Methods and arrangements for identifying objects

Abstract

In some arrangements, product packaging is digitally watermarked over most of its extent to facilitate high-throughput item identification at retail checkouts. Imagery of such packaging can be analyzed to detect digital watermarking. One claim recites a method comprising: receiving imagery representing two packaged item from a digital camera, each of the packaged items including digital watermarking encoded in packaging on the items; each of the packaged items including digital watermarking encoded in packaging on the items; using one or more programmed electronic processors, analyzing the imagery to detect the digital watermarking; accessing a spatial template associated with the digital watermarking when two or more instances of digital watermarking are detected in a single image frame; comparing spatial areas containing the two or more instances of digital watermarking within the imagery to the spatial template; carrying out an action based on the comparing. Of course, other claims, methods, apparatus and systems are described in this patent document.

Description

TECHNICAL FIELD

The present technology concerns object identification and is suited, e.g., for identifying objects at supermarket checkouts and other retail locations.

BACKGROUND AND SUMMARY

The widespread use of barcodes has greatly simplified supermarket checkout. However, many problems persist, causing both inconvenience for shoppers and added costs for retailers.

One of the difficulties is finding a barcode on a package. While experienced supermarket clerks eventually learn barcode locations for popular products, even the best clerks sometimes have difficulty with less common products. For shoppers who use self-service checkout stations, any product can be confounding.

Another issue concerns re-orienting a package so that its barcode is in position for reading.

Digital watermarking can be placed on product packaging—preferably over the majority of the package—to improve checkout speed. Methods and systems for improving watermark detection from imagery obtained from retail checkout cameras are described herein.

The foregoing and a great number of other features and advantages of the present technology will be more readily apparent from the following detailed description, which proceeds with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIGS. 1A-1F show a small can moving along a conveyor at a retail checkout station. The can enters the image frame in FIG. 1A and leaves the frame in FIG. 1F. The can is moving right to left in the images.

FIGS. 2A and 2B illustrate possible watermark detection block patterns.

FIGS. 3A-3J show composite images of a coffee can moving along a conveyor at a retail checkout station. Detection blocks are shown in the upper right quadrant of each image composite.

FIGS. 4A and 4B show detection results from test scans.

FIG. 5 shows an example detection block prioritization in an image frame.

DETAILED DESCRIPTION

In accordance with one aspect of this disclosure, the present technology concerns a method for identifying items, e.g., by a retail checkout system. A first such method involves moving an item to be purchased along a path, such as by a conveyor. A first camera arrangement captures first 2D imagery (e.g., image or video) data depicting the item when the item is at a first position along the path. Suitable 2D imaging scanners are provided, e.g., by DataLogic ADC INC., located in Eugene, Oreg., USA.

The moving item preferably includes digital watermark printed or carried on the product packaging. The digital watermarking may span a substantial portion of the packaging extent. In regions where there is no printing (e.g., white space), a yellow or other unobtrusive watermark tint can optionally be applied. (Yellow watermarking is particularly discussed, e.g., in application Ser. No. 12/774,512, filed May 5, 2010, published as application no. US 2011-0274310 A1, and U.S. Pat. No. 6,345,104, each of which is hereby incorporated herein by reference in its entirety.)

Digital watermarking patterns can be applied to items in a tiled fashion, with a single square (or other shape) watermark pattern being replicated across and down the item being watermarked. The tiles are typically embedded with an upper left corner of a first tile coincident with the upper left corner of the artwork. Tiles are then placed across and down from this starting point. In other cases, tiles are placed to coincide with or within specific spatial areas on packaging.

Each watermark pattern may have an orientation. Common practice is to embed the watermark tiles so that they are oriented in the same manner as the artwork (i.e., with “up” in the artwork corresponding to “up” in the watermark pattern). Differing watermark orientation, however, can provide an indicator of different watermarked areas. For example, a first watermarked area may include a first orientation and a second watermarked area may include a second orientation. Identification of the different watermark orientations may help distinguish different spatial areas on the product packaging.

Examples of watermarking are discussed, e.g., in assignee's U.S. Patent Application No. 61/693,106; U.S. Pat. No. 8,199,969; and published application no. US 2010-0150434 A1. Each of these patent documents is hereby incorporated herein by reference in its entirety.

Watermark embedding can be optimized too, e.g., as discussed in assignee's U.S. provisional application No. 61/749,767.

Now back to some checkout scenarios. Recall from above that an item to be purchased moves along a path, such as a conveyor. A first camera arrangement captures image data depicting the item when the item is at a first position along the path.

The next sections discuss a prioritization watermark detector blocks within a captured image frame(s), e.g., across 30 frames or more per second. A watermark detector “block” may correspond to an image area (or specific image features) that the watermark detector will analyze to determine whether a watermark signal is hidden therein. For example, with reference to FIG. 5, six (6) areas of an image are blocked out for analysis by a watermark detector. A prioritization can determine which of these blocks will be analyzed first (block 1), second (block 2) and so one through block 6. If a watermark is found in block 1 or 2, the watermark detector can be configured to stop looking at the other blocks 3-6. This is a different scheme than some traditional watermark detectors have taken. For example, some detectors would start watermark detection analysis with a block in a corner of the image (shown with dashed block in FIG. 5) and proceed to analyze blocks vertically across and horizontally down the image block by block.

Sometimes, a digital watermark detector is fed an image of larger resolution (e.g., 1024×1280 pixels) than what is covered by a single watermark detector block (e.g., 256×256 pixels). If a watermark detector is looking at image areas one detector block at a time, then it may take the watermark detector multiple runs to perform a watermark detection analysis across a whole image frame. Given constraints on computational resources from the hardware (e.g., embedded device, ARM processor, etc.), it may be difficult to cover the whole area of every frame in a timely manner (e.g., as packaged items are buzzing by on the conveyor past the camera). Therefore, it may be beneficial to limit the number of blocks that are analyzed by a watermark detector per image frame. For example, a watermark detector may only select 3-21 blocks per image frame for detection analysis.

This may not be an issue for well-marked large packages because they fill a large portion of a field of view of a camera and, thus, the chances of a single block detector being placed on a watermarked area is high. On the other hand, small packages, like cans and small boxes (e.g., a tea box), may only show up in a small portion of the camera's field of view, as shown in FIGS. 1A-1F, making the chance of a single block detector being placed on a well watermarked area very low. FIGS. 1A-1F represent images from an in-counter barcode scanner camera with a moving soft drink can, e.g., of a small Red Bull can size, moving from right to left. FIG. 1A (the entering frame) and FIG. 1F (the leaving frame) are not considered good candidates for watermark detection, because the can occupies such a small space within the whole image frame.

During a normal checkout pace, and when the camera is running at a speed of, e.g., 30 Frames Per Second (FPS), a typical small package will show up in 2 to 4 frames with suitable watermark detection presence, as shown in FIGS. 1B-1E. Since a small package covers a small area of the camera's field of view, a strategy of reading a watermark hidden in the small package's packaging from many different detection blocks across the whole image frame may have diminishing returns in terms of complexity vs. successful reads. Possibly, a watermark detector may spend time looking for watermarks in the background or on the package's boundary but not on the package itself.

When dealing with a video stream, we have found that background subtraction from the moving average of previous frames is a computationally efficient and effective method to extract the fast moving foreground objects. This method can separate static or slow moving objects (classified as background) from fast moving objects (classified as foreground), and place more single-block watermark detector blocks on more meaningful areas (e.g., in terms watermarked areas) of foreground objects.

Foreground detection can be configured to work as follows:1. Background(k+1)=alpha*Frame(k+1)+(1−alpha)*Background(k),2. Foreground(k+1)=Frame(k+1)−Background(k+1), if Frame(k+1)−Background(k+1)>threshold,where indices k or k+1 represent the incoming frame's temporal axis, alpha is the learning rate which controls how to update background from the incoming frame, and the threshold can be set to help suppress noise from illumination variations.

This process is computationally efficient because it may use pixel-wise subtraction, addition and comparison. Also memory usage is low since it does not save all previous frames but only a weighted average of some of the more recent frames. By efficient post-processing and clustering the results of each pixel, or groups of pixels, approximate information about location/shape of the foreground object can be obtained. Processing can be done in real time, or near real time.

The location and/or shape of the object can be utilized to constrain the area where watermark detector blocks should be placed. Significant savings in the computational complexity can be achieved without losing detection robustness.

Once the foreground region has been detected, we can assign detector block locations in the imagery to enhance detection. For example, the block patterns shown in FIGS. 2A and 2B can be placed over the foreground area (e.g., over the can in FIG. 1C). There are 6 detector blocks in FIG. 2A and 11 blocks in FIG. 2B. The two (2) darker blocks in FIG. 2B can be at a higher watermark resolution than the nine (9) lighter colored blocks. In some detection scenarios the patterns of these figures can be combined to identify 17 detection blocks per image frame. The imagery within each of these 17 blocks can be analyzed by a watermark detector to determine whether one or more watermarks are hidden therein. As discussed above, the detector may optionally be configured to cease detection of remaining blocks once a watermark is identified.

A first option uses a determined foreground region to trim down the FIGS. 2A & 2B patterns. For example, in the case of a combined 17 detection blocks (e.g., combining FIGS. 2A & 2B's patterns to yield 17 watermark detection blocks), a particular detection block will only be used by a watermark detector if the block falls inside (or, in some cases, overlapping with) a determined foreground region. In the case of overlapping, a detector can set a predetermined threshold, e.g., 75% or more of overlap with the foreground region.

A second option is now discussed.

As a first action in the second option, the foreground region in captured imagery can be expanded to a square region or other shaped window, enclosing all (or most of) the foreground pixels. Then the square foreground region can be divided into zones, e.g., equally spaced zones. The foreground pixels (e.g., as determined by incoming pixels minus background pixels) inside each zone can be summed together. This summation is a representation of the illumination of the foreground.

As a second action in the second option, two approaches can be used to prioritize the placement of detecting blocks (e.g., areas in which the watermark detector will look for watermark signal) inside the square foreground region, because the number of single block analysis areas may not be enough to cover the whole region.

The first approach is based on illumination (or brightness). The foreground zones are ranked according to their illumination, with those of a higher rank indicating a relatively better illumination compared to those of lower rank. We would prefer not to place the single block detectors on poor illuminated zones, so the watermark detector is configured to avoid those poor illuminated zones. Also, the watermark detector may be configured to discard or ignore zones with high illumination values because they may indicate over-saturated pixels from glare (e.g., caused by specular reflection from the packaging by a scanner illumination).

The second approach is based on a geometric position of each zone. In some cases the zones or areas near the top of the frame and near the bottom of the frame detect poorly, due to over-saturated pixels on the top and poor illuminated pixels on the bottom. So a weighting can be assigned to each zone based on its geometric locations within an image frame. For example, central zones may be weighted more significantly than zones close to the frame boundary. Or zones close to frame boundary may only be considered if no watermark is initially found in central frame zones.

To combine the two approaches, we can add a normalized illumination value of each zone with the weight of each zone from its geometric position, and then do an ascending ranking of detection blocks. Those zones with a higher value will have higher detection priority in acquiring a single-block detector.

The above second option is illustrated in FIG. 3A-3J. The minimum separation between selected detection blocks is set to a predetermined pixel value, e.g., 64 pixels between each block, to avoid choosing blocks with too much overlap (i.e. blocks that are from similar image areas). These FIG. 3 images show 10 composed frames of a coffee can imaged from an in-counter barcode scanner camera. Each image includes the incoming frame (upper left quadrant), detected square foreground region (lower left quadrant), and up to 7 detection blocks overlaid on top of the incoming frame (upper right quadrant).

We have performed some experiments to verify our process. The test datasets we used are ad-hoc captures from non-professional checkers simulating a supermarket checkout process. We have two datasets, one labeled Scan-1 and the other Scan-2. The Scan-1 dataset contains mostly small packages (e.g., small cans) and has 1025 frames of about 30 seconds recording from an in-counter barcode scanner camera, and the Scan-2 dataset contains both small and large packages and has 596 frames from the same camera. The Scan-2 dataset contains more frames with packages inside so it has more frames detected as containing watermarking.

The results of using the first option, which uses a determined foreground region to trim down the B17 pattern (including 17 detection blocks), are shown in Table 1. There are 168 frames and 53 frames detected as containing watermark from Scan-2 and Scan-1 datasets, respectively, using the fixed static B17 pattern. By switching to the flexible foreground trimmed B17 (e.g., focuses on blocks that fall within the foreground area), to get the same detection rate, on average, only 10 frames are required for Scan-2, and only 7 frames are required for Scan-1. Since Scan-1 contains more small packages, and the benefits of using foreground detection is more obvious on small packages, the savings in terms of number of blocks per frame is more significant for Scan-1.

TABLE 1
Compares flexible pattern from foreground with fixed
static pattern in placing single block detectors.
SCAN-2	168 frames detected with	168 frames detected w. fixed
	flexible foreground placement	detection pattern
	2680/275 = 9.75 block/frame	17 block/frame
SCAN-1	53 frames detected with	53 frames detected with fixed
	flexible foreground placement	detection pattern
	978/162 = 6.04 block/frame	17 block/frame

The results of using the second option are shown in FIGS. 4A and 4B. The straight-dashed lines mark the number of detected frames from Scan-2 and Scan-1 datasets using the static B17 pattern, 168 and 53, respectively. The curves indicate the number of detected frames when choosing different number of blocks for placement inside a foreground region. In general, when you have large set of detector blocks, say, e.g., 17, then the flexible block pattern give better detection results. And if you limit the number of detector blocks, say, down to 9, the flexible block pattern still gives a good detection rate with much reduced computational cost.

In other cases, a “smart watermark detector,” one that can train itself based on user or cashier habits or preferences, is preferred. For example, through a series of training check-out runs, it can be determined that cashier 1 holds packaged items at a certain angle, or at predetermined distances from the camera, or at a certain swipe speed, or places items on a conveyor at certain orientations. Other training information may include, e.g., proximity to the scanner, speed of scanning, production rotation habits, professional vs. amateur checker speed or habits or characteristics, etc. Or the detector may determine that they are only getting watermark reads from certain block areas when a certain checker checks out. All this information (or subsets of this information) can be used to adjust the watermark detector, e.g., by determine which blocks to prioritize in a detection process. For example, it might be found that cashier 1 always swipes items in front of the camera so that the packages are in the top or bottom of the field of view. Whereas, above, these block areas would typically be prioritized low. But if the detector knows that cashier 1 is checking out then these areas can be more highly prioritized. For example, these blocks are analyzed for watermarking prior to blocks located in other areas of the field of view. While this passage has focused on digital watermarking, similar user-specific information can be used to adjust, e.g., an image recognition or fingerprinting process.

A user's self-checkout habits—including how and at what speed they present objects to the check-out camera—can be monitored and then when they are determined they can be used to help configure a watermark detector, e.g., by prioritize block selection for watermark detection. A store loyalty card, customer ID or other identifier can be associated with a database or record storing the proper detector prioritization. That prioritization can then be used to inform the watermark detector on how to better process imagery for that person checking out. For example, a customer shows their loyalty card or enters (e.g., types on a touch screen or speaks to a voice recognition unit) prior to checking out. The customer's checkout profile is identified and applied to the watermark detector. For example, it might be determined that a particular customer holds packages at a certain pose. Captured imagery can be adjusted prior to watermark detection to adjust for the pose using pose estimation and homography, e.g., as discussed in assignee's U.S. patent application Ser. No. 13/789,126, filed Mar. 7, 2013, issued as U.S. Pat. No. 9,684,941, which is hereby incorporated herein by reference in its entirety.

Some checkout stations will continue to monitor barcodes even if supplemental symbologies like watermarking are present during checkout. In these cases please consider the flowing flow:

• • 1. Imagery is presented to a watermark detector.
  • 2. The watermark detector analyzes the imagery and detects a watermark. The watermark may include a payload or index or other information.
  • 3. A process is invoked that utilizes that watermark information, index or payload (or portions thereof) to create or obtain an image overlay for captured imagery. The image overlay preferably includes a barcode or other symbology that includes the watermark information, or information obtained from utilizing the watermark information. That way, if the same imagery that is analyzed for a digital watermark is then feed to a barcode reader the graphic overlay barcode will be easily recognizable even if the depicted product packaging did not display a barcode.

One challenge may occur if two or more of the same packaged items are within a single image frame. For example, 2 cans of diet Mountain Dew might be pictured in the same frame. The watermark detector finds a read, but in different, non-contiguous image areas. In such cases a watermark payload may be used to look up a spatial template. The spatial template is sized roughly to represent a particular item (e.g., diet soda). The spatial template is placed around a block area where watermarks were detected. If watermarks (or watermark components like orientation components) are located outside of the spatial template (or outside of a predetermined area or tolerance) then there is a likelihood that the image frame includes two or more watermarked objects.

The cashier can be warned to examine this area more carefully, or the system may make a determination to independently ring up both items.

In another implementation, the checkout camera includes or cooperates with special illumination. The illumination projects watermark orientation information on the packaging. The projected illumination is captured along with the packaged items. The projected orientation information is deciphered by the watermark detector to help determine positioning information including relative depth, orientation, etc. This information can be used in watermark detection, or foreground/background decisions.

In still another implementation, watermarks are used to identify certain areas on packaging. For example, a watermark signal (e.g., an orientation component) might be used to outline the nutrition facts on a package. The watermarked area is then used to create a spatial position on a reading device (in this case, e.g., a smartphone like an iPhone or Android device). An augmented reality display is overlaid on the watermarked area, e.g., as discussed in assignee's U.S. patent application Ser. No. 13/789,126 (issued as U.S. Pat. No. 9,684,941).

CONCLUDING REMARKS

This specification details a variety of embodiments. It should be understood that the methods, elements and concepts detailed in connection with one embodiment can be combined with the methods, elements and concepts detailed in connection with other embodiments and with those discussed in Appendix A. While some such arrangements have been particularly described, many have not—due to the large number of permutations and combinations. However, implementation of all such combinations is straightforward to the artisan from the provided teachings.

Although features and arrangements are described, in some cases, individually, the inventors intend that they will also be used together. Conversely, while certain methods and systems are detailed as including multiple features, the inventors conceive that—in other embodiments—the individual features thereof are usable independently.

The present specification should be read in the context of the cited references (with which the reader is presumed to be familiar). Those references disclose technologies and teachings that applicant intends be incorporated into certain embodiments of the present technology, and into which the technologies and teachings detailed herein be incorporated.

For example, with the documents cited in Appendix A, certain of the cited references teach that a single image sensor can be used, in conjunction with mirrors and other optics, to capture imagery of an object from two or more different views (as opposed to using two or more different cameras). Some such arrangements use wavelength-selective optics (e.g., dichroic mirrors) so that three different images can be projected onto common pixel elements of a single sensor, allowing separate processing of the three different images in different color channels. Other such arrangements use mirrors to project images from different viewpoints onto different rectangular sub-regions of a common sensor. Still further, other of the prior art teaches that a color (RGB) image sensor can be used to capture imagery of an object, yet object identification can proceed using just pixels of a single color (e.g., whichever color shows the highest variance in its histogram). The artisan will recognize that these and all the other arrangements taught in the cited art can utilize the methods and features detailed herein, and vice versa.

To provide a comprehensive disclosure, while complying with the statutory requirement of conciseness, applicants incorporate-by-reference each of the documents referenced herein. (Such materials are incorporated in their entireties, even if cited above in connection with specific of their teachings.)

Although not particularly illustrated, it will be recognized that the methods described above can be implemented using general purpose (or special purpose) computers, e.g., comprising one or more processors, multi-core processors, semiconductor memory, hard disks, networking connections, and input-output devices, as are conventional in the art. Software instructions for implementing the above-detailed methods can be stored on tangible media associated with such systems, e.g., disks and semiconductor memories.

While the focus in the above sections has been on digital watermark recognition, many of the above techniques will also enhance other object-identifying techniques such as a barcode, optical character recognition, image recognition, fingerprinting. For example, the foreground identification techniques above can be used to locate frame areas to derive fingerprints.

In view of the wide variety of embodiments to which the principles and features discussed above can be applied, it should be apparent that the detailed embodiments are illustrative only, and should not be taken as limiting the scope of the invention. Rather, we claim as our invention all such modifications as may come within the scope and spirit of the following claims and equivalents thereof.

Claims

1. A method comprising: receiving imagery representing two packaged items from a digital camera, each of the packaged items including digital watermarking encoded in packaging on the items;using one or more programmed electronic processors, analyzing the imagery to detect the digital watermarking;accessing a spatial template associated with the digital watermarking when two or more instances of digital watermarking are detected in a single image frame;comparing spatial areas containing the two or more instances of digital watermarking within the imagery to the spatial template;carrying out an action based on said comparing.

2. The method of claim 1 in which the action comprises determining that the imagery includes two packaged items.

3. The method claim 2 in which the action further includes charging a customer for the two packaged items.

4. The method of claim 2 in which the action further includes signaling that two items are presented for checkout.

5. The method of claim 1 in which the two or more instances of digital watermarking are detected in non-contiguous imagery areas, and said accessing a spatial template associated with the two or more instances of digital watermarking occurs when the two or more instance of digital watermarking are detected in the non-contiguous imagery areas within the single image frame.

6. The method of claim 1 in which the spatial template is sized according to the two packaged items.

7. The method of claim 6 in which the spatial template is placed around an image block area where an instance of the digital watermarking is detected.

8. The method of claim 7 in which said comparing spatial areas containing the two or more instances of digital watermarking within the imagery to the spatial template comprises determining that the digital watermarking falls within a predetermined tolerance associated with the spatial template.

9. The method of claim 1 in which the digital watermarking comprises a message component and an orientation component.

10. The method of claim 9 in which the message component is associated with the spatial component.

11. An apparatus comprising: a 2D-camera for capturing imagery;memory for storing captured imagery;one or more programmed electronic processors configured for: detecting digital watermarking from imagery captured by said 2D-camera, the imagery representing two packaged items, each of the two packaged items including digital watermarking encoded in packaging on the two packaged items, the digital watermarking comprising a message component;using the message component, obtaining a spatial template when two or more instances of digital watermarking are detected within a single image frame of the captured imagery;determining whether spatial areas within the captured imagery that are associated with the two or more instances of digital watermarking correspond in a predetermined manner to the spatial template, the determining yielding a result;executing an action based on the result of the determining.

12. The apparatus of claim 11 in which the action comprises determining that the captured imagery includes two packaged items.

13. The apparatus claim 12 in which the action further includes charging a customer for the two packaged items.

14. The apparatus of claim 12 in which the action further includes generating a signal indicating that two items are presented for checkout.

15. The apparatus of claim 11 in which the two or more instances of the digital watermarking are detected in non-contiguous imagery areas, and the obtaining a spatial template when two or more instances of digital watermarking are detected within a single image frame of the captured imagery occurs when the two or more instance of digital watermarking are detected in the non-contiguous imagery areas within the single image frame.

16. The apparatus of claim 11 in which the spatial template is sized according to the two packaged items.

17. The apparatus of claim 16 in which the spatial template is placed around an image block area where an instance of digital watermarking is detected.

18. The apparatus of claim 17 in which the determining whether spatial areas within the captured imagery that are associated with the two or more instances of digital watermarking correspond in a predetermined manner to the spatial template comprises determining that the digital watermarking falls within a predetermined tolerance associated with the spatial template.

19. The apparatus of claim 11 in which the digital watermarking comprises an orientation component.

20. An apparatus comprising: means for capturing imagery;means for storing captured imagery;means for detecting digital watermarking from imagery captured by said means for capturing imagery, the imagery representing two items, each of the two items including digital watermarking encoded thereon, the digital watermarking comprising a message component;means for obtaining a spatial template, using the message component, when two or more instances of digital watermarking are detected within a single frame of the imagery;means for evaluating whether spatial areas within the captured imagery that are associated with the two or more instances of digital watermarking correspond in a predetermined manner to the spatial template, said means for evaluating producing a result;means for controlling an action based on the result.