EP2283937A1 - Method and device for transporting objects to destinations depending on pattern images - Google Patents

Abstract

The method involves passing a sample of a classification medium with a transporting object (Ps-a,Ps-b), where the classification medium has an image recording device (1) and an assembly unit (3). The image recording device produces an image (Abb-a,Abb-b) from each object of the sample. A sorting system is equipped for triggering the transport of one object to a target point. An independent claim is also included for a device for transporting goods to target points.

Description

The invention relates to a method and a device for transporting objects, in particular postal items, to target points which depend on image patterns.

Often, a large number of similar items, such. B. many copies of a magazine issue, sent by post to the respective recipient. As a rule, some of these mailpieces can not be delivered to the recipient because the recipient address on the mailpiece is no longer valid. Such mail should often be returned to the sender. The problem arises that the sender address on the mail item is not readable from the outside or at least more difficult to read than the recipient address.

In DE 19836767 C1 For example, a method and apparatus for processing mailpieces to be returned to the respective sender will be described. For example, a mail carrier determines that a catalog transported as a mail item can not be delivered to the recipient specified on the mailpiece and provides the mailpiece with a return label. This mail again goes through a sorting process. An image of each mail item to be returned is generated. These images contain areas of images of sender addresses. Similarities of those areas in the images showing sender addresses are determined. Those images that are sufficiently similar are treated as images of postal items, all of which are to be returned to the same sender. In a video coding station, an editor reads the sender information in one of these images and enters it into a data acquisition device. All mail with similar sender images, a recognized sender is sent back to them.

In DE 102006059525 B3 A method is described for checking the transportation of mail. It is determined whether a mail item that passes through a sorting system twice is recognized correctly on the second pass without a clear identification being printed on the mailpiece.

In DE 102007038186 B4 describes a method in which a sender sends a lot of many similar mailings an image of a surface of these similar mailings to a transporter. These mail items are sorted together with other mail items in a sorting system. Using the image, it is decided whether or not a mail item to be sorted belongs to this quantity.

In EP 1622065 B1 and AU 2005203141 A1 a method is proposed in which mail items are to be sent to the same recipient.

In US 20080008376 A1 For example, a method is described for identifying a postal indicia on the envelope of a mailpiece. A "region of interest" is identified and at least one candidate is found in this "region of interest". Each candidate calculates a set of feature values for which an image of the envelope is generated and evaluated. Each candidate is classified using these characteristic values.

In US 20070239848 A1 For example, a method is described for scaling an image of a mail piece ("to normalize"). For this purpose, two images of the mailing are generated. In addition to the mailing, a logo is positioned each time. The image of the logo is used to scale the image.

The invention is based on the object, a method and an apparatus for transporting objects to target points in which image patterns on objects are evaluated during transport, avoiding the need to manually enter computer-readable descriptions of image patterns.

The object is achieved by a method having the features of claim 1 and a device having the features of claim 9. Advantageous embodiments are specified in the subclaims.

• eine Klassifizierungs-Vorrichtung,
• eine Sortieranlage und
• einen Bildmuster-Datenspeicher.

The solution according to the device comprises

• a classification device,
• a sorting plant and
• a pattern data store.

The classifying apparatus includes an image pickup device and a grouper. The sorting system also has an image pickup device and a picture pattern recognizer.

According to the solution, first a training phase and then a sorting phase are carried out. The sorting system is used at least in the sorting phase.

In the training phase, a predetermined sample of items to be transported is used. The following steps are performed according to the solution in the training phase:

The items to be transported in the sample pass through the classification device.

The image capture device generates at least one image of each subject of the sample.

The grouper automatically determines which different image patterns are displayed on at least one subject of the sample. In doing so, the grouper applies a clustering method to the images of the sample.

The thus determined image patterns are stored in the image pattern data memory.

• Der zu transportierende Gegenstand durchläuft die Sortieranlage.
• Das Bildaufnahmegerät erzeugt von dem zu transportierenden Gegenstand mindestens ein Abbild.
• Der Bildmuster-Erkenner stellt automatisch fest, ob auf dem Gegenstand ein Bildmuster dargestellt wird, das in dem Bildmuster-Datenspeicher abgespeichert ist, und wenn ja welches. Hierfür wertet der Bildmuster-Erkenner das Abbild des Gegenstands aus.
• die Sortieranlage ermittelt dann, wenn auf dem Gegenstand ein abgespeichertes Bildmuster dargestellt wird, unter Verwendung des Bildmusters einen Zielpunkt, an den dieser Gegenstand zu transportieren ist.
• die Sortieranlage löst einen Transport des Gegenstands zu diesem Zielpunkt aus.

In the sorting phase, the following steps are carried out by solution for each item to be transported:

• The object to be transported passes through the sorting system.
• The image capture device generates at least one image of the object to be transported.
• The image recognizer automatically detects whether an image pattern stored in the image pattern data memory is displayed on the object and, if so, which one. For this purpose, the image pattern recognizer evaluates the image of the object.
• the sorting system determines when the object a stored image pattern is displayed, using the image pattern, a target point to which this object is to be transported.
• the sorting system triggers a transport of the object to this destination point.

The invention eliminates the need for a processor to make computer-available descriptions of image patterns available in a training phase so that the sorting system can use those image patterns for the sorting phase. This is particularly advantageous if new patterns are added to objects on an ongoing basis, so that a training phase with manual input would always be too late.

The invention further eliminates the need for a mailer of items to communicate a computer-accessible description of a picture pattern to the sorting system. It is enough that the sender - or a third party - provides items with the image pattern.

Preferably, the sorting system is also used for the training phase. As a result, components of the sorting system, z. As the imaging device, use in both phases.

Preferably, the clustering method is performed to form groups of images, each group comprising at least two images that show the same image pattern. It is also possible to set a minimum number of images per cluster that is significantly larger than 2.

Preferably, several image features are specified. By evaluating an image, it is determined which value each feature assumes for this image. This creates a feature value vector for the image. Characteristic feature value vectors for the image patterns are stored in the image data memory. In the sorting phase, a feature value vector is generated for each object to be transported and compared with the stored feature value vectors. This makes it particularly easy to decide whether and, if so, which image pattern is displayed on the object.

The solution according to the method and the device according to the solution can be z. B. for the transport of flat mail, packages, packages, pieces of luggage or workpieces in a manufacturing use.

Fig. 1

den ersten Durchlauf von Postsendungen durch eine Sortieranlage, wobei Bildmuster per Clustering er- kannt werden;

Fig. 2

den zweiten Durchlauf von Postsendungen durch die Sortieranlage von Fig. 1, wobei die Erkennungsergeb- nisse aus dem ersten Durchlauf angewendet werden;

In the following the invention will be described with reference to an embodiment. Illustrate:

Fig. 1

the first pass of mail through a sorting system, whereby image patterns are detected by clustering;

Fig. 2

the second pass of mail through the sorting system of Fig. 1 wherein the recognition results from the first pass are applied;

In the exemplary embodiment, the method is used to recognize image patterns on mailpieces. For example, the mailpieces are copies of an issue of a journal or a catalog that will be sent by post. The mailpieces can be flat or even packages or freight items. Within a few days, a postal service provider is to send many identical copies together with other mailpieces to their respective recipients.

The picture patterns are z. As logos or emblems, with which sender provide their mail. The image pattern identifies the sender and is z. B. a protected figurative mark. Of course, the pattern may contain letters of a natural language or other characters.

The image pattern on a mailpiece should be recognized, in particular, if the mailpiece can not be delivered to the recipient whose postal address is specified on the mailpiece. For example, the receiver is unknown warped. Or the recipient has moved and has communicated his new address to the postal service provider. However, the mailpiece should not be forwarded to the new address due to a sender's "endorsement". Rather, either the mailpiece should be destroyed, and a notification about the non-delivery of the mailpiece should be sent to the sender ("sender notification"). Or the mailing provided with information about the new address should be returned to the sender. It is possible that the item of mail is provided with an "endorsement", which determines what should happen to the item of mail if the item of mail can not be sent to the address specified on the item of mail.

In the cases just described, it is necessary to determine the sender address. If an image pattern of the sender can be recognized on the mail piece, it is easier to sort the mailpieces to be returned according to the respective sender. For example, the addresses and image patterns of registered senders are stored in an image address database of the postal service provider. It is also possible to perform a procedure like this in DE 19836767 C1 is described. Recognized are those mail items that are each provided with a representation of the same postal item. The delivery address of the sender who has provided mailpieces with this image mark is determined, for which purpose an image of a mailpiece with this image mark is evaluated. The delivery address of the sender is z. B. automatically by character recognition ("Optical Character Recognition", OCR) or manually determined by video encoding.

The recognized image patterns as well as further details of the sender on the mailpiece are compared with this database. After a successful comparison, it is possible to group the mail items by sender to be able to deliver all the mailpieces to the same sender. In the simplest case, it suffices to sort the mailpieces to be returned according to sender image patterns without necessarily automatically deciphering the sender address during sorting.

In another embodiment, a sender delivers a lot of similar mailings, e.g. As the magazine copies of an issue, without delivery address and provided with its own pattern to the postal service provider. In addition, the sender sends an address list with delivery addresses to the postal service provider, preferably in computer-available form. At least one copy of the quantity of similar mailpieces is to be delivered to each delivery address in this address list. It is possible that the address list for each delivery address contains the number of copies to be delivered to this delivery address.

The postal service provider evaluates the address list and assigns each copy with an address from the address list. In this case, the similar mailpieces are to be distinguished from other mailpieces due to the same image pattern.

In a training phase, that is to say a first method step of the exemplary embodiment, a random sample with image pattern images is generated and evaluated. Fig. 1 illustrates the first process step. In a first run of mail through a sorting system, image patterns are recognized by clustering.

For this purpose, passes through a certain amount of mail Ps-a, Ps-b, ..., Ps-n a sorting system, z. B. all mail items that reach the sorting system on a particular day. A camera 1 of the sorting system generates from each mail item Ps-a, Ps-b, ... at least one image Abb-a, Abb-b, .... It is possible that several cameras produce multiple images of the same mail item from different directions to also capture sender information on the back of a flat mail piece or because the mail piece is a package and six surfaces are eligible to show sender details. It is also possible for an image Abb-a, Abb-b,... To be generated and evaluated first and only if this image does not show the recipient address or the sender address, the item of mail is turned over and another image is generated ,

Some or many of these images Abb-a, Abb-b, ... contain image patterns of different senders of the mailings Ps-a, Psb, .... Other images contain no image patterns.

The n images Abb-a, Abb-b, ... are evaluated. In one embodiment, the n images are oriented and scaled before the actual evaluation. Orientation ensures that in all images with the same image pattern, the image pattern appears in the same orientation, ie not even standing upright and once standing upside down. Preferably, the images are oriented on the basis of the respective recipient address in the image. Scaling normalizes all images to the same size, even if the images come from different sizes of mail. It is also possible, the actual evaluation without first orienting and scaling the images.

In the actual evaluation, in one embodiment, each image is first searched for the area containing information about the sender of the mailpiece. This area could contain an image of a sender's image pattern. It is of course possible that individual images do not contain any sender information.

First, an attempt is made to automatically determine this area with sender details. It is possible that it is specified that an image pattern of interest is always to be found in a certain area of the mailpiece, eg. In the upper left quadrant or the left half of the mailpiece so as to distinguish a picture pattern of a sender from a franking mark.

If it is not possible to determine a sender information in an image, in one embodiment the image is displayed on a video display device of a video coding station. An engineer marks the area containing the sender details in the displayed image, or specifies that the image does not contain any sender information. For example, the sender uses a mouse to position a rectangle around the sender specified area. Possible, but not required, is that the processor enters the sender address himself. However, he can not enter an image pattern, at least if it consists of more than one text or shows a text in a specific typography.

Predetermined m characteristics that can be calculated by evaluating a mailpiece image. For each one postal item of the random sample, a feature value vector with m feature values is generated in each case by the respective image Abb-a, Abb-b,... Of the mail item being automatically evaluated. This feature value vector indicates which m values assume the m features for this mail item.

Some of these features indicate the color distribution on the mail piece. For this purpose, the spectrum of visible colors is subdivided into finite color areas ("color channels"). For each color range, parameters of that area in the image Abb-a, Abb-b, ... are calculated, which includes pixels with hues from this color range ("color gamut area"). These parameters include, for. For example, the size of the color area area itself, the edge length of a square circumscribing the area, the diagonal of a circumscribing ellipse, the centroid, and the perimeter length of the color area area. It is possible to divide the image of the mail item into areas, eg. In four quadrants, and calculate the respective parameters of the gamut area in each quadrant.

Other features may result from the distribution and frequency of gray values, for which a grid is superimposed over the image. Such methods are for. B. known to recognize mailings without providing them with a marker ("Fingerprint", "Virtual ID").

A cluster educator 3 applies a clustering method. "Cluster analysis" or "clustering" is generally understood to mean a method in which a set of objects, which are each described by a feature value vector, are divided into groups, the objects of a group being similar to one another. "Clustering" is a method of "unsupervised learning" and is z. In the fields of machine learning, data mining, pattern recognition, image analysis, and bioinformatics Examples of clustering methods are k-means clustering, fuzzy c-means clustering, "QT clustering algorithm", "locality-sensitive hashing" and graph theory models.

For the clustering method of the embodiment, a distance measure between two feature value vectors is specified. The distances between two characteristic value vectors of the sample are calculated. For n images Abb-a, Abb-b, ... of the sample these are n * (n-1) / 2 distance calculations. All images whose distances according to this distance measure are smaller than a predetermined limit are treated as images of a group (a "cluster") showing the same image pattern.

Preferably, the groups ("clusters") are calculated so that at least two images of the sample fall into each group.

Through this application of a clustering method, the cluster-forming device 3 automatically recognizes which different image patterns are contained in the images Abb-a, Abb-b,.... The same picture pattern, z. As the same logo of a magazine publisher, can occur on different mailings of the same sender in different sizes and - due to variations in the printing process and environmental conditions when recording the images - in different configurations. Therefore, when applying the clustering method, the clustering device 3 forms groups of approximately similar image patterns.

The identified groups of similar image patterns are stored in an image database 4 or other suitable data storage. As a result, an image data base 4 with reference image patterns is automatically generated. Preferably, each group is stored by storing in the image pattern database 4 a computer-accessible description of that area in the space of the possible feature value vectors in which all the feature value vectors of this group are located. It is also possible to store a characteristic feature value vector or a characteristic image from this group. It is possible, but not necessary, to additionally extract one image pattern from the images and to store a description of this image pattern.

In a sorting phase, that is to say a second method step of the exemplary embodiment, mailpieces again pass through the sorting system. In the embodiment, the sorting system So used for both the steps of the training phase as well as the steps of the sorting phase. Fig. 2 illustrates the second pass of mail through the sorting system of FIG Fig. 1 , wherein the recognition results from the training phase, ie the first pass are applied

It is ascertained which mailpieces are to be returned to the sender and for which mailpieces a notification is to be generated to the sender. From each such mail item, as described above, an image pattern of the sender mounted on the mailpiece is detected. An image pattern recognizer 5 identifies an image pattern on a mailpiece and compares this image pattern with those reference image patterns which were automatically determined in the first method step and stored in the image pattern database 4.

In one embodiment, all mail items that are provided with different variants of the same image pattern, in the same sorting end - or the same sorting outlets, if not all mail items fit into a single sorting end - the sorting system discharged. These mail items are to be returned to the same delivery address.

In one embodiment, all mail items are spent from the same sorting end in a transport container, z. B. a postal parcel. This transport container is provided with the postal address of the sender and transported to this postal address. In this embodiment, the common sender information is not necessarily determined during sorting, but only when the transport container is labeled.

In another embodiment, the common sender address of all mail items that are to be returned to the same sender is additionally determined during the sorting, that is, in the sorting phase. For this purpose, one of these mailings selected. The sender address applied to the mailpiece is determined. First, an attempt is made to automatically determine and decipher the sender address on the selected mail piece. If this is not successful, an image of the surface of the mail item is displayed on a screen of a video coding station. An engineer marks the area in the image that shows the sender details. It is attempted to automatically read the address in this area. If this fails, the processor enters the sender details.

It is possible that the mailings of a particular sender do not contain sender information, at least not on an externally visible page. For example, an issue of a journal may show a picture pattern of the issuing publisher in the cover page, but no sender address.

In one embodiment, an image pattern address database 6 is therefore additionally specified in the second pass. This image pattern address database 6 contains several data records. Each record comprises a reference image pattern of a sender as well as a delivery address of this sender. The record may contain additional information, e.g. B. Billing information for transportation services provided to this sender.

The image pattern of the selected mailpiece is determined. This image pattern is compared with reference image patterns of data sets of the image pattern address database 6. The record with the most similar reference image pattern is automatically selected. If this automatic selection does not succeed, an editor selects the correct image pattern under automatically selected similar reference image patterns. The delivery address that is included in the record with the most similar reference image pattern is used as the delivery address for all mailpieces with this image pattern. The mailpieces are preferably transported together to this delivery address of the sender.

LIST OF REFERENCE NUMBERS

reference numeral importance 1 Camera that creates images of surfaces of mailpieces 2 discharge device 3 Clustering facility 4 Image pattern database 5 Pattern recognizer 6 Pattern-address database 7 control device Abb-a, Abb-b, ..., Abb-n Images of surfaces of the mailpieces Ps-a, Ps-b, ..., Ps-n Fig-1, Fig-2, ..., Fig-N Images of surfaces of the mailpieces Ps-1, Ps-2, ..., Ps-N m Number of characteristics for the clustering procedure N Number of samples of the sample used in the training phase N Number of mailpieces sorted in the sorting phase Ps-a, Ps-b, ..., Ps-n Mail items which pass through the sorting system in the first method step and which belong to the random sample Ps-1, Ps-2, ..., Ps-N Mail items which pass through the sorting system in the second method step

Claims (10)

Method for transporting objects to destination points,
wherein in a training phase, the steps are performed that a random sample of objects to be transported (Ps-a, Ps-b,...) passes through a classification device, - wherein the classification device comprises an image pickup device (1) and a grouper (3), - the image pickup device (1) generates at least one image (Abb-a, Abb-b, ...) of each article (Ps-a, Ps-b, ...) of the sample, - The grouper (3) automatically detects by applying a clustering method on the images (Fig-a, Abb-b, ...), which different image patterns on at least one object (Ps-a, Ps-b, .. .) of the sample, and the image patterns determined in this way are stored in an image data memory (4), and wherein in a sorting phase for each object to be transported (Ps-1, Ps-2, ...), the steps are performed that - the object to be transported (Ps-1, Ps-2, ...) passes through a sorting plant, - wherein the sorting system comprises an image pickup device (1) and a picture pattern recognizer (5), the imaging device (1) generates at least one image from the object to be transported (Ps-1, Ps-2,...), - The image pattern recognizer (5) by evaluating the image (Fig-1, Fig-2, ...) automatically determines whether an object stored in the image data memory (4) image pattern is displayed on the object, and if so which . - The sorting system, when on the object (Ps-1, Ps-2, ...) such an image pattern is displayed, using the image pattern determined a target point to which this object is to be transported, and - The sorting system triggers a transport of the object (Ps-1, Ps-2, ...) to this destination. Method according to claim 1,
characterized in that
the sorting system - Also includes the grouper (3) and - Is also used for the process steps of the training phase. A method according to claim 1 or claim 2,
characterized in that
the steps are carried out in the sorting phase for each image pattern stored in the image data memory (4) it is determined on which objects to be transported (Ps-1, Ps-2, ...) this image pattern is displayed, - a common target point is determined, to which all objects (Ps-1, Ps-2, ...) are to be transported with this image pattern, and - The sorting system triggers the transport of all objects (Ps-1, Ps-2, ...) with this pattern to the determined target point. Method according to claim 3,
characterized in that
for at least one image pattern, the determination of the target point comprises the steps that - one of the items is selected with this pattern, - By evaluating the at least one image of this selected item, the target point is determined. Method according to claim 3,
characterized in that
for at least one image pattern, the determination of the target point comprises the steps that
an image pattern destination data memory (6) is given,
in which a plurality of reference image patterns and for each reference image pattern in each case a destination information is stored,
the image pattern is compared with reference image patterns in the image pattern destination data memory (6),
as a result of the comparison, a reference image pattern is selected and
the destination point information of the selected reference image pattern is determined and used as the common destination point. Method according to one of claims 1 to 5,
characterized in that
several image characteristics are specified,
the grouper (3) in the training phase - for each image of a sample of the sample, determine the value each given feature will take on that image, in each case a characteristic value vector is calculated for each image and the feature value vectors are used to determine which different image patterns are displayed on the objects. Method according to claim 6,
characterized in that
the grouper (3) in the training phase - All images whose feature value vectors differ from each other by at most one predetermined or calculated in the training phase barrier, grouped together, and - automatically decides that in two images, which belong to two different groups, different image patterns are displayed. Method according to claim 6 or claim 7,
characterized in that
the step of storing the image patterns determined in the training phase in the image data memory (4),
the step comprises storing a computer-accessible description of an area in the space of the possible feature value vectors, and
the step of the image pattern recognizer (5) automatically determining in the sorting phase whether and, if so, which image pattern is displayed on an image,
the steps comprises that the image pattern recognizer (5) - Calculates a characteristic value vector for the image and - Compare this feature value vector with the stored in the image data memory (4) feature value vector descriptions. Device for transporting objects to destination points,
the device a classification device, - a sorting plant and a picture data memory (4) includes,
the classification device comprises an image capture device (1) and a grouper (3),
the sorting installation has an image recording device (1) and a picture pattern recognizer (5),
the device is designed so that in a training phase, the steps are performed that - a sample of objects to be transported (Ps-a, Ps-b, ...) passes through the classification device, - the image pickup device (1) generates at least one image (Abb-a, Abb-b, ...) of each article (Ps-a, Ps-b, ...) of the sample, - The grouper (3) automatically detects by applying a clustering method on the images (Fig-a, Abb-b, ...), which different image patterns on at least one object (Ps-a, Ps-b, .. .) of the sample, and - The device stores the thus determined image pattern in a pattern data memory (4), and the device is further configured to carry out the steps in a sorting phase for each object to be transported (Ps-1, Ps-2,...) - the object to be transported (Ps-1, Ps-2, ...) passes through a sorting plant, the imaging device (1) generates at least one image from the object to be transported (Ps-1, Ps-2,...), - The image pattern recognizer (5) by evaluating the image (Fig-1, Fig-2, ...) automatically determines whether an object stored in the image data memory (4) image pattern is displayed on the object, and if so which . - The sorting system, when on the object (Ps-1, Ps-2, ...) such an image pattern is displayed, using the image pattern determined a target point to which this object is to be transported, and - The sorting system triggers a transport of the object (Ps-1, Ps-2, ...) to this destination. Device according to claim 9,
characterized in that
the sorting system - Also includes the grouper (3) and - Is also designed for the implementation of the method steps of the training phase.

Images