A waste sorting method combining chemical composition analysis and the use of markers.

Abstract

The present invention relates to a method for sorting objects, the method comprising the steps of: placing objects on a conveyor belt (BCV); collecting an image of a first zone (Z1) of the belt; collecting material and / or color detection signals for cells (CL) adjacent to a second zone (Z2) of the belt; processing the detection signals to determine the detected attributes; detecting and decoding markers in the image to determine specific identifiers and attributes; determining a first command for each cell associated with a single object identifier; determining a second command for each cell associated with material and / or color attributes; and making a sorting composite decision according to the first and second sorting commands and combined rules for the cells.

Description

Technical Field

[0001] The present invention relates to the field of waste sorting, and more particularly to the field of sorting waste of package type made of plastic or fibrous materials based on various criteria such as material, color, shape, use, viscosity, presence of bio-based carbon and recycled materials.

Background Art

[0002] Generally, waste sorting relies on various techniques for directly detecting the chemical composition of the objects to be sorted. In these techniques, various detection tools are often combined with each other for more delicate diagnosis. Such tools can include, for example, spectrometers in the near-infrared or mid-infrared regions for molecular analysis of organic products (plastic, cellulose-based or bio-derived products), color cameras or spectrometers in the visible light region for analysis of the color of objects; electromagnetic induction sensors for characterizing the presence of metals, X-ray sensors or laser-induced breakdown spectroscopy (LIBS) spectrometers for analysis of atomic composition.

[0003] Similarly, there are image processing systems that include, in particular, shape recognition algorithms based on artificial intelligence. However, such systems have been found to be not very suitable for waste processing that can be significantly degraded, such as deformation or dirt. Despite the existence of a very diverse range of detection tools, it is still impossible to determine other important characteristics for recycling, such as the food or non-food use of the object, the presence of dangerous contents, the presence of a layer hiding part or all of the object (in the case of multi-layer packages), the type and shape of the object, the presence of additives in the material constituting the object, particularly the viscosity index when the object is made of PE-HD (high-density polyethylene), the ratio of recycled materials contained in the object and the proportion of bio-based carbon.

[0004] It has already been proposed to attach chemical tracers to objects being sorted in order to characterize desired properties. This thus makes properties indirectly detectable based on the principle of self-declaration. However, using such tracers presents various disadvantages. This is because the importance of such tracers must be discussed and approved by all parties involved. Associating tracers with objects carries the risk of errors and fraud. Coding capacity is limited, generally consisting of one bit of information per tracer. Tracers must ultimately be removed from objects during recycling so as not to interfere with subsequent sorting / recycling.

[0005] An international barcode system is also known that assigns a single identifier to each product number on a package. This identifier provides access to a set of technical data for the product and, therefore, all attributes that are potentially relevant for recycling. However, reading a single barcode requires the barcode to be facing the reader directly. Therefore, this marking technology is not suitable for the automated sorting of waste that is randomly present on a conveyor belt. A new version of the barcode based on QR codes (registered trademark) is in the process of standardization, but this also suffers from the same drawback of only being present in one place on the package.

[0006] RFID chip marking systems are also known, which allow for the assignment of a single identifier to each product. Comprehensive reading can be considered without the need to orient the products. However, RFID chips are expensive compared to the cost of the packaging, and they cannot accurately pinpoint the location of objects on a sorting belt.

[0007] In the 2010s, a new type of marker for packaging was proposed. This is called "Digital Watermarks," and will be referred to as "markers" below. Digital watermarks consist of adding a printed or embossed (formed with a mold or engraved) pattern to the label or packaging of an object. Whether printed or engraved, these patterns have a very high resolution of approximately 75-150 ppp, i.e., pixels with sides of 0.1-0.2 mm. The grayscale of the print or the variation in the depth of the embossing is calculated so that the marker is invisible to the naked eye, while it is always visible to a high-speed camera with appropriate lighting (monochrome or white pulsed illumination or constant illumination).

[0008] The European consortium Holy Grail 2.0 has launched an industry evaluation program in Europe since 2020 to adopt such packaging markers. This program is based on technologies developed by industry stakeholders such as Digimarc® and Filigrade®.

[0009] Compared to chemical tracers, markers do not require the addition of external substances and leave no trace during the recycling process. Furthermore, the richness of their information is orders of magnitude greater. That is, a pattern with sides of 1 cm to 2 cm can contain 10 to 50 bits of information depending on the selected technological solution. In contrast, a system equivalent to 3 bits using three different types of chemical tracers can only encode seven different situations for such marked objects (although eight combinations are possible, the complete absence coded as 0,0,0 cannot distinguish between marked and unmarked objects), which is overwhelmingly insufficient considering the number of types of marked objects. Therefore, the richness of marker information is comparable to the richness of barcode information. Unlike barcodes, these markers can be formed on the surface of an object, or even in multiple locations across the entire surface of an object, and can be automatically read without requiring the reader to present a precise side of the object.

[0010] Therefore, regardless of the orientation or condition of the object, even if it is crushed, partially obscured, or dirty, the marker remains visible. These characteristics allow for consideration of using such markers and for performing optical sorting before recycling. [Overview of the Initiative] [Problems that the invention aims to solve]

[0011] This invention proposes realizing a detection tunnel based on a row of cameras positioned on a sorting belt in combination with a conventional optical sorting system, and using physical sensors such as spectrometers in the near-infrared (NIR) and / or visible light regions to determine the composition of the materials.

[0012] Accordingly, the present invention proposes a method for sorting marked or unmarked objects using the same operation, combining direct and indirect detection by detecting markers that provide information undetectable by direct detection. In fact, for objects such as recyclable packaging, it is desirable to enable sorting based on a complex combination of criteria such as material, color, shape, use, viscosity, bio-based carbon, and the presence of recycled materials, given that markers are not necessarily present on the objects to be sorted, or if present, they are only detectable on a part of the object's surface. [Means for solving the problem]

[0013] Various embodiments relate to a method for sorting objects, comprising the steps of: placing the objects to be sorted on a belt of a conveyor belt; collecting an image flow of objects present in a first belt zone extending across the entire width of the belt; collecting material and / or color detection signals for each cell in a group of adjacent cells located in a second zone extending across the entire width of the belt; processing the detection signals for each cell to determine the detected material and / or color attributes; detecting markers in the image flow of the first zone; and for each detected marker, decoding the marker to determine an object identifier. The process includes the steps of: determining; assigning an object identifier determined by the decoding of the marker to the cell covered by each decoded marker; determining specific attributes for each object identifier determined based on the decoded marker; determining a first sorting command for each cell associated with the object identifier according to the sorting criteria and specific attributes; determining a second sorting command for each cell according to the sorting criteria and detected attributes; and determining a sorting composite decision for each cell according to the first and second sorting commands and combination rules for that cell.

[0014] In this way, multiple values ​​of specific object attributes are obtained based on object identifiers obtained by decoding markers present in the collected image flow, and material and / or color determination signals. By applying sorting criteria that can be combined with these values, marked objects mixed in with unmarked objects can be detected within the object flow to be sorted. Combining marker decoding with material detection of the objects to be sorted offers numerous advantages, particularly in terms of detection accuracy. In fact, in most cases, the detected markers do not cover the entire object. This is especially true when the object is not marked across its entire surface, when part of the label is peeling off, or when the algorithm fails to detect all markers (computation time is too long, the marked surface is excessively damaged or dirty, the object is tilted, etc.).

[0015] Furthermore, the detected material may not correspond to the material declared by the marker, for example, if the same mold is used for multiple materials, or in the case of a marker printing error. In the event of a discrepancy, priority is given to the detection signal for the material and / or color whose measurement is an actual value, rather than the declaration of inconsistency with the data obtained by decoding the marker.

[0016] Even when marking technology is employed on a large scale, unmarked products will always exist within the flow of marked objects to be sorted. These objects can be sorted using material detection devices. For example, sorting criteria can be configured to remove contaminants consisting of unmarked objects from the flow of marked objects and increase purity.

[0017] This combination also reduces redundancy and the computational output required. In fact, algorithms that detect and decode markers require enormous computational output because they use significant redundancy between consecutive images. By reducing the imaging frequency while ensuring the decoding of at least one marker per object, and by using material and / or color detection signals to complement the object map, it is possible to maintain maximum detection efficiency even with a low imaging frequency.

[0018] According to one embodiment, this method includes several steps of extending the selection composite decision of cells associated with an object identifier to adjacent cells not associated with an object identifier by applying an extended set of rules.

[0019] According to one embodiment, this method comprises several steps of controlling the sorting device to orient each object present on the belt toward a selected exit, in accordance with the sorting composite determination of the cells located in front of the sorting device.

[0020] The precision achieved by controlling the sorting device is applied regardless of which sorting system outlet is selected for each marker. This ensures that the sorting control applied to all cells corresponding to the entire object is consistent, and therefore improves sorting purity.

[0021] According to one embodiment, depending on the set of combined rules, the sorting composite decision for a cell is determined as follows: if the first and second sorting commands for the cell are the same, it conforms to either command assigned to one of the sorting system exits; if the first and second sorting commands for the cell are contradictory, the exit of the sorting system is determined according to the sorting criteria; if the second sorting command is uncertain because there is no detection indicating material and / or color in the cell, it conforms to the first sorting command for the cell; if the second sorting command for the cell is determined and it is clear that the second sorting command takes precedence by the sorting criteria considering the detected attributes, it conforms to the second sorting command for the cell; and if the first sorting command is uncertain because no marker is detected in the cell, it is determined according to the sorting criteria and the sorting composite decision of the adjacent cell.

[0022] Therefore, for all values ​​of the first and second sorting commands, processing is performed taking into account the characteristics of the detection mode, which involves image capture and image processing, and material and / or color detection.

[0023] According to one embodiment, this method comprises the steps of: selecting at least one cell of a first extended rank that is not associated with one object identifier, wherein the sorting criterion assigns priority to a first command for this cell, and the sorting criterion assigns priority to the first command for this adjacent cell, and for the adjacent cell and the selected cell, the material identified by the marker corresponds to the material detected by the sensor assembly; and assigning a value equal to the value of one cell in this nucleus to the sorting composite determination of each selected cell. has a first expansion operation including

[0024] Such an expansion operation can increase the number of cells involved in the sorting control to be activated for the sorted object, and thus can improve the accuracy of the sorting control and reduce sorting control errors.

[0025] According to one embodiment, this method consists of the following steps, namely: a step of selecting at least one cell belonging to a higher expansion rank not associated with one object identifier, for which cell, the sorting criterion assigns a priority to a first command, this cell being adjacent to at least one cell belonging to a lower expansion rank, and the material detected by the sensor assembly corresponding to the material of the selected cell, the cell selection step; and a step of assigning a value equal to the value of the cell belonging to the lower expansion rank to the sorting composite decision of each cell selected for the higher expansion rank has a subsequent expansion operation including

[0026] With this configuration, the number of cells involved in the control to be activated for the sorted object can be further increased, so the accuracy of the sorting control is further enhanced, and thus the sorting efficiency of the sorted object is further improved.

[0027] According to one embodiment, a plurality of expansion operations are performed up to a maximum number of expansion operations defined for each outlet of the sorting system.

[0028] With this configuration, the number of cells involved in the sorting control to be activated for the sorted object can be further increased, so the sorting efficiency of the sorted object is further improved, and at the same time, the risk of overflowing into adjacent sorted objects and causing sorting errors can be limited.

[0029] According to one embodiment, the maximum number of expansion operations is defined for each identifier or each value of a specific attribute.

[0030] According to one embodiment, this method applies to each cell as follows: - First Selection Command - Second selection command -Selection and combined determination, - Extended rank, and - Object identifier obtained by decoding the marker This includes constructing a detection table that has the following characteristics.

[0031] According to one embodiment, determining the attributes identified based on each object identifier determined using a decoded marker includes a search step of object identifiers in a local database that stores object identifiers extractable from the marker and specific attributes associated with each object identifier.

[0032] According to one embodiment, the method includes the steps of: receiving a selection criterion; and updating a database storing object identifiers extractable from markers and specific attributes associated with each object identifier, wherein the database update includes determining a selection command for each object identifier in accordance with the selection criterion and the specific attributes associated with the object identifier, and a first selection command for each cell associated with one object identifier is a selection command associated with the object identifier in the database.

[0033] By using a database of marked objects, a first sorting command and specific attributes determined using the object identifier can be determined simply by reading the database.

[0034] According to one embodiment, this method includes the step of deleting object identifiers from a database that have not been read for a certain period of time in order to determine a first sorting command.

[0035] This configuration reduces the need to search for objects in the database based on object identifiers to retrieve associated sorting commands, and to update sorting commands stored in the database according to sorting criteria.

[0036] According to one embodiment, the sorting criteria are based on one or more specific or detected attributes, namely: the properties of the material, color, shape, use, viscosity of the material, the proportion of biogenic material in the object, and the proportion of recycled material in the object. Therefore, multiple selection criteria can be considered, enabling complex selection processes.

[0037] According to one embodiment, the image flow is collected by alternately illuminating a first zone of a belt under two colors of illumination, so that each point of an object moving through the first zone is illuminated at least once with each color of illumination, with the two colors of illumination positioned at opposite ends of the visible spectrum, respectively.

[0038] In this way, most markers on product packaging have a simpler structure than those that must provide three colors of illumination, and even with the same total number of light sources (LEDs), the illumination is stronger and detectable regardless of the package color. According to one embodiment, the image flow is collected alternately under two different colors of lighting, with each cell being illuminated sequentially under the two colors of lighting. According to one embodiment, one of the two colors of illumination may have a wavelength of less than 460 nm, and the other of the two colors of illumination may have a wavelength of more than 630 nm.

[0039] Similarly, multiple embodiments can be involved in the sorting system, which includes a belt conveyor including a belt on which objects to be sorted are placed; a camera assembly for collecting an image flow of objects present in a first zone of the belt extending across the entire width of the belt; a sensor assembly for collecting material and / or color detection signals for each cell of a group of adjacent cells located in a second zone extending across the entire width of the belt; an image processing unit for detecting markers in the image flow; a signal processing unit for identifying detection attributes determined based on the material and / or color detection signals; a sorting device for orienting each object present on the belt toward the exit of the sorting system; and a sorting control unit configured to carry out the above method.

[0040] According to one embodiment, the system includes a lighting system comprising two groups of monochromatic light sources that alternately illuminate a first zone. [Brief explanation of the drawing]

[0041] The present invention will be better understood by the following embodiments relating to the accompanying drawings. In the drawings, the same reference numerals correspond to the same or similar elements structurally and / or functionally. [Figure 1] Figure 1 is a schematic side view of an object sorting system according to one embodiment. [Figure 2] Figure 2 is a schematic diagram showing a detection and analysis device for a sorting system according to one embodiment. [Figure 3] Figure 3 is a schematic diagram showing an image of objects detected on a conveyor belt of a sorting system according to one embodiment. [Figure 4] Figure 4 is a schematic diagram showing an image of objects detected on a conveyor belt of a sorting system according to one embodiment. [Figure 5] Figure 5 shows the contents of the sorting table in one step of a sorting method according to one embodiment. [Figure 6]Figure 6 shows an example of the contents of a sorting table in another step of a sorting method according to one embodiment. [Figure 7] Figure 7 shows the contents of the sorting table in the next step of the sorting method according to one embodiment. [Figure 8] Figure 8 shows the contents of the sorting table in the next step of the sorting method according to one embodiment. [Figure 9] Figure 9 shows the contents of the sorting table in the next step of the sorting method according to one embodiment. [Figure 10] Figure 10 shows another example of a control table for a sorting device according to one embodiment. [Figure 11] Figure 11 is a schematic side view of an object sorting system according to another embodiment. [Modes for carrying out the invention]

[0042] Figure 1 shows an object sorting system according to one embodiment. The sorting system comprises a belt conveyor BCV including a belt B1 on which objects OB to be sorted are placed. Belt B1 is driven translationally in direction D1, carrying objects into a tunnel formed from a roof T1 positioned above this belt B1. The sorting system also comprises a detection device DTD including camera assemblies IS1-IS6 and a material and / or color detection sensor assembly IVS. The detection device DTD is positioned to observe belt zone B1 downstream of the accumulation zone Z0 of objects OB to be sorted. Camera assemblies IS1-6 can be associated with illumination devices VL1, VL2, which are focused to illuminate zone Z1 of belt B1 observed by the cameras at various wavelengths. Sensor assembly IVS observes zone Z2 of belt B1 downstream of zone Z1. The sorting system also comprises a sorting device BLS positioned at the output end of the conveyor BCV. The sorting device BLS includes, for example, multiple compressed air discharge nozzles, which are arranged along a direction L1 perpendicular to direction D1 and parallel to the plane of belt B1, and each nozzle is controlled by an individual solenoid valve. When one of these valves is opened, the air discharged from the thus opened nozzle pushes the object above the nozzle toward outlet S1, otherwise the object arriving at the exit end of conveyor BCV falls toward outlet S0 (due to gravity).

[0043] The sensor assembly IVS may include at least one of the following devices: Spectrometers that operate in the near-infrared or mid-infrared wavelength range and / or the visible region. LIBS spectrometer, One or more color cameras for detecting the color of objects on Belt B1, or more generally for recognizing the appearance of objects (shape, texture, color, etc.), Inductive sensors for characterizing the presence of metals, X-ray emission / detection assembly, and A surface shape measurement system or triangulation system for detecting black objects or objects that cannot be detected by the above detection techniques.

[0044] The sorting system may also include a turbine BL to guide air above belt B1 at a speed approximately corresponding to the speed at which belt B1 moves in direction D1, so that the lightest objects OB placed on the belt are transported in direction D1 at the same speed. The air supplied by turbine BL is guided within a tunnel formed from roof T1.

[0045] Figure 2 shows a detection device DTD combined with an analytical device of a sorting system according to one embodiment. Figure 2 shows the exit zone of belt B1, which includes a sensor assembly IVS, camera assemblies IS1-IS6, associated illumination devices VL1, VL2, and a sorting device BLS including a nozzle NZ and individual solenoid valves EV. Figure 2 shows, in particular, the fields of view observed by each of the cameras IS1-IS6 along the width (direction L1) of belt B1. As shown, the zones of belt B1 observed by cameras IS1-IS6 slightly overlap each other. Each camera IS1-IS6 is connected to an image processing unit PC1-PC6.

[0046] The sensor assembly IVS is connected to the signal processing unit PCS that receives signals from the sensor assembly. The processing units PC1-PC6 and PCS communicate with the sorting control unit MPC via the network NS. A terminal TL can be used to control the control unit MPC. Illumination devices VL1, VL2 include, for example, bars of light-emitting diodes (LEDs) positioned on both sides of the detection axis of cameras IS1-IS6.

[0047] According to one known technique, cameras IS1-IS6 are monochrome, and the LED bar contains two or three sets of monochromatic LEDs, each covering a specific color in the visible spectrum. The illumination is switched on sequentially according to a cycle controlled by a specially configured electronic unit LM. Image acquisition is synchronized among all cameras IS1-IS6 and also synchronized with one of the illumination colors, and the whole is controlled by the electronic unit LM. The frame rate is adjusted from 30 to 300 fps ("frames per second" or images per second). This is the total number of images with all colors mixed together. By operating the three colors sequentially at the same rate, 100 images per second are obtained for each color at 300 fps. By using different colors, the contrast for detecting markers on objects detected by belt B1 can be maximized. For example, if a marker is printed yellow on a white background, it is almost invisible to the naked eye, but when illuminated with blue light, it appears as black on a white background in a monochrome camera. When decoding the marker, it is not necessary to know the color that illuminated the detected object.

[0048] According to another known technique, illumination of zone Z1 observed by cameras IS1-IS6 can be achieved with pulsed or continuously lit white LEDs, and the cameras IS1-IS6 used are color cameras with three channels: RVB, i.e., red, green, and blue. Thus, each image capture provides three, i.e., one monochromatic sub-image per channel. This solution allows for a higher acquisition rate per color channel, but at the expense of reduced spatial resolution because the pixels of each channel are juxtaposed on the readout circuit and therefore shared among the three channels.

[0049] The inventors noted that two different colors of illumination are sufficient to detect all markers on a product package. Furthermore, multiple measurements showed that the best contrast was achieved by alternating illumination colors at both ends of the visible spectrum, namely blue illumination at the short-wavelength end (wavelengths less than 460 nm) and red or near-infrared illumination at the long-wavelength end (wavelengths greater than 630 nm). Using a different color of illumination did not provide a significant benefit. According to one embodiment, marker detection is performed by illuminating zone Z1 with alternating blue illumination and red or near-infrared illumination. This solution is highly advantageous because it allows the illumination output of the LED bar to be distributed between two colors instead of three. This results in a 50% increase in output per color, even with the same number of light sources.

[0050] In the embodiment shown in Figure 2, the camera assembly includes six cameras IS1-IS6 on a belt B1 with a width of 1.2 meters. Each camera has a field of view at a belt position 23 cm along axis L1 and 14 cm along axis D1. Each camera is connected to individual image processing units PC1-PC6. Naturally, other configurations are possible, especially depending on the computational output of the image processing units. For example, multiple processing units may be connected to one camera, or conversely, multiple cameras may be connected to one processing unit.

[0051] Each image processing unit PC1-PC6 is configured to analyze images supplied from the connected cameras IS1-IS6 and detect object markers present in each image. The image processing units PC1-PC6 do not interact with each other, but they transmit a message to the sorting control unit MPC after each image processing is completed.

[0052] According to one embodiment, each of the messages thus transmitted to the control unit MPC includes the following: - The moment the image was taken - The image coordinates (X, Y) of each marker detected in the image, and - An identifier corresponding to each detected marker.

[0053] The sensor assembly IVS includes a scanner that periodically scans the entire width of belt B1 (along the direction L1) in zone Z2, for example. The processing unit PCS connected to the sensor assembly IVS periodically supplies data in the form of collection zones divided into one or more rows (along the direction L1) of hundreds of adjacent rectangular cells. Thus, each collection row includes, for each cell, a number indicating the cell's position (along the width L1 of belt B1) (and possibly the number of the collection row), and a code indicating the detected attribute classification, such as material, color, etc. The resolution of the sensor assembly IVS may be coarser than that of the camera assemblies IS1-IS6, as the sides of the cells in the sensor assembly IVS can be several millimeters long. However, this resolution is sufficient to form a "chemical image" in which the color of each point can indicate a given object family. Furthermore, this resolution is not necessarily related to the number of solenoid valves EV of the sorting device BLS. The scanning period of the sensor assembly IVS is independent of the period of the cameras IS1-IS6.

[0054] Alternatively, the sensor assembly IVS could include a hyperspectral camera that visualizes the entire width of belt B1 without physical movement required to scan detection rows extending across the entire belt width, provided each row is collected at a fixed collection cycle that supplies data for each cell.

[0055] The processing units PC1-PC6, PCS, and control unit MPC can be synchronized with sub-millisecond precision using known protocols such as TCP / IP or UDP. Furthermore, since the workload of the control unit MPC is relatively low, it may be considered to perform the processing carried out by the processing units PCS and MPC in a single processing unit.

[0056] The image processing and marker detection algorithms performed within processing units PC1-PC6 require relatively high computational output due to the large number of images processed per second and the high resolution of each image. Even if these processes can be optimized by parallelizing calculations using a special processor (GPU type), the marker decoding time may still remain long depending on the image. Therefore, the detection algorithm can be configured not to decode all markers in an image, but rather to stop at the first marker found and decoded in the image, or at the first marker found in a predetermined sub-part of the image.

[0057] Therefore, regardless of the details of how the software for decoding the markers works, not all markers are always decoded. Figure 3 shows an image of belt B1, objects C1 and C2 detected by processing units PC1-PC6, marker M1 shown as a cross in association with sorting control toward exit S1, marker M2 shown as a circle in association with sorting control toward exit S0, and marker M3 shown as a diamond, which was not decoded. The undecoded marker M3 can be decoded during processing of subsequent images thanks to redundancy between consecutive images.

[0058] The execution time of the image processing algorithm is known to vary. The total time between the collection and provision of results by each processing unit PC1 and PC6 can vary, for example, between 20 and 300 ms. When the distance between cameras IS1-IS6 and nozzle bar NZ equipped with solenoid valve EV is set to 1-2 m, and the travel speed of belt B1 is 2-5 m / s, it may take too long for one image processing result to be supplied to the control unit MPC during use. If this situation occurs, the control unit PCS may issue a non-blocking error message.

[0059] The sorting control unit MPC combines data received from the image processing units PC1-PC6 and the processing unit PCS to control the sorting device BLS. Therefore, the control unit MPC receives and organizes data containing all useful information within the detection table, and combines this data to control the sorting device BLS. The control unit MPC performs image processing to improve sorting, derives sorting instructions from this, and sends them to the solenoid valve EV. The detection table is designed to be recorded in a file for subsequent analysis.

[0060] The sorting control unit MPC also includes a user interface accessible from the terminal TL, which has a set of sorting criteria configuration menus, including, for example: -Selection attribute menu - Sorting and adjustment menu for marked objects - Adjustment menu for object families, and -Optionally, a graphical display of the selection test in the shape of a map showing the detection table.

[0061] The coded information within a marker can be of two types, depending on the coding capacity of the marker. If the coding capacity of the marker is small (10 to 15 bits per marker), it can be considered to directly code a list of sorting and recycling attributes, that is, several characteristics that should be used to determine the sorting instruction. For example, the characteristic of whether or not it is for "food use" can be coded with 1 bit, the characteristic of "viscosity level (MF1)" can be coded with 2 bits for "very low," "low," "medium," and "high," and the material code (plastic resin) can be coded with 3 bits in the range of 1 to 7.

[0062] If the marker's coding capacity is sufficient, for example 42 bits or more like a barcode, it is advantageous to code a single identifier for the object, also known as an SKU ("Store Keeping Unit"). The object identifier used can be one that is standardized by a GTIN ("Global Trade Identification Number") and used by the international barcode system. Of course, other options are possible as long as the object is uniquely identified. Such an object identifier serves as an input key to access the object's complete technical specifications. These specifications are accessible via a remote information system, such as the internet, and specifically provide a list of sorting / recycling attributes.

[0063] When object OB passes over belt B1, only a few milliseconds are available between detecting the object and making a sorting decision. Therefore, the available time may be insufficient to access remote information and make a sorting decision.

[0064] According to one embodiment, the sorting system includes a local database SD connected to a sorting control unit MPC, which supplies sorting commands, for example, exit S0 / exit S1, for each identifier of marked objects to be sorted. The local database SD is also capable of storing a list of specific object attribute values ​​for each object identifier. The sorting commands are modifiable and determined according to sorting criteria stored by the sorting control unit MPC and selected by the user.

[0065] Each time a marker is decoded, the object identifier extracted from this marker is used to access the local database SD. If this identifier does not already exist in the local database, a query is made to the remote database to obtain a list of specific attributes of objects corresponding to existing identifiers within seconds or minutes. The local database SD is then updated with the obtained object identifier and specific attributes, and, in some cases, with sorting commands determined based on sorting criteria stored by the sorting control unit MPC. In this way, one or more initial objects associated with an identifier that does not exist in the local database SD are sorted (sent to exit S0), while subsequent objects associated with this identifier are made available for sorting.

[0066] If an object identifier extracted from a marker exists in the local database SD, the corresponding sorting command is read, and this command encounters a sorting command resulting from the analysis of the signal sent from the sensor assembly IVS. Based on these two sorting commands, the processing unit MPC, which controls the solenoid valve EV, makes a combined sorting decision. Therefore, access to the database SD is only required once for a given sorting recipe, solely to read the eject command. This configuration avoids unnecessary calculations or disk access.

[0067] According to one embodiment, object identifiers that have not been searched for a certain period of time when determining a first sorting command are cleared from the database SD. This configuration reduces the search operation of objects in the database based on object identifiers for retrieving associated sorting commands and updating sorting indicators stored in the database according to sorting criteria. For this purpose, a read count counter can be associated with each object identifier stored in the database. This counter can be incremented by one count for each object each time an object identifier is searched in the database to determine a first sorting command. This counter can be reset from time to time, for example, at regular intervals (for example, weekly). Objects associated with a read count counter that is zero before the reset are cleared from the database SD.

[0068] According to one embodiment, a menu for selecting object attributes is displayed on the terminal TL screen. An example of such a menu is shown in Table 1 below. [Table 1]

[0069] Each attribute value that an object can possess is predetermined. An example of these values ​​is shown in Table 2 below: [Table 2]

[0070] After verifying the selections made in Table 1, the "Marked Object" menu for the material selection determined by one marker is displayed. The columns in the "Marked Object" menu are selected from the sorting attribute selection menu (Table 1) and include an additional column, "Exit," allowing the sorting command to be identified according to the specific attributes in the other columns. The "Marked Object" menu is shown, for example, in Table 3 below: [Table 3] (1), (2), (3), (4), (5): The cells in Table 3 shown in this way contain all the possible options listed in Table 2 for the corresponding attribute value listed at the beginning of the corresponding column. Table 3 is displayed when the attributes "Color," "Shape," "Application," "Viscosity," "Biogenic Material %," and "Recycled Material %" are selected in the attribute selection table (Table 1).

[0071] After selection, it is possible to identify different sorting instructions for materials of the same material but with different values ​​for the attributes thus selected, by duplicating one row of the "Marking Products" menu in cells belonging to that row and one of the columns "Color," "Shape," "Application," "Viscosity," "Biogenic Material %," and "Recycled Material %" that have different values ​​from "All."

[0072] In one embodiment, when verifying the "Marking Products" menu, each of the first selection indicators identified in the "Exit" column is registered in the local database SD for each identifier of the marked object corresponding to the attribute value shown in the row corresponding to that selection indicator in the table. Therefore, there is no need to reread the selections made in the menu during selection, and only the local database SD is searched according to the object identifier obtained by decoding one marker in one image.

[0073] The "Object Family" menu for selecting materials and colors detected by the Sensor Assembly IVS includes a set of sorting criteria that enable the determination of sorting commands for each material shown in the "Material" row of Table 2, and, in some cases, for each color shown in the "Color" row of Table 2.

[0074] An example of the "Object Family" menu is shown in Table 4 below: [Table 4] If the "Color" attribute is not selected in the object attribute selection menu (Table 1), the "Detected Color" column in the "Object Family" menu will not be displayed.

[0075] The user is prompted to select one of the options shown in each cell of the "Detected Color," "Preferred Marker," and "Exit" columns in the "Object Family" menu. If "Transparent" or "Colored" is selected in one row, this row is duplicated, thereby allowing for the identification of different sorting instructions for materials of the same material but with different "Color" attribute values. In this way, it is possible to identify different sorting criteria for two object families composed of the same material, such as two PET families, "Transparent PET" and "Colored PET." If the "Preferred Marker" cell is selected (with "Yes"), the first instruction identified in the "Exit" column of the "Marked Object" menu confirms the sorting composite decision. The sorting instruction identified in the "Exit" cell of the "Object Family" menu confirms the sorting composite decision when no markers are detected and the "Preferred Marker" cell contains the value "None."

[0076] According to one embodiment, the detection table is configured to map the entire space located between the zone observed by cameras IS1-IS6 and the ejection line at the downstream end of belt B1. The detection table takes the form of a two-dimensional array composed of records, each record representing a single rectangular cell whose height (in the D1 direction) is related to a time increment that depends on the time it takes to collect one row of cells in the sensor assembly IVS. For example, if one row of cells is collected every 4 ms and the belt is moving at a speed of 3 m / s, the height of the cells in the two-dimensional array is 12 mm. However, the selection of the time increment is also dependent on the ejection technique. The width of each cell shown in the detection table is chosen so that the cell has a shape close to a square. This width may also be related to the space between the discharge nozzles NZ of the sorting device BLS. The positions of the objects to be sorted within the detection table are updated with each time increment according to the diagram of the ring buffer. The principle of the ring buffer is known to those skilled in the art and will not be repeated here.

[0077] The detection table accumulates and combines results from all cameras IS1-IS6 and all marker images by applying a logical OR operation to all results supplied from processing units PC1-PC6 and PCS, except in special cases where conflicting results appear within the same record. Significant redundancy between images, depending on the frame rate, may cause the same cell to be specified multiple times. Each record in the detection table can aggregate this redundancy. If the results are conflicting, for example, a decision can be made to invalidate the cell for any future actions, and this event can be tracked by recording a negative value for redundancy.

[0078] Considering the width of the image captured by cameras IS1-IS6 in direction D1, the image capture tempo, and the translational speed of belt B1, one point on the belt can be viewed multiple times. Such redundancy results in a significant computational load. Therefore, it is desirable to limit this redundancy without significantly reducing the detected area. Furthermore, if the cells corresponding to the recording in the detection table are spaced further apart than the markers, each cell may contain multiple markers, and therefore these markers may lead to different sorting controls. According to one embodiment, one of the outlets S0 and S1, for example, outlet S1, is provided to receive selected objects, while the other outlet S0 receives all unselected objects. In case of conflict, outlet S0 is preferred to avoid contaminating objects sent toward outlet S1.

[0079] According to one embodiment, each record in the detection table stores the following parameters: - A first sorting command (directed to exit S0 / S1) is determined based on the detected and decoded markers. -A second sorting command (directed to exit S0 / S1) is determined based on the detected attributes (material, color, etc.) obtained by processing the signal from the sensor assembly IVS. - Selection composite decision (towards exit S0 / S1), - Expansion rank, - Object identifier obtained by decoding the marker, - The values ​​of the detected attributes (material, color), and - Count value of the redundancy of detected markers

[0080] The detected attribute values ​​and corresponding second sorting commands supplied from the processing unit PCS are stored in the detection table record, taking into account the number of rows in the table corresponding to the distance between the observation zones of cameras IS1-IS6 and the observation zones of the sensor assembly IVS.

[0081] By combining data from image processing units PC1-PC6 with data from the signal processing unit PCS sent from the sensor assembly IVS, marking errors can be detected. In fact, the material detected by the sensor assembly IVS for one cell in the detection table may not correspond to the material derived from the decoded marker. This situation can occur, for example, when the same mold is used for multiple materials or when there is a printing error in the marker. In the event of a discrepancy, priority can be given to the sensor assembly IVS whose measurement is an actual value and not a declared value.

[0082] By combining data from image processing units PC1-PC6 and processing unit PCS, the determination of the ejection surface can be improved, and therefore, the determination of the solenoid valve EV to be controlled for sorting each object can be improved. In fact, in many cases, the detected markers do not completely cover the object. This is the case when the object is not marked over its entire surface, or when part of the label is peeled off. This is also the case when image processing is unable to detect all markers (computation time is too long, the marked surface is severely damaged or too dirty, the object is tilted, etc.). In this case, the surface of the object to be considered for ejection can be extended to cover the entire cell covered by the object detected by the sensor assembly IVS, based on the cell nucleus centered on the decoded marker. In this way, a positive extension is defined around the nucleus containing the marker that identifies the ejection of the object. Similarly, a negative extension is defined around the nucleus containing the marker that identifies the object not to be ejected. The solenoid valve EV is activated in the group of cells covered by the object when the extension is positive, and not activated in the group of cells covered by the object when the extension is negative.

[0083] By combining data from image processing units PC1-PC6 and data from processing unit PCS, unmarked objects can be similarly sorted. Even when marker technology is widely used, unmarked products are always mixed in with marked objects. By using the sensor assembly IVS, a combined sorting decision can be made for these objects. Furthermore, the flow of marked objects can be separated from contaminants consisting of unmarked objects.

[0084] By combining data from image processing units PC1-PC6 with data from processing unit PCS, redundancy and the required computational output can be reduced. In fact, operating at a rate of 300 images per second requires enormous computational power in the downstream processes. By decoding at least one marker for each object and entrusting the processing unit PCS with the management of complementing the object mapping, sorting efficiency can be increased even at low frame rates.

[0085] According to one embodiment, the control unit MPC is configured to determine first and second sorting commands for each cell based on data supplied in messages transmitted from processing units PC1-PC6 and PCS. For example, the first and second sorting commands can take the value 1 (sorting toward exit S1), the value -1 (sorting toward exit S0), or the value 0 if processing units PC1-PC6 and PCS did not supply any data for the cell. The control unit MPC is then configured to determine a combined sorting decision for each cell based on the combination of the first and second sorting indicators determined for that cell. For this purpose, the control unit MPC can use the following command fusion table: [Table 5] Table 5 shows, in code form, the possible values ​​in the sorting composite decision for a single cell, depending on the first sorting command (Command 1) and the second sorting command (Command 2), and the criteria selected in the "Object Family" menu (Table 4). Table 5 similarly distinguishes between cases where the material and / or color detected by the sensor assembly IVS for a single cell matches or does not match the material and / or color identified by the marker detected for this cell ("conforming" / "non-conforming"). Exit S0 is selected when there is no data that can lead to the selection of Exit S1 (no opening of one or more corresponding solenoid valves EV). The various codes shown in Table 5 are provided, in particular, to provide additional information about the sorting composite decision values ​​available for statistical purposes.

[0086] In the display of Table 5: Code "X" indicates that the material detected by the sensor assembly IVS is a material that should be sorted with "priority marker" "present", and since the first sorting command points to S1, the corresponding cell should be sorted toward exit S1; Code "Y" indicates that there is a discrepancy between the marking associated with the corresponding cell and the material or color detected by the sensor assembly IVS, but marker detection is given priority, and this discrepancy may be caused by a marking error, so the corresponding cell should be sorted toward exit S0; The code "O" indicates that the material detected by the sensor assembly IVS is sorted with a "priority marker" and therefore the sorting composite decision for the corresponding cell is determined primarily based on the first sorting command of the cell that directs to exit S0.

[0087] Code "A" indicates that the corresponding cell should be sorted toward exit S1 because detection by the sensor assembly IVS is given priority. Code "B" indicates that no markers were detected in the corresponding cell, but exit S1 was selected by the second selection command. The code "C" indicates that the corresponding cell should be sorted toward exit S0 because detection by the sensor assembly IVS is given priority. Code "D" indicates that no markers were detected in the corresponding cell, but exit S0 was selected by the second selection command. Code "E" indicates that the sensor assembly IVS detected nothing, but a marker was detected, and therefore the corresponding cell should be sorted toward exit S1. This situation can occur in the presence of a black object or when there is a failure to detect material or color.

[0088] Code "F" indicates that the sensor assembly IVS detected nothing, but a marker was detected, and therefore the corresponding cell should be sorted toward exit S0. This situation can occur in the presence of a black object or when there is a failure to detect material or color. Codes "M" and "N" are used when the first sorting command is given priority and no markers are detected in the corresponding cell. Therefore, code "N" should be used when the second command for the cell selects exit S1, and code "M" should be used when the second command for the cell selects exit S0. The code "●" likely indicates that nothing was detected in the corresponding cell, probably because there was no object in the cell, in which case exit S0 is selected by default.

[0089] In the first step, the sensor assembly IVS determines whether the material detected is a material to be sorted based on the presence or absence of a marker (priority marker) identified for each material detectable by the "Object Family" menu. If the material is not one to be sorted based on the marker, the first command has no effect, and in this case, the combined decision is equivalent to the decision of the second command. Therefore, if the second command is in S1 and at least one marker is detected (whatever the first command may be), the sorting combined decision receives code "A". If the second command is in S1 and no markers are detected, the sorting combined decision receives code "B".

[0090] If the second instruction is in S0 and at least one marker is detected (regardless of the first instruction), the sorting composite decision receives code "C". If the second instruction is in S0 and no markers are detected, the sorting composite decision receives code "D".

[0091] If materials should be sorted according to a marker, a conformity test is performed. This test determines whether the material detected by the sensor assembly IVS for the cell corresponds to the material indicated by the marker detected for the cell. If the detected material and the material identified by the marker do not match, there is a problem of material inconformity, and the sorting composite decision receives the code "Y". Conversely, if the material detected by the sensor assembly IVS matches the material identified by the marker, the cell is considered a conforming cell. In this case, the composite decision assigned to the cell corresponds to the first command. Therefore, if the first command is in S1, the sorting composite decision receives the code "X", and otherwise (if the first command is in S0), the sorting composite decision receives the code "O". If no marker is detected at all, the second command is applied because it is the only one available. That is, if the second command is in S1, the sorting composite decision receives the code "N", and if the second command is in S0, it receives the code "M".

[0092] If no material is detected by the sensor assembly IVS, the sorting composite decision corresponds to the first command. That is, if the first command is in S1, the sorting composite decision receives the code "E", and if the first command is in S0, it receives the code "F". If no material or marker is detected for a cell, there is no sorting information available for that cell, so the sorting composite decision receives the code "●".

[0093] Table 6 below shows an example of a sorting recipe, and the sorting criteria selected within the "Object Family" menu: [Table 6]

[0094] The first row of Table 6 (below the heading line) shows that transparent PET detected by the sensor assembly IVS is sorted according to the markers. The purpose is to separate food-grade transparent PET from non-food-grade transparent PET, which is achievable solely thanks to the attributes provided by the markers. Therefore, the sorting composite decision is determined by the markers for the "transparent PET" cells. In the first row, transparent PET is sorted according to the markers, but even if there are no markers, they can be presumed to be food-grade and thus oriented to exit S1. In the second row of Table 6, since this involves contaminants, colored PET can be oriented to exit S0 regardless of the presence or absence of markers. In the third row of Table 6, since the color under the shrink label cannot be detected by the sensor assembly IVS, PET with shrink labels can be sorted according to the markers. The attributes provided by the markers orient the object to exit S1 if it is food-grade transparent PET, and to exit S0 otherwise. If no markers are detected at all, exit S0 is identified to consider the object as a contaminant. The fourth row of Table 6 relates to contaminants, so regardless of the color and whether or not a marker is present, other plastics can be oriented towards outlet S0. The fifth row of Table 6 relates to contaminants, so regardless of the color and whether or not a marker is present, fibrous materials can be oriented towards outlet S0.

[0095] In the example of a sorting recipe partially identified in Table 6, the sorting criteria selected in the "Marked Objects" menu are shown in Table 7 below: [Table 7]

[0096] Based on the sorting criteria selected in this way, outlet S1 can be selected for objects made of clear food-grade PET in bottle shape and objects made of clear food-grade PET with shrink labels. All other combinations are considered contaminants, and their respective outlets S0 are identified.

[0097] Figure 5 shows an example of a detection table, specifically illustrating an example of the distribution of sorting composite decisions within cell CL. Figure 5 shows three objects separated to some extent by "blank" codes: the upper left object OB1 (code "X") sorted toward exit S1, the lower right object OB2 (code "0") which is in contact with the first object and sorted toward exit S0, and the upper right object OB3 (code "N") which is not adjacent to objects OB1 and OB2 and has no marker.

[0098] In object OB1, several cells receive code "X" because they have detected markers that match the material or color detected by the sensor assembly IVS for those cells. Other cells in the object receive code "M" because there are no markers in the corresponding cells, and in these cells, the material and / or color detected by the sensor assembly IVS is assigned to exit S0. In object OB3, a cell receives code "N" because there are no markers in the corresponding cell, and in this cell, the material and / or color detected by the sensor assembly IVS is assigned to exit S1. In object OB2, several cells receive code "O," for example, because this relates to non-food product packaging, and the sorting criteria specify that only food packaging should be sorted toward exit S0. Other cells in the object receive code "M" because there are no markers in the corresponding cells, and in these cells, the material and / or color detected by the sensor assembly IVS is assigned to exit S0.

[0099] In one embodiment, the control unit MPC executes an extended algorithm to extend cell selection to each of the exits S0 and S1, in addition to cells covered by objects, thereby improving the control accuracy of the sorting device BLS. This algorithm exclusively processes cells that receive the code "M" or "N," i.e., cells in which the detected material should be sorted according to a marker ("priority marker"), but no marker was detected at all. Furthermore, these extensions are performed only for cells adjacent to cells that receive the code "X" or "O," and only if the material detected by the sensor assembly IVS for cell "N" or "M" is the same as the material of the adjacent cell "X" or "O."

[0100] Two types of extensions are applied. In the so-called "positive" extension, the zone of the cell that received the selected code "X" in exit S1 is expanded by assigning a non-zero extension rank to cells adjacent to a cell that received the code "X" and that received the code "N" or "M". In the so-called "negative" extension, the zone of the cell that received the selected code "O" in exit S0 is expanded by assigning a non-zero extension rank to cells adjacent to a cell that received the code "O" and that received the code "N" or "M". Therefore, the extension rank assigned to a cell that received the code "X" or "O" is always 0. The extension rank assigned to a cell adjacent to a cell that received the code "X" or "O" is 1. The extension rank assigned to a cell adjacent to a cell that received rank n is n+1. The adjacency relationships applied here can be, for example, 4-connected (considering adjacent cells located above, below, to the left, and to the right of the cell under consideration) or 8-connected (considering all adjacent cells located diagonally to the cell under consideration), depending on the number of cells considered adjacent to a single cell located outside the boundary of the zone under consideration.

[0101] According to one embodiment, multiple expansion iterations are performed by scanning the detection table with each iteration. With each iteration, one expansion rank is added around the already expanded zones, and a cell of the new rank is added. Each time an expansion rank is assigned to a cell, the detection table record corresponding to this cell can also receive the marker identifiers of the cells of lower expansion ranks adjacent to that cell.

[0102] As shown in Figure 4, when two objects of the same material are adjacent, there is a risk that the extended rank assigned by consecutive iterations will extend from the marked object into the adjacent unmarked object. In fact, Figure 4 shows images taken by cameras IS1-IS6, including a marked object C3 and an unmarked object C4. When an extended rank is assigned to the marked object, there is a risk that the area occupied by the marked object will extend to, for example, the dashed line extending into the unmarked object C4. As a result, cells covered by the unmarked object C4 will be controlled for sorting as cells covered by the marked object C3. Therefore, there is a risk that the unmarked object (C4) will be mistakenly sent to the same exit as the marked object (C3).

[0103] This risk can be controlled by the concept of numbering the expansion rank, for example, by defining the maximum expansion rank in one of the menus displayed by the terminal TL. This maximum expansion rank can be different for positive and negative expansion. It can also be specific to each material detected or identified by each identifier of the marker or marking product. In this case, the maximum expansion rank can be specified by adding a column to the "Object Family" menu and / or the "Marking Object" menu.

[0104] Therefore, each maximum expansion rank determines the number of iterations to be performed for the corresponding expansion type or corresponding product type. One of the menus displayed by the terminal TL allows you to specify whether to prioritize exit S0 or S1. Thus, if exit S0 is prioritized, negative expansion is performed before positive expansion in each iteration. If the cell identifier field is already full, it can be configured not to assign a non-zero expansion rank to a cell. Thus, cells are ensured to be assigned to the closest object of the same family in the image with a single marker.

[0105] Figure 6 shows the contents of the detection table after the first iteration of the augmentation algorithm has been performed on the table shown in Figure 5. In the representation in Figure 6, cells that receive the code "N" or "M" and an augmentation rank of 1 are indicated by the code "#" if they are adjacent to a cell that received the code "X" (positive augmentation), and by the code "e" if they are adjacent to a cell that received the code "O" (negative augmentation).

[0106] Figure 7 shows the contents of the detection table after performing a second iteration of the augmentation algorithm on the table shown in Figure 6. In the representation in Figure 7, cells that receive the code "N" or "M" and an augmentation rank of 2 are marked with the code "#2" if they are adjacent to a cell that received the code "#", and with the code "e2" if they are adjacent to a cell that received the code "e".

[0107] Figures 8 and 9 show the contents of the detection table after performing the third and fourth iterations of the augmentation algorithm on the table shown in Figure 7, where the maximum rank is set to 2 for positive augmentation and at least 4 for negative augmentation. In the representation in Figure 8, cells that receive the code "N" or "M" and augmentation rank 3 are labeled with the code "e3" if they are adjacent to cells that receive the code "e2". In the representation in Figure 9, cells that receive the code "N" or "M" and augmentation rank 4 are labeled with the code "e4" if they are adjacent to cells that receive the code "e3".

[0108] After all the iterations required by the expansion algorithm have been completed, the remaining cells to which the code "N" or "M" can be assigned belong to the zones of objects that do not contain markers, such as object OB3 in the upper right of Figures 5-9, and the zones of marked objects that the expansion process did not reach. In fact, cells to which the code "N" or "M" can be assigned that are placed beyond the maximum expansion rank for the nucleus "X" or "O" do not receive a non-zero expansion rank. Such a zone appears in the lower left of object OB1. However, most of the area of ​​object OB1 has already been selected for exit S1. Therefore, the risk of error is low.

[0109] The selection of the maximum expansion rank serves multiple purposes. First, positive expansion efficiently selects objects to be oriented towards exit S1. Similarly, it avoids overflow onto adjacent unmarked objects and prevents objects from being mistakenly oriented towards the wrong exit. Negative expansion prevents objects that should be sent to exit S0 from being inadvertently oriented towards exit S1, although these objects are not completely covered by the expansion.

[0110] According to one embodiment, the control unit MPC is configured to sequentially read each row of the detection table (extending along direction L1), and in the cells of the read row, a selection composite decision determined to be "X", "N", "A", "B", "E", and an extended rank "# <n>The corresponding object is sent to outlet S1 by opening the solenoid valve EV corresponding to the cell associated with (n=""2,3,..., maximum positive extension rank). A sorting composite decision is made in the cell of the read row using "Y", "O", "M", "C", "D", "F", and "●", and the extension rank "e <n>The solenoid valve EV corresponding to the cell associated with (n=""2,3,..., negative maximum expansion rank) remains closed. Cells in which solenoid valve EVs are opened one row at a time (extending along direction L1) are shown in gray in Figure 9. Figure 10 shows the sequence of control lines for solenoid valve EVs obtained based on the expansion table in Figure 9, substituting the sorting composite decision code for the corresponding outlet S1 or S0, with outlet S0 selected by default. The control table in Figure 10 indicates for each cell whether outlet S1 is selected, and empty cells in the control table indicate that outlet S0 is selected. When the number of cells in the width direction of the table in Figure 10 is equal to the number of solenoid valve EVs in the sorting device BLS, each cell in this table determines the control of one solenoid valve.

[0111] According to one embodiment, the control unit MPC is configured to archive rows of the detection table (extending along direction L1) after reading control of the solenoid valve EV. Each row thus archived is made indexable by the time elapsed since the start of one sequence of sorting after the determination of the sorting criteria. To limit the amount of data thus archived, it is possible to decide not to record rows of void cells.

[0112] This archiving allows for empirically-based analysis of the sorting system's performance. Archiving rows from the detection table can be used to determine the area occupied by sorted objects on belt B1 by calculating the number of cells assigned to each marker identifier. This process allows for the estimation of the number of different objects processed. This estimation can be based on the average detection area recorded by the object identifier (or SKU). This estimation can also be performed by concatenation analysis between cells corresponding to the same marker identifier. The more detailed the marker identifier, the more reliable this estimation becomes. However, the presence of clusters of objects touching the same marker identifier can lead to numerous errors, including the risk of them being interpreted as a single object. To limit the impact of these errors, a maximum area per object identifier can be determined, thereby determining whether the error pertains to a cluster or a single object.

[0113] Furthermore, discontinuous marking of objects (discontinuities in marker detection and discontinuities in detection data obtained from the sensor assembly IVS) may occur that are not corrected by the extended image processing. In such cases, the same object appears as an island of multiple cells. When dealing with the same object, conventional "closing" image processing algorithms can be used to group the islands together.

[0114] To those skilled in the art, it will be apparent that the present invention is subject to various modified embodiments and diverse applications. In particular, the present invention is not limited to sorting systems having two outlets, but is equally applicable to systems having three or more outlets. Accordingly, Figure 11 includes a sorting system that differs from the sorting system of Figure 1 in that it has three outlets S0, S1, and S2, and a second sorting device BL1 located at the outlet end of a conveyor BCV that is operable to orient the objects to be sorted toward outlet S2. When the two sorting devices BLS, BL1 are not in operation, the objects to be sorted are oriented toward outlet S0. In this case, the sorting criterion specification menu (Tables 3 and 4) and fusion table (Table 5) are adapted by introducing a third alternative value (S2) in the “Outlet” column.

[0115] Similarly, it should be noted that the control of the sorting device is arbitrary, insofar as this method can be limited to creating a list of objects placed on the conveyor. In fact, the sorting composite decision identifies a set of parameters for each detected object, namely, a set of parameters relating in particular to the object's material, color, shape, and use.

[0116] Furthermore, other rules can be applied to the combination of first and second sorting commands for making a sorting composite decision, as needed. In particular, different priorities can be utilized for sorting commands without departing from the scope of the present invention.

[0117] Similarly, the execution of the extended algorithm is optional, given that this algorithm merely expands the zones that follow the selection control and thus reduces selection errors. Furthermore, the contours of objects present in belt B1 can be detected by performing known image processing, for example, based on one or more artificial intelligence models or neural networks. The extended algorithm can also be implemented further upstream on images from cameras IS1-IS6. Artificial intelligence software can also be used to recognize other aspects of objects (shape, texture, color, etc.).

[0118] The order of the two sensor assemblies described above can also be reversed. In fact, the system can function properly even if the sensor IVS is positioned upstream in the direction of movement of belt B1 and in front of the cameras IS1-IS6 which are positioned downstream.

[0119] Given that the determination of the first selection command can be carried out without a local database, the implementation of the local database SD is equally optional. By connecting to a remote database, specific attributes (associated with the detected marker identifiers) are determined, and selection criteria are applied to these attributes.

[0120] It should also be noted that, entirely separate from the object sorting method, the detection and decoding of markers using only two alternative illuminations located at the ends of the visible spectrum constitutes part of the present invention. Accordingly, the description of the present invention similarly covers a method for reading markers on a single object, the method comprising the steps of: alternatingly illuminating the object under two colors located at the ends of the visible spectrum; collecting an image of the object when it is illuminated under each of the two illuminations; detecting one or more markers on the object in the image in which the object is illuminated under one of the two illuminations; and decoding each detected marker.

[0121] Image acquisition can be performed using a single image sensor that detects two colors of light, or using two sensors that detect two different colors of light.< / n> < / n>

Claims

1. A method for sorting objects, comprising the following steps: Steps include: placing the objects to be sorted onto the belt (B1) of the belt conveyor (BCV); A step of collecting an image flow of objects present in a first belt zone (Z1) that extends across the entire width of the belt; Steps include collecting material and / or color detection signals for each cell of an adjacent cell group (CL) located in a second zone (Z2) that extends across the entire width of the belt; A step of processing the detection signal for each cell and determining the attributes of the detected material and / or color; Steps include detecting markers in the image flow of the first zone; For each detected marker, the steps are to decode the marker and determine the object identifier; For each decoded marker, the step of assigning the object identifier determined by the decoding of the marker to the cell covered by the decoded marker; A step of determining specific attributes for each object identifier determined based on the decoded marker; For each cell associated with an object identifier, the first sorting command is determined according to the sorting criteria and specific attributes; For each cell, a second sorting command is determined according to the sorting criteria and the detected attributes; A step in which a selection composite decision is determined for each cell, according to the first and second selection commands and combination rule set for that cell.

2. The method according to claim 1, comprising several steps of extending the selection composite decision of cells associated with an object identifier to adjacent cells not associated with an object identifier by applying an extended set of rules.

3. The method according to claim 1 or 2, comprising the step of controlling a sorting device (BLS, BL1) to orient each object present on the belt toward a selected exit (S0, S1, S2) in accordance with a sorting composite determination of cells facing the sorting device.

4. A method according to any one of claims 1 to 3, which is carried out according to the set of combination rules, The selective composite decision for one cell (CL) is: If the first and second sorting instructions for the cell are the same, then conform to either of these instructions. If the first and second sorting commands for the aforementioned cell are inconsistent, assign it to one of the sorting system exits (S0, S1, S2) determined according to the sorting criteria. If the second sorting command is uncertain because there is no detection indicating the material and / or color within the cell, the first sorting command for the cell shall be followed. If the second sorting instruction for a cell has been determined, and it is clear that the second sorting instruction takes precedence based on the sorting criteria that take into account the detected attributes, then the second sorting instruction for the cell shall be followed, and The method, wherein if the first sorting command is uncertain because no marker is detected in the cell, the sorting command is determined according to the sorting criteria and the combined sorting determination of adjacent cells.

5. The first extension operation includes the following steps: A cell selection step comprising selecting at least one cell (CL) of a first extended rank that is not associated with one object identifier, wherein the selection criterion assigns priority to a first command for the cell, the cell is adjacent to the core of at least one cell associated with one object identifier, the selection criterion assigns priority to a first command for this adjacent cell, and for the adjacent cell and the selected cell, the material identified by the marker corresponds to the material detected by the sensor assembly (IVS); and The step of assigning a value equal to the value of one of the cells in the nucleus to the selection composite decision for each selected cell. The method according to any one of claims 1 to 4, including the method described in any one of claims 1 to 4.

6. The subsequent extension operation includes the following steps, namely: A cell selection step comprising selecting at least one cell (CL) belonging to a higher extended rank that is not associated with one object identifier, wherein the selection criteria assign priority to the first command for the cell, the cell is adjacent to at least one cell belonging to a lower extended rank, and the material detected by the sensor assembly (IVS) corresponds to the material of the selected cell; and The step of assigning a value equal to the value of a cell belonging to a lower expansion rank to the selection composite decision of each cell selected for a higher expansion rank. The method according to claim 5, including the method described in claim 5.

7. The method according to claim 6, wherein multiple extension operations are performed for each outlet (S0, S1, S2) of the sorting system up to a defined maximum number of extension operations.

8. The method according to claim 7, wherein the maximum number of extension operations is defined for each identifier or each value of a single identified attribute.

9. For each cell, the following: - First Selection Order - Second Selection Order - Selection and combined determination - Extended rank, and - Object identifier obtained by decoding the marker The method according to any one of claims 5 to 8, comprising constructing a detection table including

10. The method according to any one of claims 1 to 9, wherein the determination of a specific attribute based on each object identifier determined using a decoded marker includes a step of searching for object identifiers in a local database (SD) storing object identifiers that can be extracted from the marker and specific attributes associated with each object identifier.

11. The steps consist of receiving selection criteria; and The process includes updating a database (SD) that stores object identifiers extractable from markers and specific attributes associated with each object identifier, wherein the database update includes determining a selection command for each object identifier according to a selection criterion and specific attributes associated with the object identifier, and a first selection command for each cell (CL) associated with a single object identifier is a selection command associated with the object identifier in the database. The method according to any one of claims 1 to 10.

12. The selection criteria are the following attributes that are identified or detected: The properties, color, shape, and use of the material, the viscosity of the material, the proportion of biogenic material in the object, and the proportion of recycled material in the object. A method according to any one of claims 1 to 11, which targets one or more of the following attributes.

13. The method according to any one of claims 1 to 12, wherein the image flow is collected by alternately illuminating a first zone of a belt (B1) under two colors of illumination, so that each point of an object moving through the first zone is illuminated at least once with each color of illumination, the two colors of illumination are located at opposite ends of the visible spectrum, respectively.

14. Images from the image flow are acquired alternately under two different illumination colors, so that each cell (CL) is continuously illuminated under the two illumination colors, where one of the two illumination colors has a wavelength less than 460 nm and the other illumination color has a wavelength greater than 630 nm. The method according to any one of claims 1 to 13.

15. It is a sorting system, A belt conveyor (BCV) including a belt on which the objects to be sorted (OB1, OB2, OB3) are placed. Camera assemblies (IS1-IS6) for collecting image flow of objects present in a first zone (Z1) of a belt that extends across the entire width of the belt. A sensor assembly (IVS) for collecting material and / or color detection signals for each cell of an adjacent cell group (CL) located in a second zone (Z2) that extends across the entire width of the belt. Image processing units (PC1-PC6) for detecting markers within an image flow. A signal processing unit (PCS) for identifying attributes that are detected and determined based on the material and / or color detection signals. A sorting device (BLS, BL1) for orienting each object present on the belt toward the outlet of the sorting system (S0, S1, S2), and A sorting control unit (MPC) configured to carry out the method described in any one of claims 1 to 14. A system that includes this.

16. The sorting system according to claim 15, comprising a lighting system having two monochromatic light source assemblies (VL1, VL2) that alternately illuminate a first zone (Z1).

Images