generate an image sequence, the image sequence comprising a plurality of images, and wherein the captured view image is part of the image sequence.

13. Computer program with a program code for carrying out the method according to one of claims 1 to 10, when the computer program runs on a computer or microcontroller.