Method for characterising items of waste present in a parking or drop-off zone, and associated system and computer program

Abstract

The present invention relates to a method for characterising items of waste present in a parking or drop-off zone, comprising the following steps: - providing (110) an image of the zone; - determining (120) in the image a plurality of recyclable items of waste; - for each recyclable item of waste, determining (130) an area of the recyclable item of waste occupied on the surface of the zone and associating a density with the item of waste; - determining (140) a generalised density of non-recyclable material; - determining (150) a generalised area occupied by non-recyclable material; - determining (160) a proportion of recyclable items of waste in the zone from the areas and densities of the recyclable items of waste and from the generalised area and generalised density of non-recyclable material.

Description

[0001] TITLE: Method for characterizing waste present in a parking or storage area, and associated computer program and system

[0002] The present invention relates to a method for characterizing waste present in a parking or storage area.

[0003] The present invention also relates to a computer program and a system associated with such a characterization method.

[0004] The parking or storage area advantageously features a skip, a pile, or any other location that can be observed, for example, from above. More specifically, the invention allows for the characterization of the percentage of recyclable waste present in such a location.

[0005] As is well known, in current industrial and commercial waste sorting centers, trucks or other waste collection vehicles are directed by an operator upon entry based on information provided by the customer regarding the recyclable waste content. Often, this information is incomplete, leading to numerous routing errors. This negatively impacts the performance of these sites.

[0006] One solution to this problem would be to assess the rate of recyclable waste on an open truck bed at the site entrance in order to orient the truck bed according to the amount of recyclable waste it contains or on a pile of waste on the ground at the unloading point to confirm that the orientation has been effective.

[0007] However, this is rarely done in sorting centers due to the complexity and lack of precision of existing evaluation methods.

[0008] In particular, when an assessment of the recyclable waste rate is implemented on an open skip, this is usually done on the basis of a camera that films the top of the truck and it is a human who then determines whether the recyclable waste rate is high or low.

[0009] On a ground pile, there are two types of characterization methods: sorting and weighing of the different fractions, and visual characterization. However, these methods are often carried out in hazardous situations involving multiple activities, and their reliability depends on the operator present on site that day. They are also not systematic, as they depend on the personnel available on site.

[0010] The present invention aims to remedy these drawbacks of the prior art and to propose means allowing a simple and precise evaluation of a rate of recyclable waste present in a parking or storage area (open skip, pile on the ground, etc.) and this without endangering the operators of the sorting center.

[0011] To this end, the invention relates to a method for characterizing waste present in a parking or storage area, comprising the following steps:

[0012] - provision of an image of the parking or drop-off area;

[0013] - determination on the image of a set of waste and determination within this set of waste, of a plurality of recyclable waste;

[0014] - for each recyclable waste item, determination of an area of ​​this recyclable waste item occupied on the surface of the parking or storage area and association of a density to this recyclable waste item;

[0015] - determination of a generalized density of the non-recyclable material; - determination of a generalized area occupied by the non-recyclable material;

[0016] - determination of a rate of recyclable waste in the parking or drop-off area from the areas and densities of recyclable waste and from the generalized area and generalized densities of non-recyclable material.

[0017] With these features, the invention enables a simple and accurate assessment of the percentage of recyclable waste present in a parking or storage area. This is achieved using, in particular, a 2D image of the area without human intervention in the process. Recyclable waste is identified in the 2D image, and its overall percentage in the area is then assessed by making a number of simplifications and assumptions about this waste and the non-recyclable material.

[0018] In particular, it was discovered by the inventors that the use of only the densities and areas occupied by recyclable waste in combination with a generalized density and a generalized area of ​​non-recyclable material, is sufficient to determine the rate of recyclable waste with good accuracy.

[0019] The generalized density of non-recyclable material can correspond to an average density of this material, determined for example prior to the implementation of the process.

[0020] The generalized area of ​​non-recyclable material can correspond to an actual area of ​​this material on the surface of the 2D image or to an extended area taking into account intersections of the areas of recyclable waste.

[0021] The generalized density and / or generalized area can be determined / calibrated by a learning technique.

[0022] The inventors also discovered that, across all calculations, the depths of recyclable and non-recyclable waste can be considered similar. This can significantly simplify the calculations while maintaining good accuracy of the results.

[0023] According to other advantageous aspects of the invention, the method comprises one or more of the following features, taken individually or in all technically possible combinations:

[0024] - the determination of the recyclable waste rate includes determining a sum of the products of the areas of recyclable waste with the corresponding densities;

[0025] - the density of each recyclable waste is defined for each predetermined category of recyclable waste;

[0026] - Recyclable waste and the corresponding occupied areas are determined by an object recognition technique applied to the image;

[0027] - the generalized density of non-recyclable material is determined by a learning technique implemented on a database of images of parking or storage areas with known rates of recyclable waste;

[0028] - the generalized area of ​​non-recyclable material is determined from a total area of ​​the parking or storage area and the areas intersected between different recyclable wastes;

[0029] - the generalized area of ​​non-recyclable material is further determined from at least two adjustment coefficients;

[0030] - said adjustment coefficients are determined by a learning technique implemented on a basis of images of parking or storage areas with known rates of recyclable waste;

[0031] - said learning technique is implemented by using a linear regression of an expression linking a known recyclable waste rate, the densities of recyclable waste, the areas of recyclable waste, the generalized area of ​​non-recyclable material, the generalized density of non-recyclable material and a plurality of depth ratios of each recyclable waste and non-recyclable material;

[0032] - said depth ratios are considered to be approximately equal to 1.

[0033] The invention also relates to a waste characterization system, comprising technical means configured to implement the process as defined above.

[0034] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement the process as defined above. The invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the drawings in which:

[0035] - [Fig. 1] Figure 1 is a schematic view of a waste characterization installation in the case of an open skip;

[0036] - [Fig. 2] Figure 2 is a schematic view of a characterization system according to the invention, the characterization system being part of the characterization installation of Figure 1; and

[0037] - [Fig. 3] Figure 3 is a flowchart of a characterization process according to the invention, the characterization process being implemented by the characterization system of Figure 2.

[0038] Figure 1 illustrates a waste characterization installation 10 located at least partially, for example, in a waste sorting center.

[0039] With reference to this figure 1, the characterization installation 10 includes imaging means 14 and a characterization system 16.

[0040] Image capture means 14 include one or more camera(s) positioned above a parking or storage area 18 to enable image capture of this area 18.

[0041] In the example shown in Figure 1, the parking or storage area 18 shows a parking area for a skip mounted, for example, on a truck, for example, at the entrance to the sorting center. Alternatively, the skip is placed on the ground or on any other support provided for this purpose. As a further alternative, the parking or storage area 18 shows a storage area for a pile of waste formed, for example, following the unloading of such waste.

[0042] Advantageously, the image capture means 14 are arranged so as to make it possible to capture images from above of the parking or storage area 18.

[0043] Image capture means 14 allow the images taken to be transmitted to the characterization system 16 via, for example, a computer network 19.

[0044] The computer network 19 can comprise any known network. Thus, this network 19 can form a local network connecting the image acquisition means 14 to the characterization system 16 locally, or a global computer network (such as the internet, for example) which allows the image acquisition means 14 to be connected to the characterization system 16 located remotely. Alternatively, the image acquisition means 14 can be connected to the characterization system 16 by any other technically feasible means, such as cables.

[0045] The characterization system 16 is illustrated in more detail in Figure 2.

[0046] Thus, with reference to this figure 2, the characterization system 16 comprises an input module 21, a processing module 22, a detection module 24 and an output module 23.

[0047] Each of these modules 21 to 24 is implemented, for example, at least partially in the form of software.

[0048] In such a case, the characterization system 16 further includes a memory for storing such software and a processor for executing this software.

[0049] Alternatively or in addition, at least one of these modules 21 to 24 presents at least partially a programmable logic circuit such as an FPGA (Field-Programmable Gate Array).

[0050] The characterization system 16 thus forms, for example, a server consisting of one or more computers. This characterization system 16 can therefore be located, for example, locally in the corresponding sorting center or remotely from it.

[0051] The input module 21 allows external data to be received and transmitted to the processing module 22 and the detection module 24.

[0052] In particular, the input module 21 is connected via the computer network 19 to the image capture means 14 to receive the images taken by these means 14. The input module 21 is further connected to a database 25 providing an object recognition model and operating parameters of the characterization system 16, as will be explained in more detail later.

[0053] The processing module 22 allows the data received by the input module 21 to be processed in order to determine a rate of recyclable waste in the parking or drop-off area 18.

[0054] Recyclable waste refers to any waste that is worthwhile to recycle. Each recyclable waste item can, for example, be classified into a predetermined category of recyclable waste, such as cardboard, wood, and plastics. Each category may, for example, be defined by current legislation.

[0055] The detection module 24 enables the detection of objects, including recyclable waste, in images provided by the image acquisition means 14, using the object recognition model, as will be explained in more detail later. The output module 23 transmits the recyclable waste rate determined by the processing module 22 to any interested external system.

[0056] Thus, for example, the processing module 23 can transmit this information in the form of a report 30 delivered to the waste provider or the sorting center operator.

[0057] The characterization system 16 allows for the implementation of a waste characterization process in the parking or storage area 18, which will now be explained with reference to Figure 3, which presents an organizational chart of its steps.

[0058] During an initial step 110 implemented for example at the entry of a new truck into the sorting center, the image-taking means 12 take at least one image from above of the parking or storage area 18 and transmit this image to the characterization system 16.

[0059] The input module 21 of the characterization system 16 receives this image and transmits it to the detection module 24.

[0060] Alternatively, the input module 21 of the characterization system 16 receives several images of the parking or storage area 18 and transmits all of these images to the detection module 24.

[0061] In the following step 120, the detection module 24 implements the object recognition model to identify a set of waste items in the received image and to determine which items within that set are recyclable. Alternatively, the detection module 24 implements this model on several images of the parking or storage area 18. During this step, the model can, in particular, delineate the waste item from any other object / environment present in the image, such as the ground or the skip. Furthermore, two separate models can be used to determine, on the one hand, the waste item and, on the other hand, the recyclable items within that waste item.

[0062] As explained previously, the object recognition model, for example, is derived from database 25 and can detect objects of different types, including recyclable waste, using a supervised or unsupervised learning algorithm. This learning algorithm has been previously trained to recognize these objects.

[0063] In particular, the learning algorithm was previously trained to recognize recyclable waste, for example, in 2D images. In some cases, the learning algorithm was also previously trained to categorize each recyclable waste item into one of the predetermined recyclable waste categories.

[0064] Of course, the use of other object recognition algorithms is also possible. It is considered at the end of this step 120 that each recyclable waste can be identified by an index i varying from 1 to k.

[0065] In the next step 130, for each recyclable waste item i, the detection module 24 determines an area A t occupied by this waste on the surface of the parking or storage area 18. This surface may correspond to that of one or more images transmitted by the image-taking means 14. Alternatively, this surface corresponds to that of one or more modified images of the parking or storage area 18. This modification may include, for example, a correction of the corresponding images, intended to correct distortions introduced by the image-taking means 14 and / or by the angle of view-taking.

[0066] To determine an area A tFor the corresponding recyclable waste, the detection module 24 delimits the waste with a contour, thus forming a bounding box. The area of ​​such a box can be determined, for example, by dividing it into triangles or rectangles. Any other method for determining the area of ​​a shape can be used instead. For example, using one possible technique, the detection module 24 can determine a surface that optimizes the shape of the waste (for example, by pixel segmentation) and then calculate the area of ​​this optimized surface.

[0067] Furthermore, during the same step 130, the detection module 24 associates each recyclable waste item i with a density p t This density p L is associated with the corresponding waste based, for example, on the category of that waste as determined in the previous step.

[0068] The values ​​of the densities pt for predetermined waste categories are for example taken from database 25.

[0069] At the end of this step 130, the values ​​A t and p t determined during this step are, for example, transmitted to processing module 22.

[0070] In the following step 140, the processing module 22 determines a generalized density p x non-recyclable material.

[0071] Non-recyclable material means any waste that has not been recognized as recyclable waste during step 120.

[0072] Advantageously, the generalized density p x is considered as a parameter of the process which remains constant, for example, during the implementation of the process.

[0073] In some implementations of the process, this generalized density p xcan, for example, be considered as the average density of all non-recyclable material. The value of this average density can, for example, be approximately equal to 140 kg / m³ 3 Advantageously, the generalized density p x is determined prior to the implementation of the process and is for example stored in database 25.

[0074] To determine the generalized density p x For example, a machine learning technique implemented on a database of images of parking or waste disposal areas with known recyclable waste rates can be used. In some embodiments, this machine learning technique is implemented by the processing module 22 during a preliminary step 105 performed before step 110.

[0075] During this preliminary step 105, the processing module 22 can, for example, determine the generalized density px using linear regression with one or more training images having known recyclable waste rates. This linear regression can be defined by the following expression:

[0076]

[0077] Or

[0078] f is the known rate of recyclable waste in the parking or drop-off area corresponding to the training image;

[0079] Ht, p t , t are respectively the depth, density and area of ​​a recyclable waste i on the training image;

[0080] ~Â t is the visible surface occupied by waste in the parking or drop-off area of ​​the learning image;

[0081] H x , p are respectively the depth and generalized density of non-recyclable material on the training image.

[0082] Thus, a learning technique using the aforementioned linear regression can be implemented by considering the coefficients

[0083]

[0084] such as unknown values.

[0085] PÆ

[0086] Alternatively, the values ​​== can be considered close to 1. In other words, the depths of recyclable waste can be considered substantially close to that of non-recyclable material.

[0087] In such a case, only the value p is considered unknown.

[0088] At the end of this preliminary step 105, the generalized density p x is defined by this value p^ at least for an implementation of the following steps of the process.

[0089] During step 150, the processing module determines a generalized area

[0090]

[0091] of non-recyclable material. In particular, according to one embodiment, this generalized area is determined by subtracting the areas A t recyclable waste from a total area A t occupied by waste in parking or storage area 18.

[0092] This total area A t is provided for example by a segmentation algorithm, trained in an unsupervised way (off-the-shelf segmentation called remove_backg round) from one or more images of the parking or storage area 18.

[0093] In other words, according to this implementation method:

[0094]

[0095] According to another embodiment, the generalized area A x It takes into account intersections of areas containing different recyclable waste, which are due in particular to imperfect detection of this recyclable waste during step 120. These intersections are denoted by At n Aj where the indices i and j are different, and vary from 1 to k.

[0096] According to this embodiment, the generalized area

[0097]

[0098] is determined by the following expression:

[0099]

[0100] where a and p are adjustment coefficients. In such a case, the generalized area of ​​the surface occupied by waste has an area extended by the intersections of the areas of recyclable waste. The influence of these intersections on the total area is defined by the adjustment coefficients a and p.

[0101] Just like the generalized density p x The adjustment coefficients a and p can be considered as process parameters which remain constant, for example, during the implementation of the process.

[0102] In some implementations of the process, the adjustment coefficients a and are determined prior to the implementation of the process and are, for example, stored in database 25.

[0103] To determine these adjustment coefficients a and p, a machine learning technique implemented on a database of images of parking or waste disposal areas with known recyclable waste rates can be used, for example. In some embodiments, this machine learning technique is implemented by the processing module 22 during the preliminary step 105 as described above.

[0104] Unlike what was described previously, in this preliminary step 105, linear regression can be defined by the following expression:

[0105]

[0106] In this expression, the values ​​p^, â and ft are considered unknown and are determined at the end of this step 105 to initialize the p values x , a and / ? repsectives.

[0107] In the following step 160, the processing module 22 determines the rate r of recyclable waste using the areas A t and the densities p t recyclable waste as well as the generalized density p x and the generalized area

[0108]

[0109] of non-recyclable material. This rate r can be determined by the following expression:

[0110]

[0111] In particular, determining this rate involves calculating a sum of the products of the values ​​p t And

[0112]

[0113] , and the calculation of a product of the p values x etÆ^(a, ?).

[0114] This rate of recyclable waste is then transmitted by output module 23 to any interested external system.

[0115] The process can then be repeated in relation to another parking or storage area 18.

Claims

DEMANDS 1. A method for characterizing waste present in a parking or storage area (18), comprising the following steps: - provision (110) of an image of the parking or storage area (18); - determination (120) on the image of a set of waste and determination in this set of waste, of a plurality of recyclable waste; - for each recyclable waste, determination (130) of an area of ​​this recyclable waste occupied on the surface of the parking or storage area (18) and association of a density to this recyclable waste; - determination (140) of a generalized density of non-recyclable material; - determination (150) of a generalized area occupied by the non-recyclable material; - determination (160) of a rate of recyclable waste in the parking or storage area (18) from the areas and densities of the recyclable waste and from the generalized area and generalized density of the non-recyclable material, said determination (160) of a rate of recyclable waste comprising the determination of a sum of the products of the areas of the recyclable waste with the corresponding densities.

2. A method according to any one of the preceding claims, wherein the density of each recyclable waste is defined for each predetermined category of recyclable waste.

3. A method according to any one of the preceding claims, wherein recyclable waste and corresponding occupied areas are determined by an object recognition technique applied to the image.

4. A method according to any one of the preceding claims, wherein the generalized density of non-recyclable material is determined by a learning technique implemented on a basis of images of parking or storage areas having known rates of recyclable waste.

5. A method according to any one of the preceding claims, wherein the generalized area of ​​the non-recyclable material is determined from a total area of ​​the parking or storage area (18) and the areas intersected between different recyclable wastes.

6. A method according to claim 5, wherein the generalized area of ​​the non-recyclable material is further determined from at least two adjustment coefficients.

7. Method according to claim 6, wherein said adjustment coefficients are determined by a learning technique implemented on a basis of images of parking or storage areas having known rates of recyclable waste.

8. A method according to claim 6 or 4, wherein said learning technique is implemented using a linear regression of an expression linking a known recyclable waste rate, the densities of recyclable waste, the areas of recyclable waste, the generalized area of ​​non-recyclable material, the generalized density of non-recyclable material and a plurality of depth ratios of each recyclable waste and non-recyclable material.

9. Method according to claim 8, wherein said depth ratios are considered substantially equal to 1.

10. Computer program comprising software instructions which, when executed by a computer, implement the method according to any one of the preceding claims.

11. Waste characterization system (20), comprising technical means (21, 22, 23, 24) configured to implement the process according to any one of claims 1 to 9.

Images