Automatic checkout system and control method for the same

Abstract

An automatic checkout system according to one embodiment of the present invention includes: a frame loader that sequentially loads a plurality of frame images; a detector that detects at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images have been manipulated; a tracker that tracks the at least one product detected within the plurality of frame images; and a counting unit that counts the at least one product detected based on the tracking results. Additionally, a method according to one embodiment of the present invention includes a method for synthesizing an image for training an automatic checkout system from a plurality of sample images.

Description

Technology Field

[0001] The present invention relates to an automatic checkout system for automatically identifying and counting products on a checkout counter and a control method thereof. Background Technology

[0002] Recently, the retail industry has seen innovations utilizing artificial intelligence (AI) and computer vision (CV) technologies. Among these, automatic checkout (ACO) technology is considered the most critical technology in this field. In an ideal ACO system, when a customer places their selected items on the counter, the system seamlessly recognizes each product and returns an accurate list of purchased items in a single step.

[0003] Research on such automated checkout systems is being conducted to address issues such as object occlusion, motion blur, product similarity, and costs incurred due to detection or classification errors. Furthermore, when applying machine learning methods like deep learning to build automated checkout systems, it is crucial to collect training video data that reflects the specific characteristics of diverse product types and classifications, as well as the frequent updating of product lists.

[0004] In other words, accuracy, stability, and efficiency are key factors to consider in the development of automated checkout systems. Prior art literature

[0005] Korean Patent Publication No. 10-2015-0140359 The problem to be solved

[0006] The problem to be solved by the present invention is to provide an automatic checkout system and a control method thereof that counts at least one product within a plurality of frame images using a detection model generated by learning a training image based on a plurality of sample images.

[0007] However, the problems that the present invention aims to solve are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art to which the present invention pertains from the description below. means of solving the problem

[0008] An automatic checkout system according to one embodiment of the present invention comprises: a frame loader that sequentially loads a plurality of frame images; a detector that detects at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images have been manipulated; a tracker that tracks the at least one product detected within the plurality of frame images; and a counting unit that counts the at least one product detected based on the tracking result.

[0009] Additionally, the above parameters may include at least one of the number of products included in each of the plurality of sample images, the rotation angle of the sample image, the enlargement or reduction ratio of the sample image, or the gamma adjustment value of the sample image.

[0010] In addition, the above training image can be generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied.

[0011] In addition, the training image may be generated by placing the plurality of sample images to which the binary mask is applied onto the checkout image, wherein the degree of overlap between the products of each of the plurality of sample images is placed within a predetermined range of mess values.

[0012] In addition, it may further include a hand predictor that predicts at least one hand region from each of the plurality of frame images using a hand detection model.

[0013] In addition, the tracker can identify the product held in the hand among at least one of the detected products based on the predicted hand area.

[0014] Additionally, the detector can sequentially store the loaded plurality of frame images in an input queue to detect at least one product from each of the plurality of frame images, the tracker can sequentially store the detected at least one product in a detection queue to track the detected at least one product within the plurality of frame images, and the counting unit can sequentially store the tracking results in a counting queue to count the detected at least one product.

[0015] In addition, the frame loader can load the plurality of frame images according to a predetermined batch size.

[0016] A control method for an automatic checkout system according to an embodiment of the present invention comprises: a step of sequentially loading a plurality of frame images; a step of detecting at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images have been manipulated; a step of tracking the at least one product detected within the plurality of frame images; and a step of counting the at least one product detected based on the tracking result.

[0017] Additionally, the above parameters may include at least one of the number of products included in each of the plurality of sample images, the rotation angle of the sample image, the enlargement or reduction ratio of the sample image, or the gamma adjustment value of the sample image.

[0018] In addition, the above training image can be generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied.

[0019] In addition, the training image may be generated by placing the plurality of sample images to which the binary mask is applied onto the checkout image, wherein the degree of overlap between the products of each of the plurality of sample images is placed within a predetermined range of mess values.

[0020] In addition, the method may further include a step of predicting at least one hand region from each of the plurality of frame images using a hand detection model.

[0021] In addition, the step of tracking at least one product detected above can identify the product held in the hand among the at least one product detected above based on the predicted hand area.

[0022] Additionally, the step of detecting at least one product may involve sequentially storing the loaded plurality of frame images in an input queue to detect at least one product from each of the plurality of frame images, and the step of tracking the detected at least one product may involve sequentially storing the detected at least one product in a detection queue to track the detected at least one product within the plurality of frame images, and the step of counting the detected at least one product may involve sequentially storing the tracking results in a counting queue to count the detected at least one product.

[0023] In addition, the step of sequentially loading the plurality of frame images may load the plurality of frame images according to a predetermined batch size. Effects of the invention

[0024] According to one embodiment of the present invention, the accuracy of product detection, tracking, and counting can be improved. Specifically, by automatically generating a training video that reflects the specific characteristics of an automated checkout environment and using the trained detection model, the performance of the automated checkout system can be improved.

[0025] In addition, by batching the entire system and adjusting it to the queue size, the processing speed of the automated checkout system can be improved. Brief explanation of the drawing

[0026] FIG. 1 is a control block diagram of an automatic checkout system according to one embodiment of the present invention. FIG. 2 is a flowchart of a control method for an automatic checkout system according to one embodiment of the present invention. Figure 3 is a schematic diagram illustrating the operation of an automatic checkout system according to the control method of Figure 2. FIG. 4 is a schematic diagram illustrating a method for generating a training image used for learning a detection model according to an embodiment of the present invention. Figure 5 is a diagram illustrating a training video generated according to the method of Figure 4. FIG. 6 is a diagram illustrating a hand detection model used by a hand predictor according to one embodiment of the present invention. FIG. 7 is a diagram for explaining the control method of an automatic checkout system according to one embodiment of the present invention, divided into a synchronous method and an asynchronous method. Specific details for implementing the invention

[0027] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims.

[0028] In describing the embodiments of the present invention, specific descriptions of known functions or configurations will be omitted if it is determined that such detailed descriptions could unnecessarily obscure the essence of the invention. Furthermore, the terms described below are defined in consideration of their functions in the embodiments of the present invention, and these definitions may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification.

[0029] Additionally, in this specification and claims, terms including ordinal numbers, such as 'first', 'second', etc., may be used to distinguish between components. These ordinal numbers are used to distinguish identical or similar components from one another, and the meaning of the terms should not be limited by the use of such ordinal numbers. For example, the order of use or arrangement of components combined with such ordinal numbers should not be limited by the number. If necessary, each ordinal number may be used interchangeably.

[0030] In this specification, singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, terms such as 'comprising' or 'composing' are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0031] Furthermore, in the embodiments of the present invention, when it is stated that a part is connected to another part, this includes not only a direct connection but also an indirect connection through another medium. Additionally, the meaning that a part includes a certain component implies that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0032] FIG. 1 is a control block diagram of an automatic checkout system according to one embodiment of the present invention.

[0033] The automatic checkout system (100) of the present invention refers to a detection system that accurately detects the existence of each product from a combination of products selected by a user and counts them.

[0034] Generally, such detection systems can be trained using images acquired from environments identical or similar to the actual usage environment. In the case of automatic checkout counters, the training images may be images of various combinations of products placed on the counter.

[0035] However, since the types of products that need to be detected at the checkout counter are vast and inventory status is updated frequently, it may be practically difficult to train the detection model to handle all possible product combinations.

[0036] To solve this, the automatic checkout system (100) of the present invention may use a detection model generated by learning a learning image based on a plurality of sample images.

[0037] Specifically, referring to FIG. 1, an automatic checkout system (100) according to one embodiment of the present invention may include a frame loader (110); a detector (120); a tracker (130); a counting device (140); and a storage unit (150).

[0038] The frame loading device (110) can sequentially load multiple frame images. At this time, the frame image being loaded is an image containing at least one product placed on a checkout counter, and according to one embodiment, the frame image may be acquired by a camera positioned from the top of the checkout counter toward the checkout counter.

[0039] The detector (120) can detect at least one product from each of a plurality of frame images using a detection model. Here, the detection model can be generated by learning from a training image obtained by manipulating the parameters of a plurality of sample images. Such a detection model may be stored in advance in a storage unit (150) and provided to the detector (120) as needed, or may be received directly from an external device, server, or cloud.

[0040] The tracker (130) can track at least one product detected by the detector (120) within a plurality of frame images, and the counting device (140) can count at least one product detected by the detector (120) based on the tracking result.

[0041] Although not illustrated in FIG. 1, the automatic checkout system (100) may further include a hand predictor that predicts at least one hand region from each of a plurality of frame images using a hand detection model. In this case, the tracker (130) may identify a product held in the hand among at least one detected product based on the predicted hand region and proceed with tracking the product held in the hand.

[0042] The automatic checkout system (100) according to the embodiment of FIG. 1 can load and process multiple frame images in real time according to a synchronization method. Alternatively, the automatic checkout system (100) according to another embodiment can load multiple frame images according to a predetermined batch size according to an asynchronous method, store them in the queue of each component, and then process them. Specifically, the frame loader (110) loads multiple frame images according to a predetermined batch size, the detector (120) sequentially stores the loaded multiple frame images in an input queue and detects at least one product from each of the multiple frame images according to the batch size, the tracker (130) sequentially stores the detected at least one product in a detection queue and tracks at least one product detected within the multiple frame images according to the batch size, and the counting unit (140) sequentially stores the tracking results in a counting queue and counts at least one product detected according to the batch size.

[0043] At least some components of an automatic checkout system (100) according to one embodiment of the present invention may be implemented by a memory including control software programmed to perform such functions and a computing device including a microprocessor that executes such software. In this case, each component of the automatic checkout system (100) according to the embodiment of FIG. 1 may be implemented independently by a microprocessor, or at least two may be implemented by a single microprocessor. For example, among the automatic checkout system (100) according to one embodiment of the present invention, at least one of the frame loader (110), hand predictor, tracker (130), or counter (140) may be implemented by a CPU, while the object detector (120), which has a high amount of computation, may be implemented by a GPU.

[0044] Up to this point, the configuration of an automatic checkout system (100) according to one embodiment of the present invention has been described. Below, a control method for an automatic checkout system (100) according to one embodiment of the present invention will be described.

[0045] FIG. 2 is a flowchart of a control method for an automatic checkout system according to an embodiment of the present invention, and FIG. 3 is a schematic diagram illustrating the operation of an automatic checkout system according to the control method of FIG. 2. FIG. 3 is based on the premise that the automatic checkout system operates according to an asynchronous method.

[0046] Referring to FIG. 2, first, an automatic checkout system (100) according to one embodiment of the present invention can sequentially load a plurality of frame images (S210). As shown in FIG. 3, a frame loader (110) of an automatic checkout system (100) according to one embodiment of the present invention can sequentially load a plurality of frame images of product combinations on the checkout counter obtained by a camera photographing the checkout counter. At this time, the frame loader (110) can also load the camera settings.

[0047] As previously explained, the frame loader (110) can load multiple frame images according to a predetermined batch size. In the case of an asynchronous method, the frame loader (110) can sequentially store the multiple frame images loaded according to the batch size in an input queue.

[0048] Next, an automatic checkout system (100) according to one embodiment of the present invention can detect at least one product from each of a plurality of frame images using a detection model (S220). Prior to this, the automatic checkout system (100) according to one embodiment of the present invention may perform the operation of directly creating and storing a detection model or receiving and storing it from an external source.

[0049] Below, a method for generating a detection model used in the control method of FIG. 2 is explained with reference to FIG. 4 and 5.

[0050] FIG. 4 is a schematic diagram illustrating a method for generating a training image used for learning a detection model according to an embodiment of the present invention, and FIG. 5 is a diagram illustrating a training image generated according to the method of FIG. 4.

[0051] As mentioned above, it may be impossible to train a detection model to handle all possible combinations of products. Therefore, in training a detection model according to one embodiment of the present invention, sample images of each product (e.g., advertising / promotional videos, etc.) captured in a controlled environment may be collected, and then training images may be generated using the sample images.

[0052] Referring to Fig. 4, the set of goods P={p i Given}, first, the set of individual sample images S = {(I s , M s , y s )|y s {∈ P} can be collected. Here, I s means a single product video, and M s represents a binary mask where the pixels corresponding to the product are colored white, and y s may represent the ID or category of the product. One embodiment of the present invention is a set S of single product sample images. aic You can use the Retail Dataset (retail store related dataset section) of the 2022 AI City Challenge as = <116,500 images; 116 ids>. In this case, sample images can be generated from the 3D scanning results of each product.

[0053] S aicIn this case, a product sample image can be randomly selected and placed on a checkout background with randomly set position, occlusion, clutter value, and lighting environment to generate a checkout image as a training image. The set of checkout images G={gi} generated in this way can be constructed as shown in Equation 1.

[0054]

[0055] Here, N={( I n , M n , y n )}∈S means a subset consisting of n randomly selected sample images, and (a, s, g, c) refers to each I n The parameters applied to can be indicated. In this case, the values ​​of the parameters according to one embodiment may be randomly selected within the range specified in Table 1, but are not limited thereto.

[0056]

[0057] In particular, training images can be generated by manipulating the parameters of each of multiple sample images to which a binary mask has been applied. Specifically, to remove the background of a sample image, I n and M n An "AND" bitwise operation can be applied between them. Next, parameters (a, s, g) can be manipulated (rotation, scaling, gamma adjustment) for each image segment. After manipulating the sample images, a mask can be obtained by applying a Threshold Invert to each image segment. The mask thus obtained can be randomly placed on the actual checkout image acquired as the background image, such that the intersection over union (IoU) value of the bounding boxes does not exceed a messiness value c. Through this process, the training image can be generated by ensuring that the degree of overlap between products in each of the multiple sample images is placed within a predetermined range of the messiness value.

[0058] Figure 5 is a diagram illustrating training videos with different disorder levels applied. The disorder level increases from the far left to the right, and a higher disorder level may mean that the detection difficulty increases.

[0059] A detection model can be generated by training a plurality of training images generated through the above method. A detection model according to one embodiment of the present invention can be generated by training a Scaled YOLOv4 model, but is not limited thereto.

[0060] To fine-tune a pre-trained MS COCO model, a dataset according to one embodiment of the present invention may be divided and used in a standard ratio of 80% to 20% between a training dataset and a validation dataset. Additionally, since parameter manipulation is applied to the generation of computational images, the training images may undergo two editing processes as a result. Through this, overfitting in the detection model is prevented, and the model can be trained to be applicable to various scales. The detection model is trained by Stochastic Gradient Descent (SGD), and the weights determined to be most optimal for the validation dataset may be selected for inference.

[0061] The detection model generated in this way may be stored in advance in a storage unit (150) inside the automatic checkout system (100) and provided when needed, or may be received directly from a device, server, or cloud outside the automatic checkout system (100).

[0062] Referring again to FIG. 2, the detection unit of the automatic checkout system (100) according to one embodiment of the present invention can detect a bounding box as at least one product from each of a plurality of frame images using a detection model generated by the method of FIG. 4.

[0063] At this time, the detector (120) may also be configured to correspond to the batched input / output format. Additionally, at least one product detected according to the batch size may be sequentially stored in the detection queue.

[0064] Meanwhile, to recognize a product held in the hand, an automatic checkout system (100) according to one embodiment of the present invention can localize a customer's hand within a loaded frame image. Specifically, the hand predictor of the automatic checkout system (100) according to one embodiment of the present invention may use a hand detection model generated by extracting an index consisting of 21 points from each of the two hands.

[0065] FIG. 6 is a diagram illustrating a hand detection model used by a hand predictor according to one embodiment of the present invention.

[0066] Referring to Fig. 6, the hand detection model can be implemented through a MediaPipe framework in which multiple artificial intelligence models work together. It may include a hand detection model that finds and outputs bounding boxes of palms present throughout the frame video, and a hand indicator extraction model that generates a high-performance 3D structure connected by multiple points from the palm presence regions found by the hand detection model.

[0067] These hand predictors operate in parallel with detectors (120), and the localization result information obtained from each can all be used for tracking products. Through this, an automatic checkout system (100) according to one embodiment of the present invention can distinguish and track products held in the hand and products that are not.

[0068] Next, an automatic checkout system (100) according to one embodiment of the present invention can track at least one product detected by a detector (120) within a plurality of frame images (S230). To this end, first, the tracker (130) of the automatic checkout system (100) according to one embodiment of the present invention can assign a unique ID to the product when the detected product enters the ROI within the frame image. This unique ID is maintained while the product exists within the visible area of ​​the camera and may indicate that a newly detected product is currently being tracked. Through this, a single product may not be identified redundantly.

[0069] When bounding boxes are input according to detection, the detector (120) can use a SORT method for tracking multiple products. When a camera is installed on a checkout counter with the lens facing downward, memory usage efficiency and overall processing speed can be increased by using a simple tracking method such as SORT.

[0070] The SORT method can use a Kalman filter for motion prediction and a Hungarian algorithm for track assignment. The Kalman filter can be used to reset the tracking state for unmatched detection results to switch the tracking state to a new target, while matched tracking results can be used to update the tracking state of existing targets.

[0071] In this case, the state space for each target can be defined by dimensional state variables (u, v, s, r, u', v', s'). Here, u and v represent the horizontal and vertical 2D pixel coordinates of the target center, and s and r represent the bounding box area and aspect ratio, respectively. A standard Kalman filter in a uniform motion and linear observation model can designate track segment k as a target, which can be used for counting in the next step. When assigning a new detection result to an existing target, the estimation of the bounding box shape for each target can be performed by predicting its unique position in the current frame. Subsequently, an assignment cost matrix is ​​calculated, which can be determined by the IoU between the actual detected bounding box and all bounding box prediction results for that target. The IoU distance, used as a criterion for evaluating the cost of consumption for optimal matching, can be obtained from the SORT when fast and efficient performance was demonstrated while used in parallel with the Kalman filter. In other words, any detection D i Regarding which track piece T j The cost when matched can be calculated through IoU. In this case, the optimal solution can be obtained using the Hungarian algorithm.

[0072] At this time, the tracker (130) may also be configured to correspond to the batched input / output format. Additionally, tracking results according to the batch size may also be sequentially stored in a counting queue.

[0073] Finally, an automatic checkout system (100) according to one embodiment of the present invention can count at least one product detected by a detector (120) based on a tracking result (S240). Specifically, the counting device (140) of the automatic checkout system (100) according to one embodiment of the present invention counts the product at the time the product appears in the ROI according to the tracking result, and may not count duplicates while the tracking is in progress.

[0074] In the case of an asynchronous method, the counting device (140) of the automatic checkout system (100) according to one embodiment of the present invention counts products by determining that the condition for counting whether the unique ID of a product is first verified is satisfied, and if the unique ID of a counted product is verified in a subsequent counting queue, the condition for deletion is satisfied, and the corresponding queue can be deleted without counting products.

[0075] FIG. 7 is a diagram for explaining the control method of an automatic checkout system according to one embodiment of the present invention, divided into a synchronous method and an asynchronous method.

[0076] When an automatic checkout system (100) according to one embodiment of the present invention operates in a synchronous manner, the automatic checkout system (100) can detect / track / count products by loading frame images periodically acquired by a camera in real time.

[0077] On the other hand, when the automatic checkout system (100) according to one embodiment of the present invention is operated in an asynchronous manner, the automatic checkout system (100) is equipped with an input queue, a detection queue, and a counting queue to store data in the queues and process it according to the batch size.

[0078] Specifically, to increase the inference speed in an asynchronous manner, an automatic checkout system (100) according to one embodiment of the present invention may batch the entire system and adjust it to the queue size. For example, to maximize the utilization of GPU capabilities, an automatic checkout system (100) according to one embodiment of the present invention may reduce the input entering the detector (120) to the batch size. Additionally, an automatic checkout system (100) according to one embodiment of the present invention may control each component to process tasks in independent threads and may provide input and output queues between components.

[0079] When an RGB image batch I={Ii|t≤i≤t+B} captured at time [t, t+B] is input (or, I ∈ RB×C×H×W), the detector (120) results in a batch D={D i It can output |RP×F|t≤i≤t+B}. Here, B represents the batch size, and P i represents the number of detections in batch i, and F can represent a list of visual features [x1, y1, x2, y2, cls, conf], i.e., the x, y coordinates of the top-left point, the x, y coordinates of the bottom-right point, a unique ID, and a confidence score.

[0080] Since each detection result is integrated with a corresponding time step index (frame index), the results can be output in chronological order. As a result, the inference speed of the automatic checkout system (100) can be improved.

[0081] As such, according to one embodiment of the present invention, the accuracy of product detection, tracking, and counting can be improved. Specifically, the performance of an automatic checkout system can be enhanced by using a learned detection model that automatically generates training images reflecting the specific characteristics of the automatic checkout environment. Furthermore, the processing speed of the automatic checkout system can be improved by batching the entire system and adjusting it to the queue size.

[0082] Meanwhile, each step included in the automatic checkout system control method according to the above-described embodiment may be implemented on a computer-readable recording medium that records a computer program programmed to perform such steps.

[0083] In addition, each step included in the automatic checkout system control method according to the above-described embodiment may be implemented by a computer program programmed to perform such steps.

[0084] The above description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential quality of the present invention. Accordingly, the embodiments disclosed in the present invention are intended to explain, not limit, the technical concept of the present invention, and the scope of the technical concept of the present invention is not limited by such embodiments. The scope of protection of the present invention shall be interpreted by the claims below, and all technical concepts within the equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0085] 100: Automatic checkout system 110: Frame Loader 120: Detector 130: Tracker 140: Counting machine 150: Storage section

Claims

Claim 1 delete Claim 2 An automatic checkout system comprising: a frame loader for sequentially loading multiple frame images; a detector for detecting at least one product from each of the multiple frame images using a detection model generated by learning a training image in which parameters of multiple sample images are manipulated; a tracker for tracking the at least one product detected within the multiple frame images; and a counting unit for counting the at least one product detected based on the result of tracking the product, wherein the parameters include at least one of the number of products included in each of the multiple sample images, the rotation angle of the sample image, the enlargement or reduction ratio of the sample image, or the gamma adjustment value of the sample image. Claim 3 In claim 2, the training image is an automatic calculator system generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied. Claim 4 An automatic checkout system comprising: a frame loader that sequentially loads a plurality of frame images; a detector that detects at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images are manipulated; a tracker that tracks the at least one product detected within the plurality of frame images; and a counting unit that counts the at least one product detected based on the result of tracking the product, wherein the training image is generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied, and the plurality of sample images to which the binary mask is applied are placed on a checkout image, wherein the degree of overlap between products of each of the plurality of sample images is placed within a predetermined range of messy values. Claim 5 An automatic checkout system according to claim 4, further comprising a hand predictor that predicts at least one hand region from each of the plurality of frame images using a hand detection model. Claim 6 In claim 5, the tracker is an automatic checkout system that identifies a product held in the hand among at least one detected product based on the predicted hand area. Claim 7 An automatic checkout system comprising: a frame loader that sequentially loads a plurality of frame images; a detector that detects at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images are manipulated; a tracker that tracks the at least one product detected within the plurality of frame images; and a counting unit that counts the at least one product detected based on the result of tracking the product, wherein the detector sequentially stores the loaded plurality of frame images in an input queue to detect the at least one product from each of the plurality of frame images, the tracker sequentially stores the at least one product detected within the plurality of frame images to track the at least one product detected within the plurality of frame images, and the counting unit sequentially stores the result of tracking the product in a counting queue to count the at least one product detected. Claim 8 An automatic checkout system comprising: a frame loader that sequentially loads a plurality of frame images; a detector that detects at least one product from each of the plurality of frame images using a detection model generated by learning a training image in which parameters of a plurality of sample images have been manipulated; a tracker that tracks the at least one product detected within the plurality of frame images; and a counting device that counts the at least one product detected based on the result of tracking the product, wherein the frame loader loads the plurality of frame images according to a predetermined batch size. Claim 9 delete Claim 10 A control method for an automatic checkout system, comprising: a step in which a frame loading device of the automatic checkout system sequentially loads a plurality of frame images; a step in which a detector of the automatic checkout system detects at least one product from each of the plurality of frame images using a detection model generated by learning a learning image in which parameters of a plurality of sample images are manipulated; a step in which a tracker of the automatic checkout system tracks the at least one product detected within the plurality of frame images; and a step in which a counting device of the automatic checkout system counts the at least one product detected based on the result of tracking the product, wherein the parameters include at least one of the number of products included in each of the plurality of sample images, the rotation angle of the sample image, the enlargement or reduction ratio of the sample image, or the gamma adjustment value of the sample image. Claim 11 In claim 10, the control method of an automatic calculator system wherein the learning image is generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied. Claim 12 A control method for an automatic checkout system, comprising: a step in which a frame loading device of the automatic checkout system sequentially loads a plurality of frame images; a step in which a detector of the automatic checkout system detects at least one product from each of the plurality of frame images using a detection model generated by learning a learning image in which parameters of a plurality of sample images are manipulated; a step in which a tracker of the automatic checkout system tracks the at least one product detected within the plurality of frame images; and a step in which a counting device of the automatic checkout system counts the at least one product detected based on the result of tracking the product, wherein the learning image is generated by manipulating the parameters of each of the plurality of sample images to which a binary mask is applied, and the plurality of sample images to which the binary mask is applied are placed in a checkout image, wherein the degree of overlap between products of each of the plurality of sample images is placed within a predetermined range of messy values. Claim 13 A control method for an automatic checkout system according to claim 12, further comprising the step of a hand predictor of the automatic checkout system predicting at least one hand region from each of the plurality of frame images using a hand detection model. Claim 14 In claim 13, the step of tracking at least one detected product is a control method for an automatic checkout system that identifies a product held in the hand among at least one detected product based on the predicted hand area. Claim 15 A control method for an automatic checkout system, comprising: a step in which a frame loading device of the automatic checkout system sequentially loads a plurality of frame images; a step in which a detector of the automatic checkout system detects at least one product from each of the plurality of frame images using a detection model generated by learning a learning image in which parameters of a plurality of sample images are manipulated; a step in which a tracker of the automatic checkout system tracks the at least one product detected within the plurality of frame images; and a step in which a counting device of the automatic checkout system counts the at least one product detected based on the result of tracking the product, wherein the step of detecting the at least one product is to sequentially store the loaded plurality of frame images in an input queue to detect the at least one product from each of the plurality of frame images, the step of tracking the at least one product detected is to sequentially store the at least one product detected within the plurality of frame images in a detection queue, and the step of counting the at least one product detected is to sequentially store the result of tracking the product in a counting queue to count the at least one product detected. Claim 16 In claim 15, the step of sequentially loading the plurality of frame images is a control method for an automatic checkout system that loads the plurality of frame images according to a predetermined batch size. Claim 17 delete Claim 18 delete

Images