A weighing and settlement integrated machine settlement method and system

Abstract

The application discloses a kind of settlement methods and systems of weighing settlement integrated machine, it is related to food sales technical field;Among them method includes: obtaining the RGB image and depth image of target customer;Extract the face frame and body frame of target customer in RGB image;According to the center coordinate information of face frame obtained in RGB image;According to the depth information of body frame obtained in depth image;When satisfying to be identified condition, the face subgraph in RGB image is intercepted, and face recognition is carried out, and the corresponding ID information is obtained;Binding body frame and ID information, to carry out target tracking to body frame, when satisfying unbinding condition, unbinding is cancelled, and the meal price corresponding to ID information is generated;Unbinding condition includes: body frame tracking loss or the frame depth distance between current frame body frame and previous frame body frame exceeds preset threshold;The application reduces tray binding step, saves merchant cost, improves customer dining efficiency, and can also reduce the risk of false tracking.

Description

Technical Field

[0001] This invention relates to the field of food sales technology, and in particular to a settlement method and system for an integrated weighing and settlement machine. Background Technology

[0002] With the rapid advancement of technology and the increasingly fast pace of modern life, self-service systems have widely penetrated various industries, especially in the catering sector. The application of self-service food collection and payment systems has become a significant trend in improving service efficiency and optimizing customer experience. In current self-service food collection systems, the integrated weighing and payment machine serves as the core terminal device and is particularly prevalent. Users first bind their personal payment method to a tray equipped with a PFID or one-dimensional / two-dimensional barcode by scanning a QR code, then place the tray in the food collection recognition area. The device identifies the tray information to determine the entire process from start to finish. During the food collection process, the system calculates the weight of the food and generates the corresponding fee. After the food is collected, the system automatically settles the payment through the user's pre-authorized payment platform, achieving full automation and convenience.

[0003] The core of this self-service food pickup system lies in how to link the price of the food pickup with the person picking it up. Currently, the main models on the market include: 1. Embedding an RFID chip in the tray; users scan the code to pick up the tray, binding the tray to the user, and then the device identifies the chip in the tray to link the user to the food pickup price; 2. Printing one-dimensional or two-dimensional codes on the tray for scanning to replace the RFID chip, reducing tray costs; 3. Using multiple surveillance cameras to identify and track the current person picking up food; This solution uses facial recognition to replace the previous method of using a tray as a medium to bind the food pickup price, and uses pedestrian tracking to determine the start and end of the food pickup process.

[0004] However, all of the above models have shortcomings: In the first model, a large number of chip-embedded trays need to be purchased, which is costly, and diners need to scan the code to bind the trays in advance, adding steps to the dining process; In the second model, a large number of trays also need to be purchased. Although the cost of chips is replaced by printing one-dimensional or two-dimensional codes, they are easily damaged during disinfection and cleaning, and the dining process is also increased; In the third model, although the method of using multiple cameras to track people reduces the steps of binding trays, it has the problem of high cost, because an additional monitoring camera needs to be purchased for every one or two scales, and a private server is needed for every few scales; The installation of monitoring cameras has high requirements for the on-site environment, is difficult to construct, and has low flexibility.

[0005] Meanwhile, in scenarios involving uniforms, relying solely on pure RGB images for target tracking carries certain risks, especially in crowded places like cafeterias. Due to the uniformity of the uniforms, when multiple people wearing the same uniform move in the frame, the target tracking algorithm may mistakenly "inherit" the bounding box of one pedestrian as belonging to another person with similar attire. This misidentification is particularly serious in payment scenarios, potentially leading to incorrect identity verification and improper deductions. While lowering the bounding box matching threshold might seem to solve some of the mistracking problems, this approach actually makes the tracking conditions for the same person more stringent, thus reducing the accuracy and stability of the tracking. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a settlement method and system for an integrated weighing and settlement machine, which can reduce the steps of tray binding, save costs for merchants, improve customer dining efficiency, and reduce the risk of mistracking.

[0007] To achieve the above objectives, the present invention provides a settlement method for a weighing and settlement integrated machine, comprising:

[0008] Acquire the RGB and depth images of the target customer;

[0009] Extract the face bounding box and body bounding box of the target customer from the RGB image;

[0010] Based on the RGB image, obtain the center coordinate information of the face bounding box, and based on the depth image, obtain the depth information of the human body bounding box;

[0011] When the recognition conditions are met, a face sub-image is captured from the RGB image; the recognition conditions include: determining that the face frame is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face frame, and determining that the corresponding human body frame is located within a preset area based on the depth information;

[0012] Face recognition is performed based on the aforementioned face sub-image to obtain the corresponding ID information;

[0013] Bind the human body bounding box and the ID information to perform target tracking on the human body bounding box;

[0014] When the unbinding conditions are met, the binding between the human body bounding box and the ID information is released, and a meal price corresponding to the ID information is generated; the unbinding conditions include: the inter-frame depth distance between the human body bounding box of the current frame and the human body bounding box of the previous frame exceeds a preset threshold.

[0015] Optionally, extracting the target customer's face bounding box and body bounding box from the RGB image includes:

[0016] The pose model of YOLO v8 is used to detect the human skeleton points of the target customer in the RGB image;

[0017] The human skeleton points are converted into human body frames and human face frames.

[0018] Optionally, before obtaining the depth information of the human body bounding box based on the depth image, the method further includes:

[0019] Collect the point set X on the x-axis of the RGB image. RGB and the set of points X on the x-axis of the depth image TOF ;

[0020] Using the mapping function F on the point set X RGB And point set X TOF To perform coordination; the expression for the mapping function F is: F = [(X RGB ,1) T (X RGB ,1)] -1 (X RGB ,1) T X TOF ;

[0021] In the formula, T represents the transpose of the matrix.

[0022] Optionally, based on the depth image, the depth information of the human bounding box is obtained, including:

[0023] Obtain the TOF data corresponding to the depth image;

[0024] The left boundary L of the human body frame RGB and right boundary R RGB The median D mapped onto the TOF data is used as the depth information; the expression for the median D is:

[0025] D = Median(d([x i ,1]·F)),i=[L RGB L RGB +1, ..., R RGb -1, R RGR ];

[0026] In the formula, i represents the set of x-axis components of the human body bounding box from the left boundary to the right boundary; x i represents the element in the x-component vector of the RGB image coordinates; F represents the mapping function between the RGB camera and the TOF camera with respect to the x-axis coordinates; d(·) represents the depth information output by the TOF camera; Median() represents median filtering.

[0027] Optionally, the expression for the preset threshold is:

[0028]

[0029] In the formula, t represents the preset threshold; D diff An empirical value representing the inter-frame depth distance; D max Indicates D diff Weighted maximum effective distance; D min Indicates D diff Weighted minimum effective distance; D oli This indicates the depth distance of the human bounding box in the previous frame.

[0030] Optionally, generating a meal pickup price corresponding to the ID information includes:

[0031] Based on the weight of the first plate in the weighing area initially obtained during the binding of the human body frame and the ID information, and the weight of the second plate in the weighing area last obtained, a meal price corresponding to the ID information is generated.

[0032] Optionally, acquire the RGB image and depth image of the target customer, including:

[0033] The target customer's RGB image is captured by shooting at a horizontal angle using an RGB camera;

[0034] The depth image of the target customer is obtained by shooting downwards at a preset tilt angle using a TOF camera; the RGB camera and the TOF camera are integrated into the weighing and settlement machine.

[0035] Optionally, acquire the RGB image and depth image of the target customer, including:

[0036] RGB images of the target customer are acquired by setting up multiple RGB cameras in different locations;

[0037] Based on the RGB image, a depth image of the target customer is obtained through camera calibration and depth calculation.

[0038] The present invention also provides a settlement system for a weighing and settlement integrated machine, comprising:

[0039] The image acquisition module is used to acquire the RGB image and depth image of the target customer;

[0040] Image processing module, used for:

[0041] Extract the face bounding box and body bounding box of the target customer from the RGB image;

[0042] Based on the RGB image, obtain the center coordinate information of the face bounding box, and based on the depth image, obtain the depth information of the human body bounding box;

[0043] The face recognition module is used for:

[0044] When the recognition conditions are met, a face sub-image is captured from the RGB image; the recognition conditions include: determining that the face frame is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face frame, and determining that the corresponding human body frame is located within a preset area based on the depth information;

[0045] Face recognition is performed based on the aforementioned face sub-image to obtain the corresponding ID information;

[0046] The tracking pricing module is used for:

[0047] Bind the human body bounding box and the ID information to perform target tracking on the human body bounding box;

[0048] When the unbinding conditions are met, the binding between the human body bounding box and the ID information is released, and a meal price corresponding to the ID information is generated; the unbinding conditions include: the inter-frame depth distance between the human body bounding box of the current frame and the human body bounding box of the previous frame exceeds a preset threshold.

[0049] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0050] The settlement method of the weighing and settlement machine provided by this invention eliminates the need for customers to bind trays. It uses a face recognition and human body binding scheme based on RGB and depth images, replacing the previous dining steps that required pre-binding trays. This reduces costs for merchants and simplifies the dining process for diners. In acquiring the RGB and depth images of the target customer, only the installation of corresponding RGB and TOF cameras is required, making deployment relatively simple and requiring no additional construction. When performing multi-target tracking, the combination of RGB and depth images effectively reduces the risk of mistracking that may occur when relying solely on RGB images for visual tracking. Compared with existing multi-target pedestrian tracking algorithms, it reduces the possibility of human body tracking being "inherited" without altering the tracking effect for the same person. Attached Figure Description

[0051] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.

[0052] Figure 1 This is a schematic diagram of the settlement method of the weighing and settlement integrated machine according to an embodiment of the present invention;

[0053] Figure 2This is a schematic diagram illustrating the target customer tracking process according to an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of the module structure of the settlement system of the weighing and settlement integrated machine according to an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The purpose of this invention is to provide a settlement method and system for a weighing and settlement integrated machine, which can reduce the pallet binding steps, save costs for merchants and improve customer dining efficiency, while also reducing the risk of mistracking.

[0057] Please see Figure 1 , Figure 1 This is a flowchart illustrating the settlement method of the integrated weighing and settlement machine. The settlement method of the aforementioned integrated weighing and settlement machine includes:

[0058] Step 101: Obtain the RGB image and depth image of the target customer.

[0059] In the application, an RGB camera can be used to capture an RGB image of the target customer at a horizontal angle; and a TOF camera can be used to capture a depth image of the target customer at a preset tilt angle. The RGB and TOF cameras are integrated into the weighing and settlement machine. An 8x8 pixel TOF camera can be used as an example.

[0060] The RGB camera shoots at a horizontal angle, focusing on capturing the facial information and body dynamics of the target customers; while the TOF camera shoots downwards at a certain tilt angle (preset tilt angle), which can adapt to diners of different heights and ensure that the distance dynamics of each target customer can be accurately captured.

[0061] Compared to the previous solution of installing multiple surveillance cameras, this invention only requires adding a TOF camera and binding the position of the RGB camera, which reduces hardware costs and deployment difficulty, making it more economical and easier to implement.

[0062] In applications, RGB images of the target customer can be acquired by using multiple RGB cameras set up at different locations; based on the RGB images, depth images of the target customer can be obtained through camera calibration and depth calculation.

[0063] In this system, multiple RGB cameras, such as two RGB cameras, are set up in different locations. Through camera calibration and depth calculation, the location of the target customer can be determined and the corresponding depth image can be obtained.

[0064] Alternatively, the same effect can be achieved by setting up an RGBD camera. An RGBD camera can acquire both RGB images and depth images; the depth information acquisition for the RGBD camera can be done using a TOF camera, other structured light cameras, or binocular RGB cameras, with proper pixel-to-pixel calibration between the sensors.

[0065] Step 102: Extract the target customer's face bounding box and body bounding box from the RGB image.

[0066] In applications, deep learning models (such as YOLO, SSD, etc.) can be used to process RGB images and extract the face bounding boxes and body bounding boxes of the target customer.

[0067] In one embodiment, step 102 above includes:

[0068] The pose model of YOLO v8 is used to detect the human skeleton points of the target customer in RGB images;

[0069] Convert human skeleton points into human body bounding boxes and human face bounding boxes.

[0070] Step 103: Obtain the center coordinates of the face bounding box based on the RGB image, and obtain the depth information of the human body bounding box based on the depth image.

[0071] In applications, the real-time location of the target customer can be determined from the depth image. Then, by fusing the depth image and the RGB image, the position of the target customer's face bounding box on the RGB image, i.e., the center coordinate information, can be obtained. Similarly, the distance between the target customer's body bounding box and the TOF camera, i.e., the depth information, can also be obtained.

[0072] In one embodiment, to ensure that the information between the RGB camera and the TOF camera can be fused, the following operations may be performed before step 103:

[0073] Collect the point set X on the x-axis of the RGB image RGB and the set of points X on the x-axis of the depth image. TOF ;

[0074] Using the mapping function F on the point set X RGB And point set X TOF To perform coordination; the expression for the mapping function F is: F = [(X RGB ,1) T (X RGB ,1)] -1(X RGB ,1) T X TOF Among them, collaboration means establishing a point set X. RGB And point set X TOF The mapping relationship between them.

[0075] In one embodiment, obtaining the depth information of the human body bounding box based on the depth image in step 103 above specifically includes:

[0076] Acquire TOF data corresponding to the depth image; wherein, the TOF data includes distance information between each pixel in the human bounding box and the TOF camera;

[0077] Set the left boundary L of the human body frame RGB and right boundary R RGB The median D mapped onto the TOF data is used as depth information; the expression for the median D is:

[0078] D = Median(d([x i ,1]·F)),i=[L RGB L RGB +1, ..., R RGB -1, R RGB ];

[0079] In the formula, i represents the set of x-axis components of the human body bounding box from the left boundary to the right boundary; x i represents the element in the x-component vector of the RGB image coordinates; F represents the mapping function between the RGB camera and the TOF camera with respect to the x-axis coordinates; d(·) represents the depth information output by the TOF camera; Median() represents median filtering.

[0080] Step 104: When the recognition conditions are met, extract the face sub-image from the RGB image.

[0081] The conditions to be identified include: determining that the face bounding box is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face bounding box, and determining that the corresponding human body bounding box is located within the preset area based on the depth information.

[0082] Here, the target recognition location refers to the position range of the face bounding box in the RGB image to facilitate face recognition; in application, when the detected face bounding box meets the preset conditions (recognition location matches, liveness detection passes), and the corresponding human bounding box meets a specific distance range (D) on the TOF camera, the target recognition location is determined. min D max When this is done, a facial image of the target customer can be extracted from an RGB image. It can be understood that the preset area corresponds to D... min To D maxThe range between; for example, a preset area can be set to the area within a certain distance from the TOF camera, such as 0.1 meters to 1 meter.

[0083] When performing liveness detection, one can determine whether a person is alive within the face frame based on detailed features in the RGB image (such as skin texture, blinking, etc.). Furthermore, one can combine it with an IR camera (infrared camera) for liveness detection. By taking advantage of the fact that screens (such as mobile phone displays) cannot image under IR and that the texture of photographs does not match the image, one can effectively deal with situations where someone else's photo is used for face recognition and prevent false recognition.

[0084] Step 105: Perform face recognition based on the face sub-image to obtain the corresponding ID information.

[0085] In applications, captured facial images can be input into a facial recognition system, and the corresponding customer ID information can be obtained by comparing them with pre-stored facial information in the database.

[0086] Step 106: Bind the human bounding box and ID information to perform target tracking on the human bounding box.

[0087] In the application, the obtained ID information is bound to the previously detected human bounding box to achieve continuous tracking of the target customer while ensuring that the currently tracked individual has the correct identity.

[0088] Step 107: When the unbinding conditions are met, unbind the human body frame and ID information, and generate the meal price corresponding to the ID information.

[0089] The unbinding conditions include: the inter-frame depth distance between the human body bounding box in the current frame and the human body bounding box in the previous frame exceeds a preset threshold.

[0090] Please see Figure 2 During the process of diners (target customers) retrieving food, the system can continuously track the currently bound human body, locking onto the target until the tracking signal is interrupted, at which point the binding is released. Given that uniforms may cause confusion during tracking, this invention sets an empirical value D for the inter-frame depth distance. diff To enable adaptive judgment; in applications, the depth distance D of the human body bounding box in the previous frame can be used. old The inter-frame distance threshold T (preset threshold) required to generate the current frame.

[0091] For example, the expression for the preset threshold T can be:

[0092]

[0093] In the formula, t represents the preset threshold; D diff An empirical value representing the inter-frame depth distance; Dmax Indicates D diff Weighted maximum effective distance; D min Indicates D diff Weighted minimum effective distance; D old This indicates the depth distance of the human bounding box in the previous frame.

[0094] The inter-frame depth distance refers to the difference between the first distance between the human bounding box and the TOF camera in the previous frame and the second distance between the human bounding box and the TOF camera in the next frame, which are adjacent images. The distance between the human bounding box and the TOF camera can be represented by the median D mentioned above.

[0095] This means that regardless of whether the target is temporarily lost during the tracking process for some reason, as long as the inter-frame depth distance of the tracked target exceeds the preset threshold, i.e.: DD old If the value is greater than t, the system will abandon marking the object, actively end the current food collection process, and generate a collection price; thus reducing improper deductions due to tracking errors. This design improves the stability and reliability of the self-service weighing and payment machine, ensuring a better dining experience for diners.

[0096] In one embodiment, generating a meal pickup price corresponding to the ID information includes:

[0097] Based on the weight of the first plate in the weighing area initially obtained during the binding of the human body frame and ID information, and the weight of the second plate in the weighing area last obtained, a meal price corresponding to the ID information is generated.

[0098] The settlement method provided by this invention can be applied to integrated weighing and settlement machines for canteen dishes. Previous canteen weighing and settlement systems often required attaching a tray to complete the food dispensing and pricing process. This invention, however, utilizes facial recognition and human body tracking technology to completely replace the cumbersome tray-attaching process. This not only significantly reduces the canteen's hardware costs, decreasing equipment investment and maintenance expenses, but also simplifies the operation steps for diners, improving the dining experience. Furthermore, by utilizing depth images from a TOF camera, it effectively reduces the risk of mistracking that may occur when relying solely on RGB images for visual tracking.

[0099] Corresponding to the aforementioned application function implementation method embodiments, the present invention also provides a settlement system for a weighing and settlement integrated machine and corresponding embodiments.

[0100] Please see Figure 3 , Figure 3 This is a schematic diagram of the module structure of the settlement system of the aforementioned weighing and settlement integrated machine. The settlement system includes:

[0101] Image acquisition module 31 is used to acquire the RGB image and depth image of the target customer;

[0102] Image processing module 32, used for:

[0103] Extract the face bounding box and body bounding box of the target customer from the RGB image;

[0104] Based on the RGB image, obtain the center coordinates of the face bounding box, and based on the depth image, obtain the depth information of the human body bounding box;

[0105] Face recognition module 33 is used for:

[0106] When the recognition conditions are met, a face sub-image is extracted from the RGB image; the recognition conditions include: determining that the face bounding box is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face bounding box, and determining that the corresponding human body bounding box is located within the preset area based on the depth information;

[0107] Face recognition is performed based on face sub-images to obtain the corresponding ID information;

[0108] Tracking pricing module 34 is used for:

[0109] Bind the human bounding box and ID information to perform target tracking on the human bounding box;

[0110] When the unbinding conditions are met, the binding between the human body bounding box and the ID information is released, and a meal price corresponding to the ID information is generated. The unbinding conditions include: the inter-frame depth distance between the human body bounding box in the current frame and the human body bounding box in the previous frame exceeds a preset threshold.

[0111] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated further here.

[0112] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A settlement method for a weighing and settlement integrated machine, characterized in that, include: Acquire the RGB and depth images of the target customer; Extract the face bounding box and body bounding box of the target customer from the RGB image; Based on the RGB image, obtain the center coordinate information of the face bounding box, and based on the depth image, obtain the depth information of the human body bounding box; When the recognition conditions are met, a face sub-image is captured from the RGB image; the recognition conditions include: determining that the face frame is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face frame, and determining that the corresponding human body frame is located within a preset area based on the depth information; Face recognition is performed based on the face sub-image to obtain the corresponding ID information; the human body bounding box and the ID information are bound together to perform target tracking on the human body bounding box; When the unbinding conditions are met, the binding between the human body bounding box and the ID information is released, and a meal price corresponding to the ID information is generated; the unbinding conditions include: the inter-frame depth distance between the human body bounding box of the current frame and the human body bounding box of the previous frame exceeds a preset threshold. Extracting the face bounding box and body bounding box of the target customer from the RGB image includes: The pose model of YOLO v8 is used to detect the human skeleton points of the target customer in the RGB image; the human skeleton points are converted into human body bounding boxes and face bounding boxes; Before obtaining the depth information of the human body bounding box based on the depth image, the method further includes: Collect the point set X on the x-axis of the RGB image. RGB and the set of points X on the x-axis of the depth image TOF ; Using the mapping function F on the point set X RGB And point set X TOF To perform coordination; the expression for the mapping function F is: F = [(X RGB ,1) T (X RGB ,1)] -1 (X RGB ,1) T X TOF .

2. The settlement method of the weighing and settlement integrated machine according to claim 1, characterized in that, Based on the depth image, the depth information of the human body bounding box is obtained, including: obtaining TOF data corresponding to the depth image; and defining the left boundary L of the human body bounding box. RGB and right boundary X RGB The median D mapped onto the TOF data is used as the depth information; the expression for the median D is: D = Median(s([xi, 1]·F)), i = [LRGB, LRGB+1, ..., RRGB-1, RRGB]; In the formula, i represents the set of x-axis components of the human body bounding box from the left boundary to the right boundary; x i represents the element in the x-component vector of the RGB image coordinates; F represents the mapping function between the RGB camera and the TOF camera about the x-axis coordinates; d(·) represents the depth information output by the TOF camera; Median() represents median filtering.

3. The settlement method of the weighing and settlement integrated machine according to claim 1, characterized in that, The expression for the preset threshold is: In the formula, t represents the preset threshold; D diff An empirical value representing the inter-frame depth distance; D max Indicates D diff Weighted maximum effective distance; D min Indicates D diff Weighted minimum effective distance; D old This indicates the depth distance of the human bounding box in the previous frame.

4. The settlement method of the weighing and settlement integrated machine according to claim 1, characterized in that, Generating the meal price corresponding to the ID information includes: Based on the weight of the first plate in the weighing area initially obtained during the binding of the human body frame and the ID information, and the weight of the second plate in the weighing area last obtained, a meal price corresponding to the ID information is generated.

5. The settlement method of the weighing and settlement integrated machine according to claim 1, characterized in that, Obtain the RGB and depth images of the target customer, including: The target customer's RGB image is captured by shooting at a horizontal angle using an RGB camera; The depth image of the target customer is obtained by shooting downwards at a preset tilt angle using a TOF camera; the RGB camera and the TOF camera are integrated into the weighing and settlement machine.

6. The settlement method of the weighing and settlement integrated machine according to claim 1, characterized in that, Obtain the RGB and depth images of the target customer, including: RGB images of the target customer are acquired by setting up multiple RGB cameras in different locations; Based on the RGB image, a depth image of the target customer is obtained through camera calibration and depth calculation.

7. A settlement system for a weighing and settlement integrated machine, characterized in that, include: The image acquisition module is used to acquire the RGB image and depth image of the target customer; Image processing module, used for: Extract the face bounding box and body bounding box of the target customer from the RGB image; Based on the RGB image, the center coordinate information of the face bounding box is obtained, and based on the depth image, the depth information of the human body bounding box is obtained; wherein, extracting the face bounding box and human body bounding box of the target customer in the RGB image includes: using the pose model of YOLOv8 to detect the human skeleton points of the target customer in the RGB image; and converting the human skeleton points into human body bounding boxes and face bounding boxes; Before obtaining the depth information of the human body bounding box based on the depth image, the method further includes: Collect the point set X on the x-axis of the RGB image. RGB and the set of points X on the x-axis of the depth image TOF ; Using the mapping function F on the point set X RGB And point set X TOF To perform coordination; the expression for the mapping function F is: F = [(X RGB ,1) T (X RGB ,1)] -1 (X RGB ,1) T X TOF ; The face recognition module is used for: When the recognition conditions are met, a face sub-image is captured from the RGB image; the recognition conditions include: determining that the face frame is located at the target recognition position based on the center coordinate information, determining that the target customer is a live person based on the face frame, and determining that the corresponding human body frame is located within a preset area based on the depth information; Face recognition is performed based on the aforementioned face sub-image to obtain the corresponding ID information; The tracking pricing module is used for: The human body bounding box and the ID information are bound together to perform target tracking on the human body bounding box; when the unbinding condition is met, the binding between the human body bounding box and the ID information is released, and a meal price corresponding to the ID information is generated; the unbinding condition includes: the inter-frame depth distance between the human body bounding box of the current frame and the human body bounding box of the previous frame exceeds a preset threshold.

Images