Edge computing-based single-item separation methods and devices, sorting methods and systems

Abstract

The present invention discloses a single-piece separation method and device based on edge computing, a sorting method and system, wherein the single-piece separation method comprises the following steps: S1: obtaining an RGB image and a depth image on the current transmission line; S2: aligning the RGB image and the depth image; S3: performing target detection on the RGB image and the depth image respectively, and obtaining an RGB image detection result and a depth image detection result; S4: fusing the RGB image detection result and the depth image detection result to obtain the layout and form of the objects on the transmission line; S5: adjusting the separation order and speed of the array belt according to the positional relationship between the layout form of the objects and the array belt. The present invention collects RGB images and depth images, aligns and fuses them, obtains the layout and form of the objects and their relationship with the array belt, can greatly improve the accuracy of separation, and is deployed at the separation site, can greatly improve the speed and efficiency of sorting.

Description

Technical Field

[0001] The present invention relates to the field of automated processing technology, and in particular to a single piece separation method and device, and a sorting method and system. Background Art

[0002] With the rapid development of the e-commerce industry, online shopping has become the main way of shopping. In order to improve efficiency, online shopping usually packs items into objects or stores them in warehouses. The warehouse will send the items purchased by users to users through logistics. There will also be centralized warehouses at the user's destination, where objects from all over the world will be concentrated in the warehouse, sorted, and then delivered to specific customer addresses.

[0003] With hundreds of millions of express items delivered every day, higher and higher requirements are placed on the automated processing of items. In recent years, express sorting systems with cross-belt sorters, matrix sorting, and AGV robot sorting as the main forms have been deployed in large quantities, greatly improving the automation level of express item sorting. However, the separation of individual items at the front end of the sorting system still relies mainly on manual work, and the object processing capacity and site turnover rate are greatly limited. The main reason restricting the sorting efficiency is mainly in the object separation link.

[0004] To this end, artificial intelligence technology based on image processing and cloud computing have begun to be applied in sorting, which has greatly improved the automation and intelligence levels of sorting. However, due to the complexity of the sorting scenarios, the collected image data is huge, which leads to high computing costs and network transmission lags, making it difficult to further improve the sorting efficiency. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the invention proposes a single-piece separation method based on edge computing, comprising the steps of:

[0006] S1: Get the RGB image and depth image on the current transmission line;

[0007] S2: Align the RGB image and the depth image;

[0008] S3: Perform target detection on the RGB image and the depth image respectively to obtain the RGB image detection result and the depth image detection result;

[0009] S4: Fusing the RGB image detection results and the depth image detection results to obtain the layout and shape of the objects on the transmission line;

[0010] S5: According to the layout of the objects and the positional relationship between the array belts, the separation order and speed of the array belts are adjusted.

[0011] Further, step S2 includes:

[0012] S20: Get the intrinsic parameter matrix of the RGB camera The intrinsic parameter matrix of the depth camera And the external parameter rotation matrix of the RGB camera and the depth camera and the offset matrix

[0013] S21: traverse the coordinates of each pixel point in the depth image, and calculate the coordinates of the pixel point in the depth image corresponding to the pixel point in the RGB image, including:

[0014] The world coordinates are calculated based on the pixel coordinates of the depth image. The formula is:

[0015]

[0016] The external parameter matrix aligns the world coordinates of the depth image and the world coordinates of the RGB image. The formula is:

[0017]

[0018] The pixel coordinates of the RGB image are calculated according to the world coordinates of the RGB image to align the pixel coordinates of the depth image and the pixel coordinates of the RGB image. The formula is:

[0019]

[0020] Furthermore, target detection is performed on RGB images by using a model trained using the YOLACT deep network.

[0021] Furthermore, performing target detection on the depth image includes:

[0022] The depth image is converted to a grayscale image, and the conversion formula is:

[0023]

[0024] Layer the grayscale image according to its histogram;

[0025] Convert the grayscale image after stratification into a binary image;

[0026] Segment the binary image and obtain the number of objects.

[0027] Furthermore, S4 includes:

[0028] S40: RGBDObject detected by RGB image i (i=1, ..., N) and the target DEPTHObject detected by the depth image j (j=1, ..., M) are sorted;

[0029] S41: Search for sorted RGBObject i (i=1,...,N) and DEPTHObject j (j=1, ..., M) overlapping targets, if RGBObject i With DEPTHObject j If they overlap and the overlapping area is greater than 1 / 2 of the smaller area of ​​the two objects, the two objects are merged and recorded as RGBDObject. k ;

[0030] S42: For RGBObject that has not been merged i and DEPTHOject j If the target area is larger than the preset object area, it is retained and marked as RGBDObject k , otherwise discard;

[0031] S43: Get all target detection results RGBDObject k (k=1, ..., K), K is all the detected targets, and the targets are sorted.

[0032] Further, adjusting the separation order and speed of the array belt in S5 includes:

[0033] S50: a pre-fixed value;

[0034] S51: Input object location information;

[0035] S52: When there is no object on the transmission line, each small belt is set to run at a low speed. If there is an object on the transmission line, step S53 is executed:

[0036] S53: Calculate the maximum X value, minimum X value, maximum Y value and minimum Y value of the minimum circumscribed rectangle of each object;

[0037] S54: Calculate the range of the small belt pressed by each object according to the minimum circumscribed rectangle of each object;

[0038] S55: Calculate the running speed of each object in the current image according to the condition of the small belt pressed by the object and the speed of each small conveyor belt in the separation area calculated from the previous frame of image;

[0039] S56: setting the speed of the object;

[0040] S57: According to the result calculated in step S56, a speed value is assigned to each small conveyor belt.

[0041] The present invention also discloses a single-piece separation device based on edge computing, comprising an image processing module and a control module, wherein:

[0042] The image processing module is used to collect and process RGB images and depth images to obtain the layout and shape of objects on the transmission line;

[0043] The control module adjusts the separation order and speed of the array belt according to the layout and shape of the objects.

[0044] The present invention also discloses a sorting method, comprising the steps of:

[0045] Inducting objects onto a conveyor belt;

[0046] Distribute the imported objects;

[0047] Separate objects;

[0048] Center or edge the objects to achieve single-row transmission of objects;

[0049] Transfer the objects transported in a single row to the subsequent processing equipment,

[0050] The method for separating the objects is the single-piece separation method.

[0051] The present invention also discloses a sorting system, comprising a transmission device, wherein the transmission device comprises:

[0052] Induction unit: inducts objects onto the conveyor belt;

[0053] Dispersed unit: Disperse the imported objects;

[0054] Separation unit: separates objects, and the separation unit is the single-piece separation device based on edge computing;

[0055] Centering unit: Center or edge the objects to achieve single-row transmission of objects;

[0056] Export unit: transfers the objects transported in a single column to the subsequent processing equipment.

[0057] Compared with the prior art, the present invention simultaneously collects RGB images and depth images, analyzes the RGB images and depth images, and then aligns and fuses them to obtain the layout and shape of the objects and their positional relationship with the array belt, which can greatly improve the separation accuracy. The algorithm is deployed at the separation site, which can greatly improve the separation speed and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of a single-piece separation method based on edge computing of the present invention;

[0059] Figure 2 A schematic diagram of object detection in a certain frame of image according to an embodiment of the present application;

[0060] Figure 3 This is a schematic diagram of an application scenario of a single-piece separation device based on edge computing in this application;

[0061] Figure 4 This is a schematic diagram of the structure of a sorting system according to an embodiment of the present application. DETAILED DESCRIPTION

[0062] The following describes the implementation of the present invention through specific examples and in conjunction with the accompanying drawings. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific examples, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0063] The objects in this application generally refer to target objects flowing on the logistics line or production line, including products, parcels, express packages, packaging boxes, etc. In the process of automated sorting, object single-piece separation is mainly used in the front-end part of various automated sorting equipment. Its purpose is to automatically separate the objects placed randomly on the transmission line or belt conveyor, and output them in a single column in an orderly manner at a certain spacing, so as to achieve effective docking with the automatic sorting equipment.

[0064] In the sorting scenario, there are a large number of objects, and they are required to be sorted quickly. Therefore, the amount of image data collected is very large. Coupled with the complex algorithms, a processing platform with powerful computing power is required. Cloud computing is an effective solution to the computing power demand and has also been applied in some fields. However, the sorting scenarios are mainly in closed factory environments, or even in remote industrial parks. Therefore, the huge amount of image data is transmitted to the background server for calculation, analysis, and then feedback control instructions are issued. Due to cloud computing and special scenarios, it is inevitable that it will be affected by high latency, network instability, etc., which in turn affects the efficiency of sorting. In order to solve the defects of the prior art, the present application discloses a single-piece separation method based on edge computing, which directly deploys image analysis processing and control at the site closest to the sorting, so as to give full play to the low latency and high security characteristics of the edge, thereby greatly improving the sorting efficiency. The single-piece separation method based on edge computing of the present application includes the steps of:

[0065] S1: Get the RGB image and depth image on the current transmission line;

[0066] S2: Align the RGB image and the depth image;

[0067] S3: Perform target detection on the RGB image and the depth image respectively to obtain the RGB image detection result and the depth image detection result;

[0068] S4: Fusing the RGB image detection results and the depth image detection results to obtain the layout and shape of the objects on the transmission line;

[0069] S5: According to the layout of the objects and the positional relationship between the array belts, the separation order and speed of the array belts are adjusted.

[0070] This application uses a camera (such as an RGB-D camera) to simultaneously collect RGB images and depth images, align the RGB images and depth images, and then perform target detection on the RGB images and depth images respectively, and fuse the two detection results to output the layout and form of the object, and then set the separation order and speed of the objects based on the layout and form of the objects and their positional relationship with the array belt, based on the spacing required for separation, and convert the speed of the objects into the speed of the array belt and output it to the controller that controls the operation of the belt. The array belt will run at different speeds, and the single-piece separation of the objects can be achieved through the different transmission speeds of the small belt. The technical solution of this application can increase the separation accuracy to more than 99%, and the efficiency can reach more than 6,000 pieces per hour.

[0071] The technical solution of the present application is described in detail below in combination with embodiments and various preferred methods.

[0072] S1: Collect the RGB image and depth image on the current transmission line:

[0073] In an embodiment of the present application, an RGB-D camera is used to simultaneously capture RGB images and depth images. The depth image is an image or image channel containing information related to the distance of the surface of the scene object from the viewpoint, and each pixel value is the actual distance between the sensor and the object.

[0074] S2: Align the RGB image and the depth image.

[0075] In this single-piece separation system, it is necessary to determine which small belts in the separation area the object is located on based on its position. In order to obtain more accurate position information of the object, positioning and segmentation are performed based on the RGB image and the depth image respectively. First, the RGB image and the depth image are aligned. After the alignment operation is completed, all pixels in the RGB image correspond to all pixels in the depth image one by one. As a preferred embodiment, the alignment method of the present application includes the following steps S20 and S21:

[0076] S20: Get the intrinsic parameter matrix of the RGB camera The intrinsic parameter matrix of the depth camera And the external parameter rotation matrix of the RGB camera and the depth camera and the offset matrix

[0077] S21: traverse the coordinates of each pixel point in the depth image, and calculate the coordinates of the pixel point in the depth image corresponding to the pixel point in the RGB image, including:

[0078] The world coordinates are calculated based on the pixel coordinates of the depth image. The formula is:

[0079]

[0080] The external parameter matrix aligns the world coordinates of the depth image and the world coordinates of the RGB image. The formula is:

[0081]

[0082] The pixel coordinates of the RGB image are calculated according to the world coordinates of the RGB image. The formula is:

[0083]

[0084] depth_j and depth_i represent the pixel coordinates of the depth image, Z depth_c Indicates the depth value of the coordinate (depth_j, depth_i). depth_c and Y depth_c Represents the world coordinates of the depth image.

[0085] rgb_j and rgb_i represent the pixel coordinates of the rgb image, Z rgb_c Indicates the depth value of the coordinate (rgb_j, rgb_i). rgb_c and Y rgb_c Represents the world coordinates of an RGB image.

[0086] Through these three formulas, the alignment of the depth image pixel coordinates and the RGB image pixel coordinates is achieved.

[0087] S3: Perform target detection on the RGB image and the depth image respectively to obtain the RGB image detection result and the depth image detection result.

[0088] The embodiment of the present application uses the YOLACT deep network to train the target detection of RGB images. YOLACT is a simple, fully convolutional real-time instance segmentation network model. First, a large amount of RGB image data is used to iteratively train the YOLACT network to modify the weight parameters in the network so that the weight parameters reach the global optimum. Then, using the trained weight parameters, the YOLACT network forward reasoning operation is performed on the input RGB image, and finally the network output result is post-processed to obtain the final target detection result.

[0089] The specific parameter information of the YOLACT network model training is as follows: The optimizer for weight parameter update uses SGD, and a total of 80k iterations of training are performed, with an initial learning rate of 10 -3 , weight decay is set to 5×10 -4 After training, input the test RGB image and output the target detection result RGBObject in the image. i (i=1, ..., N), where N is the number of objects detected.

[0090] In the target detection of the depth image of the embodiment of the present application, the depth image data is the distance data between the camera and the object and the conveyor belt, which reflects the height of the object. However, for the scene applied by the present application, since objects of different heights are close together, the segmentation difficulty is increased. Therefore, it is necessary to perform layered processing on the depth data, and then perform target segmentation on the depth data of different levels, in order to realize the segmentation detection of multiple objects. As a preferred implementation method, the method is as follows:

[0091] Convert the depth image to a grayscale image. According to the maximum and minimum heights of the objects specified by the algorithm, convert the depth image to a grayscale image. The conversion formula is as follows:

[0092]

[0093] Among them, pixelValue represents the pixel value in the depth image, boxMinHeight represents the minimum height of the object to be detected, boxMaxHeight represents the maximum height of the object to be detected. When the pixel value pixelValue in the depth image is greater than the minimum height boxMinHeight and less than the maximum height of the object to be detected, its value is the above formula, and the value is 255 in other cases.

[0094] Layering the grayscale image according to the grayscale image histogram: According to the threshold of the number of pixels in the grayscale image histogram, the grayscale image histogram is divided into multiple different and valid pixel ranges. These pixel ranges are the pixel values ​​displayed in each layer of the image after the image is layered.

[0095] Then the grayscale image after stratification is converted into a binary image;

[0096] Finally, the binary image is segmented to obtain the number of objects: the multiple binary images obtained after the grayscale image is layered are used to mark the pixels as targets using the connected region analysis algorithm. Finally, the maximum and minimum thresholds of the number of connected pixels in the connected region are used to determine the effective connected region, which is the target DEPTHObject. j (j=1, ..., M), M is the number of detected targets.

[0097] S4: Fusing the RGB image detection results and the depth image detection results to obtain the layout and shape of objects on the transmission line. As a preferred implementation, the steps include:

[0098] S40: RGBObject detected by RGB image respectively i (i=1, ..., N) and the target DEPTHObject detected by the depth image j (j=1, ..., M) are sorted, for example, from left to right and from top to bottom;

[0099] S41: Search for sorted RGBObject i (i=1,...,N) and DEPTHObject j (j=1, ..., M) overlapping targets, if RGBObject i With DEPTHObject j If they overlap and the overlapping area is greater than 1 / 2 of the smaller area of ​​the two objects, the two objects are merged and recorded as RGBDObject. k

[0100] S42: For RGBObject that has not been merged i and DEPTHObject j If the target area is larger than the preset object area, it is retained and marked as RGBDObject k , otherwise discarded.

[0101] S43: Get all target detection results RGBDObject k (k=1, ..., K), K represents all detected targets, and the targets are sorted, for example, from left to right and from top to bottom.

[0102] After the above step S4, the layout and shape of the objects on the transmission line are obtained, and then the speed of the belt array on the transmission line can be adjusted according to the layout and shape of the objects, thereby achieving separation of the objects.

[0103] S5: According to the layout of the objects and the positional relationship between the array belts, the separation order and speed of the array belts are adjusted.

[0104] The adjustment of the array belt is based on the object position information and separation spacing requirements output by the current frame image segmentation module, combined with the speed value of each small belt corresponding to the previous frame image, to calculate the current speed of each small belt in the separation area of ​​the current frame. By setting different speed values ​​for each small belt in the separation area, the object is pulled to achieve the goal of this system.

[0105] See also Figure 2 , Figure 2 A schematic diagram of the layout and shape of objects is obtained for the embodiment of the present application. The transmission line described in S1 refers to the transmission area where the objects are to be separated. As shown in the figure, in the belt array, there are three objects in the figure, the object with serial number 1, and there are a total of 6 small belts carrying the object. If you want to separate the object from other objects, you only need to adjust the speed of the 6 small belts.

[0106] As a preferred implementation method of control adjustment, the method is as follows:

[0107] S50: pre-define some fixed values, including high speed value, low speed value, stop line and separation distance. The high speed value refers to the maximum speed of the belt in the separation zone, the low speed value refers to the normal speed of the belt in the separation zone, the stop line refers to the stop position of the object waiting for separation, and the separation distance refers to the distance between the object and the previous object when it leaves the separation zone.

[0108] S51: Input object position information RGBObject k (k=1,...,K).

[0109] S52: When there is no object on the transmission line, each small belt is set to run at a low speed. If there is an object on the transmission line, step S53 is executed:

[0110] S53: Calculate the maximum X value, minimum X value, maximum Y value and minimum Y value of the minimum circumscribed rectangle of each object.

[0111] S54: Calculate the range of the small belt pressed by each object based on the minimum circumscribed rectangle of each object.

[0112] S55: Calculate the running speed of each object in the current image according to the condition of the small belt pressed by the object and the speed of each small conveyor belt in the separation area calculated in the previous frame image.

[0113] S56: Set the speed of the object:

[0114] If it is the first object, the conveyor belt pressed by the object is set to the high speed value.

[0115] If it is not the first object, the distance between this object and the object in front needs to be determined. If the distance is greater than the preset separation distance, the conveyor belt occupied by the current object is set to high speed;

[0116] If the interval is less than or equal to the separation spacing, it is also necessary to determine whether the current object position is greater than the object stop line. If the current object position is greater than or equal to the stop line, the speed of the conveyor belt occupied by the current object is set to stop; if the current object position is less than the stop line, the speed of the conveyor belt occupied by the current object is set to low speed.

[0117] S57: According to the result calculated in step S56, a speed value is assigned to each small conveyor belt.

[0118] The present application also provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the single-piece separation method.

[0119] Based on the above embodiments, the present application further provides a single-piece separation device based on edge computing, including an image processing module and a control module, wherein:

[0120] The image processing module is used to collect and process RGB images and depth images to obtain the layout and shape of objects on the transmission line;

[0121] The control module adjusts the separation order and speed of the array belt according to the layout and shape of the objects.

[0122] See also Figure 3 , Figure 3 For the application scenario of the single-piece separation device based on edge computing in this application, the single-piece separation device based on edge computing in this application can be integrated into a processing platform of an AI smart box, and the image processing module or image processing algorithm, the control module or control strategy algorithm are deployed in the AI ​​smart box. The deep network acceleration inference framework of the AI ​​smart box is fully utilized to realize efficient real-time processing of the image processing algorithm and the control strategy algorithm, and more than 20 frames can be processed per second, which fully meets the real-time processing requirements of the single-piece separation system. Since it is integrated into an AI smart box, it can be deployed and installed at the position closest to the transmission line according to the production site, thereby greatly improving the separation efficiency.

[0123] The specific deployment includes three stages: the training stage of the network model, the conversion stage of the network model format, and the optimization stage of the model structure by the TensorRT inference optimizer.

[0124] Network model training phase: Network model training refers to training the YOLACT network model on the server.

[0125] Network model format conversion stage: The trained network model structure and weight parameters are saved as .onnx files in ONNX format.

[0126] The TensorRT inference optimizer optimizes the model structure: Use the TensorRT inference optimizer to optimize the ONNX model structure and generate the TensorRT runtime engine.

[0127] In addition, based on the above embodiment, the present application also provides a sorting method, comprising the steps of:

[0128] Inducting objects onto a conveyor belt;

[0129] Distribute the imported objects;

[0130] Separate objects;

[0131] Center or edge the objects to achieve single-row transmission of objects;

[0132] Transfer the objects transported in a single row to the subsequent processing equipment,

[0133] The method for separating the objects is the single-piece separation method described in the above embodiment.

[0134] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of a sorting system in an embodiment of the present application. Based on the above embodiment, the present application provides a sorting system, including a transmission device, wherein the transmission device includes:

[0135] Import unit: imports objects onto the conveyor belt;

[0136] Dispersed unit: Disperse the imported objects;

[0137] Separation unit: separates the objects, and the separation unit is a single-piece separation device based on edge computing in the above embodiment;

[0138] Centering unit: Center or edge the objects to achieve single-row transmission of objects;

[0139] Export unit: transfers the objects transported in a single column to the subsequent processing equipment.

[0140] The camera detects and updates the object status of the separation unit at a processing rate of more than 20 frames per second, and then sends the collected image to the processing platform for information processing. The processing platform is the above-mentioned AI smart box. As a processing platform, the AI ​​smart box can generate control instructions to assign values ​​to the motors on the belt array, and then separate the objects on the belt and send them to the centering unit.

[0141] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Any person skilled in the art may modify and alter the above embodiments without violating the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be as set forth in the claims.

Claims

1. A single-piece separation method based on edge computing, characterized in that: Includes steps: S1: Get the RGB image and depth image on the current transmission line; S2: Align the RGB image and the depth image; S3: Perform target detection on the RGB image and the depth image respectively to obtain the RGB image detection result and the depth image detection result; S4: Fusing the RGB image detection results and the depth image detection results to obtain the layout and shape of objects on the transmission line, including: S40: RGBObject detected by RGB image respectively i (i=1, ..., N) and the target DEPTHObject detected by the depth image j (h=1, ..., M) are sorted; S41: Search for sorted RGBObject i (i=1,...,N) and DEPTHObject j (j=1, ..., M) overlapping targets, if RGBObject i With DEPTHObject j If they overlap and the overlapping area is greater than 1 / 2 of the smaller area of ​​the two objects, the two objects are merged and recorded as RGBDObject. k ; S42: For RGBObiect that has not been merged i and DEPTHObject j If the target area is larger than the preset object area, it is retained and marked as RGBDObject k , otherwise discard; S43: Get all target detection results RGBDObject k (k=1, ..., K), K is all the detected targets, and the targets are sorted; S5: According to the layout of the objects and the positional relationship between the array belts, the separation order and speed of the array belts are adjusted.

2. The single piece separation method according to claim 1, characterized in that: Step S2 includes: S20: Get the intrinsic parameter matrix of the RGB camera The intrinsic parameter matrix of the depth camera And the external parameter rotation matrix of the RGB camera and the depth camera and the offset matrix S21: traverse the coordinates of each pixel point in the depth image, and calculate the coordinates of the pixel point in the depth image corresponding to the pixel point in the RGB image, including: The world coordinates are calculated based on the pixel coordinates of the depth image. The formula is: The external parameter matrix aligns the world coordinates of the depth image and the world coordinates of the RGB image. The formula is: The pixel coordinates of the RGB image are calculated according to the world coordinates of the RGB image to align the pixel coordinates of the depth image and the pixel coordinates of the RGB image. The formula is:

3. The single piece separation method according to claim 1, characterized in that: The target detection for RGB images is: detection is performed using the model trained with the YOLACT deep network.

4. The single piece separation method according to claim 1, characterized in that: Object detection on depth images includes: The depth image is converted to a grayscale image, and the conversion formula is: Layer the grayscale image according to its histogram; Convert the grayscale image after stratification into a binary image; Segment the binary image and obtain the number of objects.

5. The single piece separation method according to claim 1, characterized in that: Adjustment of the separation sequence and speed of the array belts in S5 includes: S50: a pre-fixed value; S51: Input object location information; S52: When there is no object on the transmission line, each small belt is set to run at a low speed. If there is an object on the transmission line, step S53 is executed: S53: Calculate the maximum X value, minimum X value, maximum Y value and minimum Y value of the minimum circumscribed rectangle of each object; S54: Calculate the range of the small belt pressed by each object according to the minimum circumscribed rectangle of each object; S55: Calculate the running speed of each object in the current image according to the condition of the small belt pressed by the object and the speed of each small conveyor belt in the separation area calculated from the previous frame of image; S56: setting the speed of the object; S57: According to the result calculated in step S56, a speed value is assigned to each small conveyor belt.

6. A single-piece separation device based on edge computing, used to implement the single-piece separation method described in any one of claims 1 to 5, characterized in that: It includes an image processing module and a control module, wherein: The image processing module is used to collect and process RGB images and depth images to obtain the layout and shape of objects on the transmission line; The control module adjusts the separation order and speed of the array belt according to the layout and shape of the objects.

7. A sorting method comprising the steps of: Inducting objects onto a conveyor belt; Distribute the imported objects; Separate objects; Center or edge the objects to achieve single-row transmission of objects; Transfer the objects transported in a single row to the subsequent processing equipment, It is characterized in that the method for separating objects is the single-piece separation method described in any one of claims 1 to 5.

8. A sorting system, comprising a transmission device, characterized in that: The transmission device includes: Import unit: imports objects onto the conveyor belt; Dispersed unit: Disperse the imported objects; Separation unit: to separate objects, the separation unit is the single-piece separation device based on edge computing as described in claim 6; Centering unit: Center or edge the objects to achieve single-row transmission of objects; Export unit: transfers the objects transported in a single column to the subsequent processing equipment.

Images