Remote two-dimensional code scanner

Abstract

The invention, which belongs to the technical field of two-dimensional code identification, discloses a long-distance two-dimensional code scanner comprising a multi-modal image acquisition and control module, a scene understanding and target object identification module, a feature knowledge base, a two-dimensional code fine processing module, a decoding module and an optimization feedback module. Through cooperation of multiple modules, the remote two-dimensional code recognition features of context sensing, intelligent processing and self-learning capabilities are achieved, and the defects of remote recognition, complex environment adaptability, processing intelligence and the like in the prior art can be effectively overcome. An efficient and reliable two-dimensional code scanner is provided for various application scenes requiring long-distance and high-robustness two-dimensional code recognition.

Description

Technical Field

[0001] The present invention belongs to the technical field of two-dimensional code recognition, and in particular relates to a long-distance two-dimensional code scanner. Background Art

[0002] As a convenient and low-cost method for encoding and carrying information, QR codes (Quick Response Codes) have been widely used in numerous fields, including logistics, manufacturing, retail, payment, and ticketing. In the wave of industrial automation and intelligent upgrades, automatic QR code recognition through cameras has become a key step in improving efficiency and reducing manual operations.

[0003] In many practical application scenarios, automatic QR code recognition faces many challenges, especially in long-distance and complex environments. The specific challenges are as follows:

[0004] Traditional QR code scanning devices or ordinary cameras are usually designed for close range, making it difficult to clearly capture small QR codes at long distances (e.g., several to tens of meters away). Even when using cameras with optical zoom, recognition rates are often low due to issues such as inaccurate focus, jitter, and difficulty quickly locking onto the target.

[0005] Outdoor or industrial environments are complex and changeable. Factors such as strong light, backlight, shadows, low illumination at night, rainy and foggy weather, screen reflections, QR code surface contamination (such as dust, oil, scratches), and display refresh rates can seriously affect the quality of QR code images, causing ordinary recognition algorithms to fail.

[0006] Most existing QR code recognition solutions lack an understanding of the specific context and objects in which the QR code appears. They typically employ generic image processing and decoding strategies and are unable to optimize for the specific context of the QR code (e.g., whether it is printed on paper or displayed on an LED screen; on a fixed sign or on a moving package), which limits recognition efficiency.

[0007] Therefore, a long-distance QR code scanner is needed that can intelligently perceive the QR code context and adaptively adjust the image acquisition, processing and decoding strategies accordingly, so as to effectively address the above-mentioned defects and improve the success rate and speed of recognition. Summary of the Invention

[0008] In response to the above shortcomings, the present invention provides a long-distance QR code scanner, which includes a multimodal image acquisition and control module, a scene understanding and target object recognition module, a feature knowledge base, a QR code fine processing module, a decoding module, and an optimization feedback module;

[0009] The multimodal image acquisition and control module is used to capture high-quality images of the target QR code and its environment from a distance, and to provide precise visual guidance and synchronize image data management;

[0010] The scene understanding and target object recognition module receives the raw image data stream provided by the scene understanding and target object recognition module, analyzes the image content, identifies key objects, preliminarily locates the QR code, and generates a context label describing the specific scene in which the QR code is located;

[0011] The feature knowledge base receives contextual tags from the scene understanding and target object recognition modules for query, provides relevant prior knowledge to the QR code image refinement module and decoding module, and receives updates from the optimization feedback module;

[0012] The QR code fine processing module accurately tracks the target QR code image, enhances key structures, optimizes overall quality, and performs geometric correction based on the prior knowledge of the feature knowledge base, and outputs a finely processed QR code image.

[0013] The decoding module attempts to decode the refined QR code image, verifies the content, and performs multi-engine collaborative arbitration based on the prior knowledge of the feature knowledge base, and outputs the decoding result and confidence level.

[0014] The optimization feedback module is used to coordinate and control the multimodal image acquisition and control module, the QR code image refinement processing module and the decoding module, receive the decoding results of the decoding module, and update the feature knowledge base.

[0015] Furthermore, the multimodal image acquisition and control module includes:

[0016] A telephoto camera module for long-distance zoom capture of QR codes;

[0017] Wide-angle camera module for wide-angle recognition scenes;

[0018] A pan / tilt control system for carrying telephoto camera modules and wide-angle camera modules;

[0019] Auxiliary lighting unit for night lighting and image data management and synchronization unit for coordinating data acquisition of various cameras.

[0020] Furthermore, the scene understanding and target object recognition module includes:

[0021] An image preprocessing submodule that performs denoising, dehazing, and sharpening on the captured images;

[0022] Key object detection and classification submodule for identifying key objects in pre-processed images using deep learning models;

[0023] A detection and positioning submodule for locating preliminary potential QR code regions based on color, shape, and texture features, thereby outputting the coordinates and approximate sizes of one or more QR code candidate regions;

[0024] It also includes a context association and target locking submodule for determining the object to which the QR code is most likely attached based on the identified key object results and the preliminary positioning results of the QR code, and outputting a precise context label of the QR code for controlling the precise aiming of the telephoto camera module.

[0025] Furthermore, the two-dimensional code fine processing module includes:

[0026] A target ROI tracking and stabilization extraction submodule for receiving the preliminary position and context label of the QR code provided by the scene understanding and target object recognition module and outputting the image frame of the multimodal image acquisition and control module as a high-quality ROI image;

[0027] A predictive key structure location and enhancement submodule that receives high-quality ROI images and prior knowledge from a feature knowledge base and outputs clear key structure point information;

[0028] The environment and material adaptive enhancement submodule is used to adaptively improve the overall quality of the high-quality ROI image based on the surface characteristics and environmental challenge information of the QR code provided by the feature knowledge base, and output the enhanced ROI image as an overall enhanced ROI image;

[0029] It also includes a precise geometric correction and standardization submodule for performing precise geometric transformation on the two-dimensional code image according to the key structure point information and the enhanced ROI image, and outputting the result as a standardized two-dimensional code image.

[0030] Furthermore, the decoding module includes:

[0031] A priority parameterized decoding attempt submodule for receiving the image output by the QR code image refinement processing module and the prior knowledge of decoding parameters provided by the feature knowledge base, and outputting preliminary decoding data;

[0032] A data content and structure conformity verification submodule for verifying the validity and reliability of the preliminary decoded data based on the prior knowledge of the feature knowledge base;

[0033] The multi-decoding engine coordination and arbitration submodule is used to process the image from the QR code image refinement processing module again when the priority parameterized decoding attempt submodule fails to decode, and refer to the feature knowledge base for arbitration. The decoding result is output to the data content and structure conformity verification submodule for verification, and finally serves as the evaluation basis for the decoding quality and confidence comprehensive evaluation submodule;

[0034] It also includes a decoding quality and confidence comprehensive evaluation submodule that is responsible for comprehensively evaluating the multi-source information in the entire decoding process and outputting the final decoding result and a confidence score indicating its reliability.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. Through a high-resolution telephoto camera, a pan-tilt system, and a target ROI tracking and stable extraction submodule, it can accurately capture and stably track the target QR code from a distance, solving the problems of "invisibility" and "unstable vision" of traditional devices, and making up for the shortcomings of traditional scanning devices' short working distance and the poor long-distance imaging quality of ordinary cameras;

[0037] 2. By analyzing the environment through the scene understanding and object recognition modules and combining them with context-specific prior knowledge from the feature knowledge base, the environment and material adaptive enhancement submodule intelligently selects and applies the optimal image enhancement strategy (such as de-glare, HDR, dehazing, and contrast enhancement) to effectively address common environmental disturbances such as lighting changes, screen reflections, and surface contamination.

[0038] 3. By identifying the specific context of the QR code (e.g., "weighbridge instrument panel QR code," "equipment nameplate QR code") and obtaining prior knowledge of the typical version, fault tolerance level, data format, and common challenges of the QR code in that context from the feature knowledge base, the image processing and decoding process is made more guided, thereby improving efficiency and accuracy and reducing invalid attempts.

[0039] 4. The predictive key structure location and enhancement submodule can proactively search for and enhance key structural points blurred by local interference based on the predicted QR code version. The precise geometric correction and normalization submodule can accurately correct QR code deformation caused by shooting angle or surface curvature, reducing the reliance on perfect QR code printing / display quality and improving the tolerance and recognition ability of QR codes with certain degrees of damage, deformation, or unclear key parts.

[0040] 5. Through the knowledge base and model self-learning update sub-module, the system can learn from successful and failed recognition cases and continuously optimize the scene understanding model, image processing strategy and prior knowledge in the feature knowledge base. DETAILED DESCRIPTION

[0041] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] This embodiment provides a long-distance two-dimensional code scanner, including a multimodal image acquisition and control module, a scene understanding and target object recognition module, a feature knowledge base, a two-dimensional code fine processing module, a decoding module, and an optimization feedback module.

[0043] The multimodal image acquisition and control module is composed of a telephoto camera module, a wide-angle camera module, a pan-tilt system, an auxiliary lighting unit, and an image data management and synchronization unit. The telephoto camera module is used for long-distance optical zoom to capture clear images of the QR code on the weighing system and its surrounding environment. The wide-angle camera module provides a broad scene field of view. The pan-tilt system (i.e., the traditional pan-tilt system) is used to carry the telephoto camera module and the wide-angle camera module, which can assist in quickly locating the approximate position of the vehicle and the weighing scale, and provide more comprehensive contextual information (such as weather and overall lighting) for the scene understanding module. It can withstand outdoor environments and achieve stability and Precise horizontal, pitch and zoom control is required to track and aim at the target QR code. The auxiliary lighting unit is composed of LED lights and light sensors. Under low-contrast conditions such as night, shadows or backlight, it provides uniform and controllable fill light (such as infrared or white light LEI array) to the target area to reduce reflections and improve the quality of the QR code image. Finally, the image data management and synchronization unit is responsible for coordinating the data acquisition of each camera, ensuring the synchronization of the image frame timestamps, managing the transmission and compression (if necessary) of the data stream, and distributing the image stream with metadata (such as shooting parameters and gimbal position) to the subsequent QR code image refinement processing module.

[0044] In this embodiment, the camera in the telephoto camera module is a high-resolution telephoto camera with a sensor resolution of ≥12MP and an optical zoom capability of ≥20x. It has fast and accurate autofocus (to adapt to vehicle movement or vibration) and automatic exposure (to adapt to changes in outdoor light), and supports remote control of focus and aperture; the camera in the wide-angle camera module is a wide-angle environmental perception camera with a sensor resolution of ≥5MP and a wide dynamic range (WDR) to cope with high-contrast scenes; in order to ensure the life of outdoor use, the pan-tilt system needs to have a protection level ≥IP66, a rotation accuracy of ≤0.1 degrees, wind resistance, and preset position and cruise scanning functions, while the brightness is adjustable and can be linked with the camera exposure.

[0045] The scene understanding and target object recognition module includes an image preprocessing submodule, a key object detection and classification submodule, a QR code preliminary detection and positioning submodule, and a context association and target locking submodule. The image preprocessing submodule is used to perform denoising (such as Gaussian filtering), defogging (for outdoor environments), sharpening, and WDR processing on the input image, which has the function of improving image quality and facilitating subsequent recognition. The key object detection and classification submodule uses a targeted training deep learning model (such as the YOLO series, EfficientDet, or FasterR-CNN) to identify key objects in the scene, such as "weighing platform", "weighing instrument / display", and "vehicle (specific parts such as the front of the vehicle and license plate area)". , "operation booth", "signboard", etc., thereby outputting the object carrying the QR code (such as the weighing platform); the QR code preliminary detection and positioning submodule can use an efficient QR code detection algorithm (which can combine color, shape, and texture features) in wide-angle images or low-magnification telephoto images to quickly scan and roughly locate potential QR code areas, thereby outputting the coordinates and approximate sizes of one or more QR code candidate areas; the context association and target locking submodule is used to integrate the object recognition results with the QR code preliminary positioning results, determine the object to which the QR code is most likely to be attached (such as "the QR code on the weighing instrument display", "the QR code on the operating column next to the weighing scale"), and output the precise context label of the target QR code, which is used to guide the precise aiming of the telephoto lens.

[0046] It should be noted that the contextual label of the QR code is stored in the feature knowledge base, which records the characteristics of the QR code in various scenarios and attached objects (i.e., context) as well as the characteristics that the QR code usually has (such as the expected content format and structure description). When the system recognizes the specific context corresponding to the QR code, it will query this knowledge base to obtain relevant prior knowledge to guide the subsequent image processing and decoding process (i.e., the typical version information corresponding to the recognized QR code), and the feature knowledge base can be dynamically updated (manual input and online resource acquisition and entry).

[0047] The QR code fine processing module includes the target ROI tracking and stable extraction submodule, the predictive key structure positioning and enhancement submodule, the environment and material adaptive enhancement submodule, and the precise geometric correction and standardization submodule, as follows:

[0048] The target ROI tracking and stable extraction submodule is used to receive the context label and approximate location of the target QR code, control the telephoto camera to accurately align, use image tracking algorithms (such as KCF and CSRT) to stably track the QR code area in consecutive frames, and extract a high-quality region of interest (ROI) image containing the complete QR code, minimizing motion blur and geometric deformation.

[0049] Based on information about typical QR code versions provided by the feature knowledge base, the predictive key structure localization and enhancement submodule estimates the number and approximate layout of key structural elements (such as the positioning pattern (three large squares), the correction pattern (small squares), and the timing pattern distributed along the edges of rows and columns) that the QR code should contain. It then actively searches for these key structures in the extracted ROI image. It is important to note that when the precise boundaries and center points of these key structures are difficult to accurately identify due to localized, subtle interference (such as a slight reflective spot overlapping the edge of a positioning pattern, a small area of ​​stain obscuring a correction pattern, or low contrast making a small timing pattern difficult to distinguish from the background), this submodule uses highly targeted image processing methods (such as contrast stretching for very small areas, edge sharpening, sub-pixel feature point extraction algorithms, or model-based small-scale inpainting) to enhance the clarity and recognizability of these specific structural areas, ensuring their accurate localization. If some key structural points are difficult to identify directly due to slight damage or blur, this submodule will also attempt to repair or interpolate them based on their known geometric relationships in the QR code and surrounding pixel information, providing the most accurate and reliable key structural reference points possible for the precision geometry correction and standardization submodule;

[0050] The Environment and Material Adaptive Enhancement submodule intelligently selects and applies the most appropriate image enhancement strategy based on the overall surface characteristics of the QR code (e.g., "glass panels are prone to large-scale specular reflections") obtained from the feature knowledge base and common environmental challenges that affect the overall image quality of the QR code in this scenario (e.g., "strong outdoor sunlight causes overexposure or severe lack of contrast across the entire QR code area," "lack of ambient light causes an overall dark and noisy image," and "fog or heavy dust in the air causes overall blurriness of the image"). This improves the overall quality of the entire QR code region of interest (ROI), for example:

[0051] A. If the feature knowledge base indicates that the current surface material (such as glass) is prone to large-scale reflections or glare under current lighting conditions (such as direct sunlight) that affect the readability of the entire QR code, the image de-reflection algorithm is enabled, or (if supported by the hardware) the polarization filter is adjusted to suppress this global interference;

[0052] B. If the scene has a strong overall contrast between light and dark, or if the overall lighting is insufficient, resulting in underexposure, high dynamic range (HDR) image synthesis and global or local adaptive contrast enhancement techniques (such as the CLAHE algorithm) will be used to improve the dynamic range and visual readability of the entire QR code image.

[0053] C. If the QR code image is judged to be blurred or low-contrast due to dust, fog, or screen aging, appropriate image sharpening processing (such as unsharp masking), dehazing algorithms, or noise suppression may be performed;

[0054] The purpose of this submodule is to improve the overall signal-to-noise ratio, contrast, and clarity of the QR code image, reduce global interference, and provide a higher-quality input image for the subsequent decoding module to successfully perform binarization and pattern recognition.

[0055] The precise geometric correction and standardization submodule uses the key structural points (especially the three large positioning patterns and possible correction patterns) that are clarified and accurately located in the predictive key structure positioning and enhancement submodule to perform high-precision geometric transformation operations (such as affine transformation or perspective transformation). Its purpose is to correct and restore the deformed QR code image due to the tilted shooting angle or the curved surface of the object to a standard plane image as if it were taken vertically from directly above. The ultimate goal is to make each module of the QR code appear as a regular square as much as possible and arrange it on a neat grid.

[0056] It should be explained that the predictive key structure positioning and enhancement submodule and the environment and material adaptive enhancement submodule seem to have similar functions, but the predictive key structure positioning and enhancement submodule focuses more on the precision of "points" and "structures", and performs fine positioning and repair of the "skeleton" (key structure) of the QR code to ensure the accuracy of its geometric information, while the environment and material adaptive enhancement submodule focuses more on the overall optimization of the "surface", and performs overall conditioning and beautification of the "skin" (the entire image area) of the QR code to enhance the overall visual effect and readability.

[0057] The decoding module includes a priority parameterized decoding attempt submodule, a data content and structure compliance verification submodule, a multi-decoding engine collaboration and arbitration submodule, and a decoding quality and confidence comprehensive evaluation submodule, as follows:

[0058] The main QR code decoding engine built into the priority parameterized decoding attempt submodule (such as the optimized ZXing or ZBar library) will first attempt to decode using the typical QR code version and typical fault tolerance parameter combination obtained from the feature knowledge base and predicted for the current context. If the decoding attempt fails, other versions and fault tolerance parameter combinations will be automatically tried in a preset probability order or an order of effectiveness based on experience.

[0059] After the QR code is initially decoded to obtain the original data string, the data content and structure conformity verification submodule will perform strict verification based on the expected content format and structure description of this type of QR code in the feature knowledge base (for example, the feature knowledge base may describe what character rules and length the weighing ticket number should conform to, and what standard date and time format the timestamp should be in). If the decoded weighing ticket number data does not conform to the preset rule of starting with "WB" and followed by 12 digits, the reliability of the decoding result will be reduced, and it may even be directly judged as an incorrect decoding and discarded;

[0060] The multi-decoding engine collaboration and arbitration submodule can integrate multiple different open source or commercial QR code decoding libraries. When the main decoding engine fails to decode successfully, or the decoded result conforms to the format but its internal evaluation quality is not high, the system can activate one or more backup decoding engines to try to decode the same pre-processed QR code image. If multiple engines give results, the system will compare the consistency of their results, the decoding confidence reported by each, and then make a comprehensive judgment (i.e. arbitration) based on the format requirements in the feature knowledge base to select the most reliable decoding result.

[0061] The decoding quality and confidence comprehensive assessment submodule integrates various information collected throughout the decoding process, such as the clarity of key structural points located in the predictive key structure location and enhancement submodule, the amount of error correction code actually used during the decoding process (how much error correction capacity is used indicates how much damage the original image has suffered), whether the decoded data content perfectly matches the format predicted in the feature knowledge base, and (if the multi-decoding engine collaboration and arbitration submodule is used) whether the results of multiple decoding engines are consistent. Ultimately, it provides a quantitative assessment score that represents the reliability of the decoding operation.

[0062] The optimization feedback module includes the main controller, closed-loop feedback adjustment submodule, knowledge base and model self-learning update submodule, and human-computer interaction and remote operation and maintenance interface, as follows:

[0063] The main controller is responsible for coordinating the orderly work of each module, managing the task scheduling process and state transition within the system, and handling various interrupt requests and abnormal situations that may occur;

[0064] The triggering condition of the closed-loop feedback adjustment submodule is: after the entire recognition process is completed, if the QR code fails to be successfully decoded, or the confidence score of the decoding result is lower than the preset reliability threshold, this submodule will be started. This submodule will analyze at which stage the problem occurred and the possible reasons (for example: the scene understanding part of the scene understanding and target object recognition module fails to accurately identify the object carrying the QR code, the image processing of the QR code fine processing module fails to clearly locate the key structural points of the QR code, or the decoder of the decoding module times out after all attempts). It then intelligently adjusts certain parameters or strategies and initiates a new round of recognition attempts until the QR code is successfully recognized. For example:

[0065] Instruct the multimodal image acquisition and control module: fine-tune the pan / tilt camera's shooting angle, change the lens's focal length or focus point, or enhance / change the intensity and angle of auxiliary lighting;

[0066] Instruct the QR code fine processing module: Try using different image enhancement parameter combinations or switch to another geometric correction algorithm;

[0067] Instruction decoding module: instructs the decoding engine to try a wider range of QR code versions and fault tolerance level parameters;

[0068] The knowledge base and model self-learning update submodule is used to continuously collect case data for each QR code recognition (regardless of success or failure). This data includes: the original captured image, the recognized scene context information, the parameters used in each processing step, the true parameters of the QR code (if they can be confirmed later in some way, such as manual verification or acquisition from an associated system), the final decoding result, etc. Subsequently, through supervised learning methods, the object recognition model in the scene understanding and target object recognition module and the AI ​​model used to locate the key structure of the QR code in the QR code fine processing module (image refinement) can be regularly retrained or fine-tuned to improve their accuracy and robustness. In addition, the data can be analyzed to update and optimize the various contextual feature information stored in the feature knowledge base.

[0069] The human-computer interaction and remote operation and maintenance interface provides a user interface (which can be a local display or a remote web interface) for real-time display of the video stream captured by the camera, the intermediate process of the system's QR code recognition, the final decoding result, and the current operating status of the entire system; it can also implement manual intervention, such as remote manual control of pan-tilt alignment, system calibration or calibration operations, review of detailed operation logs for problem troubleshooting, management and editing of content in the feature knowledge base, etc.

[0070] S11. When in use, the wide-angle camera module needs to continuously monitor a designated location such as the entrance to the weighbridge so that it can continuously and preliminarily analyze the wide-angle image to determine whether there is a vehicle entering;

[0071] S12. When the multimodal image acquisition and control module detects the vehicle entering, the wide-angle camera module detects the vehicle entering the weighbridge area. The pan / tilt system roughly adjusts the telephoto camera module toward the vehicle and the electronic scale based on the vehicle position information in the wide-angle image. The telephoto camera module begins capturing images of the vehicle and the electronic scale area. Simultaneously, the image data management and synchronization unit sends the image stream to the scene understanding and target object recognition module.

[0072] S21, the image preprocessing submodule processes the received image;

[0073] S22, the key object detection and classification submodule identifies "vehicle", "weighing platform", and "electronic scale display screen", and the QR code preliminary detection and positioning submodule roughly locates the QR code in the electronic scale display screen area;

[0074] S23, the context association and target locking submodule generates a context label: "dynamic weighing QR code on the electronic scale display screen", and sends this label and the approximate location information of the QR code to the QR code fine processing module, and at the same time sends the context label to the feature knowledge base for query;

[0075] S31. The feature knowledge base receives the context tag, queries and extracts typical features of the “dynamic weighing QR code on the electronic scale display” (such as typical version, fault tolerance level, data format, common environmental challenges such as screen reflection, etc.), and sends this prior knowledge to the QR code fine processing module and decoding module respectively;

[0076] S32, the target ROI tracking and stable extraction submodule accurately controls the telephoto camera to align and stably track the QR code according to the approximate position of the input QR code, and extracts the ROI;

[0077] S33, the predictive key structure positioning and enhancement submodule locates and enhances the key structure points of the QR code based on the typical version information obtained from the feature knowledge base;

[0078] S34, the environment and material adaptive enhancement submodule performs overall enhancement on the ROI image based on the environmental challenge information (such as screen reflection) obtained from the feature knowledge base;

[0079] S35, the precision geometric correction and standardization submodule uses the enhanced key structural points to perform geometric correction and standardization on the two-dimensional code image, and sends the processed high-quality two-dimensional code image to the decoding module;

[0080] S41, the priority parameterized decoding attempt submodule decodes the received QR code image using the typical version and fault tolerance level parameters obtained from the feature knowledge base;

[0081] S42, the data content and structure conformity verification submodule verifies the decoded data against the expected data format obtained from the feature knowledge base. If necessary, the multi-decoding engine collaborates with the arbitration submodule to perform auxiliary decoding and result selection;

[0082] S43, the decoding quality and confidence comprehensive evaluation submodule outputs the final decoding result (weighing information) and confidence score, and the decoding result and confidence score are sent to the optimization feedback module (and may also be directly output to the external business system);

[0083] S5. The main controller receives the decoding result and confidence level:

[0084] If the decoding is successful and the confidence level is high, the results are displayed through human-computer interaction and remote operation and maintenance interfaces, and the knowledge base and model self-learning update submodules are controlled to record this success case for subsequent knowledge base and model optimization.

[0085] If decoding fails or the confidence is low, the closed-loop feedback adjustment submodule is activated to analyze the cause of the failure and perform the following operations based on the cause of the failure:

[0086] 1) Send instructions to the multimodal image acquisition and control module to adjust shooting parameters (such as gimbal fine-tuning, zoom, and fill light);

[0087] 2) Send instructions to the QR code fine processing module to try different image enhancement parameters;

[0088] 3) Send instructions to the decoding module to try a wider range of decoding parameters;

[0089] 4) Return to step S21 or S31 and perform a new round of recognition attempts until success or the maximum number of attempts is reached;

[0090] At the same time, the knowledge base and model self-learning update submodules record failure cases for analysis and improvement.

[0091] It should be noted that the structure described in the present invention can be implemented in a variety of different forms and is not limited to the described embodiments. Any equivalent transformations made by ordinary technicians in this field using the contents of the present invention specification, or directly or indirectly applied to other related technical fields, such as the loading and unloading of other items, are included in the scope of protection of the present invention.

Claims

1. A long-distance two-dimensional code scanner, characterized by: It includes multimodal image acquisition and control module, scene understanding and target object recognition module, feature knowledge base, QR code fine processing module, decoding module and optimization feedback module; The multimodal image acquisition and control module is used to capture high-quality images of the target QR code and its environment from a distance, and to provide precise visual guidance and synchronize image data management; The scene understanding and target object recognition module receives the raw image data stream provided by the scene understanding and target object recognition module, analyzes the image content, identifies key objects, preliminarily locates the QR code, and generates a context label describing the specific scene in which the QR code is located; The feature knowledge base receives contextual tags from the scene understanding and target object recognition modules for query, provides relevant prior knowledge to the QR code image refinement module and decoding module, and receives updates from the optimization feedback module; The QR code fine processing module accurately tracks the target QR code image, enhances key structures, optimizes overall quality, and performs geometric correction based on the prior knowledge of the feature knowledge base, and outputs a finely processed QR code image. The decoding module attempts to decode the refined QR code image, verifies the content, and performs multi-engine collaborative arbitration based on the prior knowledge of the feature knowledge base, and outputs the decoding result and confidence level. The optimization feedback module is used to coordinate and control the multimodal image acquisition and control module, the QR code image refinement processing module and the decoding module, receive the decoding results of the decoding module, and update the feature knowledge base.

2. The long-distance two-dimensional code scanner according to claim 1, wherein: The multimodal image acquisition and control module includes: A telephoto camera module for long-distance zoom capture of QR codes; Wide-angle camera module for wide-angle recognition scenes; A pan / tilt control system for carrying telephoto camera modules and wide-angle camera modules; Auxiliary lighting unit for night lighting and image data management and synchronization unit for coordinating data acquisition of various cameras.

3. The long-distance two-dimensional code scanner according to claim 1, wherein: The scene understanding and target object recognition module includes: An image preprocessing submodule that performs denoising, dehazing, and sharpening on the captured images; Key object detection and classification submodule for identifying key objects in pre-processed images using deep learning models; A detection and positioning submodule for locating preliminary potential QR code regions based on color, shape, and texture features, thereby outputting the coordinates and approximate sizes of one or more QR code candidate regions; It also includes a context association and target locking submodule for determining the object to which the QR code is most likely attached based on the identified key object results and the preliminary positioning results of the QR code, and outputting a precise context label of the QR code for controlling the precise aiming of the telephoto camera module.

4. The long-distance two-dimensional code scanner according to claim 1, wherein: The two-dimensional code fine processing module includes: A target ROI tracking and stabilization extraction submodule for receiving the preliminary position and context label of the QR code provided by the scene understanding and target object recognition module and outputting the image frame of the multimodal image acquisition and control module as a high-quality ROI image; A predictive key structure location and enhancement submodule that receives high-quality ROI images and prior knowledge from a feature knowledge base and outputs clear key structure point information; The environment and material adaptive enhancement submodule is used to adaptively improve the overall quality of the high-quality ROI image based on the surface characteristics and environmental challenge information of the QR code provided by the feature knowledge base, and output the enhanced ROI image as an overall enhanced ROI image; It also includes a precise geometric correction and standardization submodule for performing precise geometric transformation on the two-dimensional code image according to the key structure point information and the enhanced ROI image, and outputting the result as a standardized two-dimensional code image.

5. The long-distance two-dimensional code scanner according to claim 1, wherein: The decoding module includes: A priority parameterized decoding attempt submodule for receiving the image output by the QR code image refinement processing module and the prior knowledge of decoding parameters provided by the feature knowledge base, and outputting preliminary decoding data; A data content and structure conformity verification submodule for verifying the validity and reliability of the preliminary decoded data based on the prior knowledge of the feature knowledge base; The multi-decoding engine coordination and arbitration submodule is used to process the image from the QR code image refinement processing module again when the priority parameterized decoding attempt submodule fails to decode, and refer to the feature knowledge base for arbitration. The decoding result is output to the data content and structure conformity verification submodule for verification, and finally serves as the evaluation basis for the decoding quality and confidence comprehensive evaluation submodule; It also includes a decoding quality and confidence comprehensive evaluation submodule that is responsible for comprehensively evaluating the multi-source information in the entire decoding process and outputting the final decoding result and a confidence score indicating its reliability.