AGV forklift pallet ultra-wide detection early warning system, method and equipment

Abstract

The invention discloses an AGV forklift pallet ultra-width detection early warning system, method and device, and aims to solve the problems of resource waste, misrecognition and low system reliability caused by indifference continuous acquisition and processing of image data in the prior art. The method or system comprises the following steps that firstly, the distance between the AGV forklift and a target object is detected; if the distance meets a first preset condition, collecting a two-dimensional digital image of the target area; identifying whether a target tray exists in the target area based on the image; if yes, positioning coordinates of two sides of the front surface of the tray and calculating the width size of the tray; and if the width size meets a second preset condition, early warning is triggered, and the current working process of the AGV forklift is stopped. By optimizing the detection logic and the data acquisition strategy, the redundant calculation load of an image processing system is remarkably reduced, the recognition reliability in a complex dynamic scene is improved, and meanwhile, the comprehensive requirements of AGV equipment for high efficiency, stability and energy conservation in a high-density storage environment are met.

Description

Technical Field

[0001] The present invention relates to the technical field of AGV forklift manufacturing, and in particular to an AGV forklift pallet over-width detection and warning system, method and equipment. Background Art

[0002] In automated logistics systems, AGVs (Automated Guided Vehicles) and forklifts are widely used for pallet handling and stacking. To ensure safety and efficiency during transportation, real-time inspection is required to ensure that pallet dimensions meet pre-set standards, particularly to prevent accidents such as collisions and jams caused by excessive pallet width. Visual recognition-based inspection solutions are becoming increasingly mainstream. These solutions continuously capture image data from the work area using cameras, employ image processing algorithms to calculate pallet width in real time, and trigger warnings.

[0003] However, this type of visual inspection system faces significant technical bottlenecks in practical applications: since AGV forklifts need to move and adjust their posture frequently during operation, their cameras often need to remain in continuous working state to capture potential targets. This indiscriminate way of collecting and processing image data causes the system to process a large amount of redundant information (such as empty areas, non-pallet objects, etc.), which not only causes a serious waste of computing resources, but also causes misidentification due to factors such as changes in ambient light and background interference. For example, when the AGV has not yet approached the target object, the system may mistakenly judge the edge features of stacked goods in the distance as the pallet boundary, thereby triggering an erroneous warning. Such problems directly increase the power consumption and hardware loss of the equipment, while reducing the overall reliability of the detection system, becoming a key obstacle to the intelligent upgrade of AGV forklifts in high-density warehousing scenarios. Summary of the Invention

[0004] The primary objective of this invention is to significantly reduce the redundant computational load of the image processing system while ensuring accurate detection of over-width pallets on AGV forklifts, while also improving recognition reliability in complex dynamic scenarios. Specifically, by optimizing detection logic and data acquisition strategies, this approach addresses the resource waste and false triggering issues inherent in existing technologies caused by indiscriminate, continuous operation, thereby meeting the comprehensive requirements for efficiency, stability, and energy efficiency of AGV equipment in high-density warehouse environments.

[0005] To achieve the above objectives, the first aspect of the present invention provides an AGV forklift pallet over-width detection and warning system, comprising: A distance measurement and condition judgment module is used to detect the distance between the AGV forklift and the target object and determine whether the distance meets a first preset condition; an image collector, connected to the distance measurement and condition judgment module, for collecting a two-dimensional digital image of the target area when the distance meets a first preset condition; an image recognition and positioning module, connected to the image collector and the distance measurement and condition judgment module respectively, for identifying whether a target pallet exists in the target area based on the two-dimensional digital image and the distance, and locating the coordinates of both sides of the front of the target pallet; A calculation module, connected to the image recognition and positioning module, for calculating the width of the target pallet based on the coordinates on both sides of the front side; The judgment execution module is connected to the calculation module and is used to judge whether the width dimension meets the second preset condition. If so, an early warning operation is performed; otherwise, the current working process of the AGV forklift is terminated.

[0006] Furthermore, the distance measurement and condition judgment module includes: Positioning unit, used to detect whether the AGV forklift has entered the target area; An ultrasonic transmitting unit, connected to the positioning unit, is used to transmit an ultrasonic pulse signal when detecting that the AGV forklift enters the target area; an ultrasonic receiving unit, connected to the ultrasonic transmitting unit, and configured to receive a reflected signal from a target object; The distance calculation unit is used to generate real-time distance data based on the time difference between the transmitted signal and the reflected signal.

[0007] Furthermore, the distance measurement and condition judgment module also includes: Cycle counting unit, used to record the number of consecutive measurements; A condition judgment unit is connected to the cycle counting unit and is used to confirm that the distance meets the first preset condition when the distance measurement values ​​are all lower than the preset distance threshold for three consecutive times.

[0008] Furthermore, the image recognition and positioning module includes: A dual-modal feature extraction unit, configured to extract edge gradient feature parameters from the two-dimensional digital image using a Canny edge detection algorithm and extract geometric feature parameters using an ORB feature matching algorithm; a feature fusion unit connected to the dual-modal feature extraction unit, configured to fuse the edge gradient feature parameter with the geometric feature parameter to generate a fused feature parameter; A topology matching unit is connected to the feature fusion unit and is used to determine the existence of the target pallet based on the matching results of the fused feature parameters and the preset database. The judgment conditions include detecting at least three raised features that meet the foot size threshold and the spatial arrangement of the raised feature parameters meeting the tripod pallet topology constraint.

[0009] Furthermore, the bimodal feature extraction unit includes: FAST corner detection subunit, used to detect corners in the image in real time when performing ORB feature matching; The BRIEF descriptor generation subunit is used to generate binary descriptors of corner points when performing ORB feature matching.

[0010] Furthermore, the image recognition and positioning module also includes: The Hough transform processing unit is used to establish a coordinate system based on the center point of the tripod tray's foot and fit the boundary line using the parametric equation ρ = xcosθ + ysinθ; where ρ represents the vertical distance from the line to the coordinate origin (polar diameter), and θ represents the angle between the line and the positive direction of the x-axis (polar angle); The edge merging unit is used to determine that two lines are the same physical edge when the polar diameter difference between them is ≤ 5 mm. ρ1 and ρ2 represent the vertical distances (polar diameters) between the two lines and the coordinate origin, respectively, and δ = 5 mm is the pallet thickness tolerance.

[0011] Furthermore, the calculation module includes: The original width calculation unit is used to calculate the original width value W based on the coordinates on both sides raw ; The error compensation unit is used to perform dynamic compensation calculations, and its formula is: , where ΔD=2cm is the preset absolute value of the ranging error, D ref is the real-time ranging threshold, α is the material compensation coefficient and α∈[0.1,0.3].

[0012] A second aspect of the present invention provides an AGV forklift pallet over-width detection and early warning method, comprising: Detect the distance between the AGV forklift and the target object; If the distance satisfies a first preset condition, acquiring a two-dimensional digital image of the target area; Based on the two-dimensional digital image of the target area, identifying whether a target pallet exists in the target area; If a target pallet is identified to exist in the target area, the coordinates of both sides of the front of the target pallet are located; Calculate the width of the target pallet based on the coordinates of both sides of the front side of the target pallet; If the width of the target pallet meets the second preset condition, an early warning is triggered, and the current workflow of the AGV forklift is terminated.

[0013] A third aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the AGV forklift pallet over-width detection and warning method.

[0014] In a fourth aspect of the present invention, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the AGV forklift pallet over-width detection and warning method are implemented.

[0015] The present invention provides an AGV forklift pallet overwidth detection and early warning method and system with at least the following beneficial effects: The present invention avoids redundant information processing caused by indiscriminate continuous work by collecting images only when the spacing meets the first preset condition, and significantly reduces the computational load of the image processing system; by optimizing the detection logic and data acquisition strategy, it reduces the impact of ambient light changes and background interference on the recognition results, and improves the recognition reliability in complex dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the structure of an AGV forklift pallet overwidth detection and warning system provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a distance measurement and condition judgment module provided in an embodiment of the present invention; Figure 3 A schematic diagram of a flow chart of an AGV forklift pallet overwidth detection and early warning method provided by an embodiment of the present invention; Figure 4 A schematic flow chart of a method for an AGV forklift to identify the presence of a target pallet in a target area provided by an embodiment of the present invention; Reference numerals: Distance measurement and condition judgment module-100, positioning unit-110, ultrasonic transmitting unit-120, ultrasonic receiving unit-130, distance calculation unit-140, image collector-200, image recognition and positioning module-300, calculation module-400, judgment execution module-500. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0018] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0019] Example 1 See also Figure 1 , Figure 1 The following is a schematic diagram of a pallet overwidth detection and warning system for an AGV forklift, provided as an example in this embodiment. The system comprises multiple functional modules working together to implement overwidth pallet detection and warning functions. The core components of the system include a distance measurement and condition determination module 100, an image acquisition device 200, an image recognition and positioning module 300, a calculation module 400, and a judgment execution module 500. Each module implements data exchange and control command transmission through standardized interfaces, specifically including: The distance measurement and condition judgment module 100 is used to detect the distance between the AGV forklift and the target object and determine whether the distance meets the first preset condition; An image collector 200, connected to the distance measurement and condition judgment module 100, is used to collect a two-dimensional digital image of the target area when the distance meets a first preset condition; The image recognition and positioning module 300 is connected to the image acquisition device 200 and the distance measurement and condition judgment module 100, respectively, and is used to identify whether there is a target pallet in the target area based on the two-dimensional digital image and the distance, and locate the coordinates of the front side of the target pallet; A calculation module 400, connected to the image recognition and positioning module 300, is used to calculate the width of the target pallet based on the coordinates of both sides of the front side; The judgment execution module 500 is connected to the calculation module 400 and is used to judge whether the width dimension meets the second preset condition. If so, an early warning operation is performed; otherwise, the current working process of the AGV forklift is terminated.

[0020] In this embodiment, the terms involved are explained as follows: Target object: refers to the pallet or other goods that the AGV forklift needs to detect and handle during operation.

[0021] Two-dimensional digital image: A two-dimensional plane image captured by a camera, used for image processing and recognition.

[0022] Target pallet: refers to the pallet whose width dimension needs to be detected, usually the object to be transported and stacked by AGV forklift.

[0023] The first preset condition: refers to the condition that the distance between the AGV forklift and the target object reaches a certain set threshold, triggering image acquisition.

[0024] The second preset condition: refers to the condition that the width of the target pallet exceeds a certain set threshold, triggering an early warning.

[0025] In this exemplary embodiment, the AGV forklift pallet overwidth detection and warning system achieves efficient and reliable detection and warning functions through the collaborative work of multiple modules. The specific implementation plan is as follows: The ranging and condition determination module 100 serves as the system's front-end sensing unit, integrating a laser ranging component and an ultrasonic submodule. For example, a pulsed laser sensor with a specific wavelength is mounted on the front end of the fork and scans the work area at a fixed frequency. Upon detecting that the AGV has entered a predefined area with a radius of several meters centered on the fork, the ultrasonic detection unit is immediately activated. This unit emits pulses at a specific frequency and calculates the real-time distance using the time-of-flight method. For example, when the detection value is less than or equal to a set threshold, the image acquisition process is triggered. This threshold can be dynamically adjusted based on vehicle model parameters.

[0026] The image collector 200 utilizes an industrial-grade imaging device, such as a global shutter CMOS sensor, offering megapixel resolution and high frame rate capture. Its field of view covers a specific area in front of the forks. Dynamic exposure control technology adapts to ambient light variations, automatically extending exposure time to milliseconds in low-light conditions. The collector operates only when the ranging module is triggered, capturing a specific width detection area centered on the fork centerline, effectively suppressing background interference.

[0027] The image recognition and positioning module 300 achieves precise recognition through a multi-stage algorithm. The preprocessing stage uses a Gaussian filter kernel of a specific size for noise reduction, combined with a dual-threshold Canny operator to extract contour features. The feature recognition layer uses a deep learning network to analyze the pallet texture and detect raised structures that meet geometric requirements, such as requiring the detection of at least three feature regions with projected areas exceeding a set value. The topology verification stage verifies the legitimacy of the target using spatial constraints, such as requiring that the spacing between feature points does not exceed a set ratio. Once the target is confirmed, a sub-pixel positioning algorithm is used to obtain boundary coordinates. Coordinate transformations are then performed using pre-calibrated parameters, such as using lens parameters for a specific focal length to eliminate perspective distortion.

[0028] Calculation module 400 performs dimensional calculations and error compensation based on the coordinate data. Basic calculation formulas are combined with dynamic compensation models. For example, by incorporating distance measurement error correction terms and material coefficients, differentiated compensation strategies are employed for wooden, plastic, and metal pallets. Compensation parameters are obtained through a table lookup method to ensure measurement accuracy under different working conditions.

[0029] The judgment and execution module 500 implements a hierarchical response control: when the calibrated width exceeds a safety threshold, it simultaneously triggers an audible and visual warning and an emergency stop mechanism. For example, a specific frequency alarm device is used to cut off power output within milliseconds via a high-speed bus, and event data is uploaded to the management system. The system response time from detection to execution is in the millisecond range, significantly reducing the risk of collision.

[0030] The system in this exemplary embodiment achieves comprehensive performance improvements through resource optimization strategies: a dynamic thresholding mechanism filters out most invalid images, significantly reducing single-frame processing time; multi-feature fusion recognition keeps the false positive rate low; and adaptive exposure and coordinate compensation technologies ensure detection accuracy in low-light environments. Extended function modules can integrate environmental perception units, for example, automatically adjusting detection parameters in areas with dense shelves or enabling HDR imaging mode in bright light conditions. A self-calibration mechanism continuously monitors detection quality and dynamically optimizes imaging parameters and recognition thresholds, forming a closed-loop control system.

[0031] This exemplary system utilizes a modular design to achieve precise control of the inspection process, significantly improving warehouse safety. In typical application scenarios, the system effectively identifies both standardized pallets and non-standard carriers, adapting to diverse operating environments through dynamic parameter adjustments. Experiments have shown that the system maintains high recognition accuracy and low false positive rates even in complex lighting conditions and when inspecting multiple pallet types, making it particularly suitable for high-density automated warehousing systems and intelligent logistics scenarios.

[0032] Example 2 See also Figure 1 and Figure 2 , Figure 2 The following is a schematic diagram of the structure of a distance measurement and condition judgment module 100, provided as an example for this embodiment. This system comprises multiple functional modules working together to implement over-width pallet detection and warning functions. The core components of the system include the distance measurement and condition judgment module 100, an image acquisition device 200, an image recognition and positioning module 300, a calculation module 400, and a judgment execution module 500. These modules implement data exchange and control command transmission through standardized interfaces, specifically including: The distance measurement and condition judgment module 100 is used to detect the distance between the AGV forklift and the target object and determine whether the distance meets the first preset condition; An image collector 200, connected to the distance measurement and condition judgment module 100, is used to collect a two-dimensional digital image of the target area when the distance meets a first preset condition; The image recognition and positioning module 300 is connected to the image acquisition device 200 and the distance measurement and condition judgment module 100, respectively, and is used to identify whether there is a target pallet in the target area based on the two-dimensional digital image and the distance, and locate the coordinates of the front side of the target pallet; A calculation module 400, connected to the image recognition and positioning module 300, is used to calculate the width of the target pallet based on the coordinates of both sides of the front side; The judgment execution module 500 is connected to the calculation module 400 and is used to judge whether the width dimension meets the second preset condition. If so, an early warning operation is performed; otherwise, the current working process of the AGV forklift is terminated.

[0033] The distance measurement and condition judgment module 100 includes: Positioning unit 110, used to detect whether the AGV forklift enters the target area; The ultrasonic transmitting unit 120 is connected to the positioning unit 110 and is used to transmit an ultrasonic pulse signal when detecting that the AGV forklift enters the target area; an ultrasonic receiving unit, connected to the ultrasonic transmitting unit 120, for receiving a reflected signal from a target object; The distance calculation unit 140 is configured to generate real-time distance data according to the time difference between the transmitted signal and the reflected signal.

[0034] In this embodiment, the terms involved in the embodiments of the present invention are explained as follows: Ultrasonic pulse signal: refers to a short-duration pulse signal emitted by the ultrasonic transmitting unit 120 and is used to measure the distance to a target object.

[0035] Reflected signal: refers to the signal reflected back after the ultrasonic pulse signal encounters the target object, which is used to calculate the time difference.

[0036] Based on the above embodiment 1, this embodiment uses ultrasonic pulse signals for distance detection, which has the following advantages: High precision and real-time performance: Ultrasonic ranging technology can quickly and accurately measure the distance to the target object and is suitable for real-time detection in dynamic environments.

[0037] Environmental adaptability: Ultrasonic ranging is less affected by changes in ambient light and can work stably under complex lighting conditions.

[0038] Cost-effectiveness: The ultrasonic transmitter unit 120 is low-cost and easy to integrate into the AGV forklift system, making it suitable for large-scale applications.

[0039] Reliability: By transmitting ultrasonic pulse signals and receiving reflected signals, it can effectively avoid false detection problems caused by visual interference or background noise.

[0040] In this embodiment, the distance measurement and condition determination module 100 dynamically detects the distance between the AGV and the target object through the coordinated operation of multiple units. This module first determines the AGV's operating area using the positioning unit 110, for example by using a laser navigation system to monitor the vehicle's position in real time. The subsequent distance measurement process is triggered when the AGV is detected entering a sector-shaped target area with a radius of 2 meters centered on the forklift's front end.

[0041] The ultrasonic transmitter unit 120 forms a linkage mechanism with the positioning unit 110, automatically activating upon receiving a zone entry signal. Installed at the front end of the AGV's forks, this unit uses a 40kHz ultrasonic transducer to emit detection signals at a periodic pulse frequency of 10Hz, ensuring real-time tracking of changes in the distance to the target object.

[0042] The ultrasonic receiving unit works in conjunction with the transmitting unit to continuously monitor echo signals reflected from the target object. This unit is equipped with a bandpass filter circuit, with filtering parameters set to a center frequency of 40kHz, for example, to effectively suppress ambient noise interference. Upon receiving a valid reflected signal, the receiving unit immediately records the signal's return time, providing a baseline time parameter for the distance calculation unit 140.

[0043] The distance calculation unit 140 uses the time-of-flight method to determine the real-time distance by precisely measuring the time difference Δt between the transmission and reception of ultrasonic pulses. For example, a high-precision timer is used to capture the time difference Δt. Combined with the speed of sound at room temperature of 343 meters per second, the actual distance is calculated using the formula D = (v × Δt) / 2. The calculation unit also includes a built-in threshold comparison function that compares the measurement result with a preset safety distance of 1.5 meters. It then generates a signal indicating that the first preset condition has been met, triggering the subsequent image acquisition process.

[0044] This module effectively balances system power consumption and detection response speed through a hierarchical detection mechanism, maintaining a low-power monitoring state in non-operating areas and activating ultrasonic detection only when necessary. For example, during unloaded driving, positioning unit 110 continuously monitors the area status and activates the full ranging process when a potential operating area is detected, significantly reducing the energy consumption associated with continuous high-frequency detection. Furthermore, the modular design allows each functional unit to be independently optimized and upgraded. For example, by replacing a high-performance ultrasonic transducer or upgrading the signal processing algorithm, detection accuracy and anti-interference capabilities in specific scenarios can be improved.

[0045] Through the methods of the embodiments of the present invention, the accuracy, real-time performance and environmental adaptability of the distance detection between the AGV forklift and the target object can be significantly improved. Ultrasonic ranging technology can accurately measure the distance, avoid the problem of false detection caused by changes in ambient light or background noise, and reduce the false detection rate. The ultrasonic transmitting unit 120 has a simple structure and low cost, and is easy to integrate into the existing AGV forklift system, thereby simplifying the system structure and reducing the implementation cost. Under complex lighting conditions, ultrasonic ranging can still work stably to ensure high accuracy of detection. Based on Example 1, the present invention further optimizes the detection method, providing reliable technical support for the intelligent upgrade of AGV forklifts in high-density warehousing environments.

[0046] In a preferred embodiment, the distance measurement and condition determination module 100 further includes: Cycle counting unit, used to record the number of consecutive measurements; A condition judgment unit is connected to the cycle counting unit and is used to confirm that the distance meets the first preset condition when the distance measurement values ​​are all lower than the preset distance threshold for three consecutive times.

[0047] In this exemplary embodiment, the distance measurement and condition determination module 100 achieves accurate distance determination through the collaborative operation of a cycle counting unit and a condition determination unit. This module uses a laser or ultrasonic distance measurement unit to collect distance data at a fixed frequency, ensuring real-time performance. Each acquired distance value is dynamically compared with a preset threshold, which can be adjusted based on different operational scenarios.

[0048] The loop counting unit uses a ring buffer mechanism to store recent measurement results, for example, retaining three consecutive readings and automatically updating them when new data is generated. This unit also includes a built-in threshold comparison module that generates status flags in real time for subsequent logical analysis. The conditional judgment unit performs complex logical analysis based on the loop counting unit's output. For example, a trigger signal is generated only when multiple consecutive measurement values ​​fall below a preset threshold, filtering out single measurement anomalies or environmental interference. This logic is implemented using a hardware description language to ensure stability even at high frequencies.

[0049] This embodiment enhances reliability through a dual-redundancy design. A continuous measurement mechanism prevents false triggering due to accidental interference, such as dynamic adjustments by the AGV or brief displacement of the target object, preventing the system from mistakenly initiating image acquisition. The dynamic threshold adjustment function adapts to different operating scenarios, for example, tightening the detection criteria in narrow passages while relaxing the threshold to maintain sensitivity in open areas. This design significantly reduces false alarm rates while maintaining rapid response, making it suitable for highly dynamic operating environments.

[0050] In terms of resource optimization, this embodiment reduces the operating hours of the image acquisition module through a conditional trigger mechanism, activating it only when conditions are met and remaining in low-power standby mode. This design effectively reduces average system power consumption, extending mobile device battery life, while also reducing ineffective image processing and optimizing computing resource utilization. For example, the load on the image processing unit is significantly reduced, providing ample margin for multitasking.

[0051] This embodiment achieves a balance between reliability, response speed, and energy efficiency while ensuring detection accuracy through software and hardware collaborative design, providing a robust and economical spacing detection solution for automated warehousing equipment.

[0052] In a preferred embodiment, the image recognition and positioning module 300 includes: A dual-modal feature extraction unit, configured to extract edge gradient feature parameters from the two-dimensional digital image using a Canny edge detection algorithm and extract geometric feature parameters using an ORB feature matching algorithm; a feature fusion unit connected to the dual-modal feature extraction unit, configured to fuse the edge gradient feature parameter with the geometric feature parameter to generate a fused feature parameter; A topology matching unit is connected to the feature fusion unit and is used to determine the existence of the target pallet based on the matching results of the fused feature parameters and the preset database. The judgment conditions include detecting at least three raised features that meet the foot size threshold and the spatial arrangement of the raised feature parameters meeting the tripod pallet topology constraint.

[0053] In this embodiment, the terms involved are explained as follows: Canny edge detection algorithm: A classic image processing algorithm used to extract edge features in images, achieving high-precision edge extraction through dual threshold detection and edge tracking.

[0054] ORB feature matching algorithm: A feature point-based image matching algorithm that extracts the geometric features of an image by detecting key points and calculating their descriptors.

[0055] Edge gradient feature parameters: refers to the gradient information of the image edge extracted by the Canny edge detection algorithm, which is used to describe the contour features of the target object.

[0056] Geometric feature parameters: refers to the geometric information of image key points extracted by the ORB feature matching algorithm, which is used to describe the shape and structural characteristics of the target object.

[0057] Fusion feature parameters: refers to the feature parameters generated by combining edge gradient feature parameters with geometric feature parameters, which are used to comprehensively describe the characteristics of the target object.

[0058] Foot size threshold: refers to the minimum size threshold that the projected area of ​​the target pallet's foot protrusion feature in the image must meet.

[0059] Topological constraints of a tripod pallet: refers to the geometric relationship that the three raised features of the target pallet must satisfy in their spatial arrangement, for example, the maximum spacing deviation between the three points must not exceed a certain ratio.

[0060] In this exemplary embodiment, the image recognition and positioning module 300 achieves accurate identification of the target pallet through a multi-stage processing unit. Its core components include a bimodal feature extraction unit, a feature fusion unit, and a topology matching unit. The bimodal feature extraction unit first performs parallel feature extraction on the input two-dimensional digital image. It uses the Canny edge detection algorithm to construct a contour feature set, for example, by suppressing noise through Gaussian filtering, calculating the image gradient amplitude and direction information, and then extracting clear edges through dual threshold detection. Simultaneously, it uses the ORB feature matching algorithm to generate geometric feature parameters, for example, through FAST keypoint detection and BRIEF descriptor calculation, to obtain a rotationally invariant feature vector.

[0061] The feature fusion unit performs a weighted fusion of the two feature parameters, for example, establishing a mapping relationship between edge gradient intensity and geometric feature response to generate a comprehensive feature vector containing both contour and structural information. This fusion strategy effectively combines the accuracy of edge detection with the robustness of feature matching, maintaining the stability of feature descriptions even in scenes with varying lighting intensity.

[0062] The topology matching unit incorporates a pre-defined pallet feature database containing multi-dimensional feature templates for standardized pallets. The matching process employs a two-level verification mechanism: First, candidate targets are screened based on feature vector similarity, for example, by setting a similarity threshold to filter out non-pallet objects. Next, topology constraint verification is performed, requiring the detection of at least three raised features that meet a foot size threshold and whose spatial arrangement satisfies the geometric constraints of a three-legged pallet. For example, the feature point spacing deviation must not exceed a set ratio to ensure that the identified object conforms to the ISO standard pallet structure.

[0063] This module significantly improves recognition reliability in complex scenarios through a design that combines bimodal feature fusion with topology verification. Experiments show that in warehouse environments with dense shelves and uneven lighting, the module maintains a high level of pallet recognition accuracy and significantly reduces the misidentification rate. Furthermore, the introduction of the ORB feature matching algorithm improves feature extraction efficiency. Combined with the edge accuracy of the Canny operator, the processing time for a single frame of image is shortened while ensuring real-time performance, meeting the real-time requirements of high-speed AGV forklift operations. This design is particularly suitable for highly automated warehousing systems, providing a reliable target positioning foundation for over-width detection and warning functions.

[0064] In a preferred embodiment, the bimodal feature extraction unit includes: FAST corner detection subunit, used to detect corners in the image in real time when performing ORB feature matching; The BRIEF descriptor generation subunit is used to generate binary descriptors of corner points when performing ORB feature matching.

[0065] In this embodiment, the terms involved are explained as follows: ORB feature matching algorithm: a feature point-based image matching algorithm that combines FAST corner detection and BRIEF descriptor generation technology to extract the geometric features of the image.

[0066] FAST corner points: refer to key points in an image detected by the FAST (Features from Accelerated Segment Test) algorithm. These points are usually located in areas with significant changes in the image, such as edges or corners.

[0067] BRIEF descriptor: A binary descriptor used to describe the local features of FAST corner points. It generates a binary string by comparing the grayscale values ​​of pixel pairs around the corner point. It has the characteristics of high computational efficiency and small storage space.

[0068] In this exemplary embodiment, the dual-modal feature extraction unit integrates the FAST corner detection subunit with the BRIEF descriptor generation subunit to achieve efficient geometric feature extraction. The FAST corner detection subunit uses a fast detection algorithm based on grayscale differences. For example, by comparing the grayscale value difference between a pixel and its neighbors, a threshold is set to quickly screen potential corners. This threshold can be dynamically adjusted based on image contrast, for example, set to a typical value to balance detection sensitivity and computational efficiency, ensuring accurate localization of key feature points in complex backgrounds.

[0069] The BRIEF descriptor generation subunit and the FAST detection subunit form a pipelined processing architecture, performing local feature encoding on detected corner points. For example, a specific number of pixel pairs are selected around the corner point and a compact feature descriptor is generated through binary encoding. The pixel pair selection strategy can adopt a randomized or structured pattern, for example, selecting 256 pixel pairs to generate a 256-bit binary descriptor, significantly reducing computational complexity while maintaining feature uniqueness.

[0070] The dual-modal feature extraction unit achieves real-time processing through hardware acceleration technology, such as using an FPGA or dedicated image processing chip to perform corner detection and descriptor generation in parallel. This design significantly accelerates feature extraction and reduces single-frame image processing time, meeting the real-time requirements of high-speed AGV forklift operations. Furthermore, the storage and matching efficiency of binary descriptors is significantly improved compared to traditional floating-point descriptors, effectively reducing system memory usage and comparison latency.

[0071] This embodiment achieves a balance between feature extraction accuracy and computational resource consumption through hardware implementation of the FAST-BRIEF algorithm. Experiments show that under varying lighting conditions in warehouse scenarios, the module maintains a high corner point recurrence rate and significantly improves feature matching accuracy. Compared to pure software implementations, this design reduces feature extraction power consumption and is particularly suitable for embedded deployments in mobile AGVs, providing efficient and reliable feature data support for ultra-wide detection and early warning systems.

[0072] In a preferred embodiment, the image recognition and positioning module 300 further includes: The Hough transform processing unit is used to establish a coordinate system based on the center point of the tripod tray's foot and fit the boundary line using the parametric equation ρ = xcosθ + ysinθ; where ρ represents the vertical distance from the line to the coordinate origin (polar diameter), and θ represents the angle between the line and the positive direction of the x-axis (polar angle); The edge merging unit is used to determine that two lines are the same physical edge when the polar diameter difference between them is ≤ 5 mm. ρ1 and ρ2 represent the vertical distances (polar diameters) between the two lines and the coordinate origin, respectively, and δ = 5 mm is the pallet thickness tolerance.

[0073] In this embodiment, the terms involved are explained as follows: Cartesian coordinate system: A two-dimensional rectangular coordinate system based on the center points of the three legs of a tripod pallet, used to describe the position and boundaries of the target pallet.

[0074] Hough transform parametric equation: A mathematical method for detecting straight lines in an image. A straight line is represented by the polar coordinate parametric equation ρ = xcosθ + ysinθ, where ρ is the perpendicular distance from the line to the origin (polar diameter) and θ is the angle between the line and the positive x-axis (polar angle).

[0075] Two-dimensional histogram: A statistical histogram established in the (ρ,θ) parameter space, used to record the frequency of different line parameters in the Hough transform.

[0076] Threshold T: The threshold used to filter out the line parameter pairs with higher frequency in the Hough transform parameter space, indicating the significance of the straight line in the image.

[0077] Pallet thickness tolerance δ: A tolerance parameter used to determine whether two straight lines are the same physical edge, usually set to 5 mm.

[0078] In this exemplary embodiment, the image recognition and positioning module 300 achieves precise positioning of the target pallet's boundaries through the collaborative work of a Hough transform processing unit and an edge merging unit. The Hough transform processing unit first constructs a local coordinate system based on the center points of the three legs of the tripod pallet. For example, this system uses the least squares method to fit the center points of the leg contours as the coordinate origin and defines principal axis directions to eliminate the influence of different pallet postures. This unit then performs parameterized line detection: Based on the Hough transform principle, the polar coordinate equation ρ = xcosθ + ysinθ is used to transform the image space. This discretized parameter space is then used to achieve a parametric representation of line features.

[0079] In practice, the Hough transform processing unit traverses the θ parameter space with a set step size. For example, the range from 0 to π radians is divided into equal parts, generating candidate line parameters in increments of 1 degree. For each θ value, the corresponding polar radius ρ is calculated and the voting matrix in the parameter space is updated. Ultimately, a two-dimensional histogram is formed to represent the frequency of occurrence of each parameter combination. This process is parallelized using a hardware acceleration module.

[0080] This embodiment further defines the specific method of locating the coordinates of the two sides of the front of the target pallet based on the above embodiment. The method of using the Hough transform parameter equation and the two-dimensional histogram has the following advantages: Improve edge detection accuracy: Hough transform can accurately fit the straight edges in the image and is suitable for high-precision positioning of the target pallet boundary.

[0081] Enhanced robustness: Through two-dimensional histogram and threshold screening, noise and interference can be effectively filtered out, improving the robustness of edge detection.

[0082] Adaptability to complex scenes: Hough transform has certain adaptability to image rotation and partial occlusion, and can meet the needs of edge detection in complex dynamic environments.

[0083] Optimize computational efficiency: By setting the step size and threshold, the computational complexity can be reduced while ensuring accuracy, thus meeting real-time requirements.

[0084] In a preferred embodiment, the calculation module 400 includes: The original width calculation unit is used to calculate the original width value W based on the coordinates on both sides raw ; The error compensation unit is used to perform dynamic compensation calculations, and its formula is: , where ΔD=2cm is the preset absolute value of the ranging error, D refis the real-time ranging threshold, α is the material compensation coefficient and α∈[0.1,0.3].

[0085] In this exemplary embodiment, the calculation module 400 achieves precise measurement and dynamic correction of the pallet's actual width through the collaborative operation of the original width calculation unit and the error compensation unit. The original width calculation unit first receives coordinate data from the image recognition and positioning module 300 and calculates the horizontal distance between the two boundary points on the front side of the pallet using, for example, the Euclidean distance formula. This unit also includes a built-in coordinate conversion module that converts position parameters in the pixel coordinate system into actual dimensions in the world coordinate system. For example, this module uses a pre-calibrated camera intrinsic parameter matrix to eliminate the effects of perspective distortion, ensuring the geometric accuracy of the width calculation result.

[0086] The error compensation unit dynamically corrects the original measurement values ​​based on a physical model and empirical parameters. The compensation formula utilizes a nonlinear correction model, with a preset absolute value of the ranging error, ΔD, set to, for example, 2 centimeters, reflecting the system-level error of the ultrasonic or laser ranging module. A real-time ranging threshold, Dref, serves as a normalization benchmark, with a dynamic threshold of, for example, 1.5 meters, adapted to different operating scenarios. The material compensation coefficient, α, is configured based on the pallet's material characteristics, for example, 0.15 for wooden pallets, 0.25 for plastic pallets, and 0.30 for metal pallets. This parameter is obtained through the material recognition module or manual input. The compensation unit quickly matches the α value using a table lookup and calculates the width correction based on the real-time ranging data. For example, when the ΔD / Dref ratio is 0.013, the width correction coefficient for a wooden pallet is 1.00195, achieving millimeter-level accuracy compensation.

[0087] The dual units achieve efficient computing through a pipelined processing architecture: the original width calculation unit uses a fixed-point arithmetic accelerator card to perform floating-point operations, reducing the time required for single calculations; the error compensation unit uses a pre-built compensation parameter table for rapid table lookup, avoiding delays caused by real-time calculations. This design enables the system to maintain measurement accuracy while shortening the single-frame processing cycle, meeting the real-time requirements of high-speed AGV forklift operation scenarios. Experiments have shown that the width measurement error is significantly reduced after dynamic compensation, with the error less than ±1 mm in the wooden pallet scenario and less than ±1.5 mm in the metal pallet scenario, significantly improving the reliability of over-width detection.

[0088] This module effectively overcomes the systematic errors associated with a single measurement method through a compensation strategy that combines physical modeling with data-driven approaches. This makes it particularly suitable for intelligent warehousing scenarios involving mixed operations involving multi-material pallets. The parameters in the compensation formula can be dynamically optimized based on the operating environment. For example, a machine learning algorithm continuously updates the matching relationship between ΔD and α, forming an adaptive optimization mechanism that reliably ensures the precise operation of AGV forklifts.

[0089] Example 3 See also Figure 3 , Figure 3 This is a flow chart of a method for detecting and warning an over-width pallet on an AGV forklift provided by way of example in an embodiment of the present application. The method comprises: S100, detecting the distance between the AGV forklift and the target object; S200: If the distance satisfies a first preset condition, acquiring a two-dimensional digital image of the target area; S300, identifying whether a target pallet exists in the target area based on the two-dimensional digital image of the target area; S400, if the target pallet exists in the identified target area, locate the coordinates of both sides of the front of the target pallet; S500, calculating the width of the target pallet according to the coordinates of both sides of the front side of the target pallet; S600: If the width of the target pallet meets the second preset condition, trigger an early warning and terminate the current workflow of the AGV forklift.

[0090] In the prior art, AGV forklifts need to frequently move and adjust their posture during operation, and the camera is constantly working to capture potential targets, causing the system to process a large amount of redundant information, such as empty areas and non-pallet objects. This indiscriminate collection and processing of image data not only wastes computing resources, but also causes misidentification due to changes in ambient light and background interference, increasing device power consumption and hardware loss, and reducing the overall reliability of the detection system. Therefore, the present invention solves the resource waste and false triggering problems caused by indiscriminate continuous operation in the prior art by optimizing the detection logic and data collection strategy, thereby meeting the comprehensive requirements of AGV equipment for efficiency, stability, and energy saving in high-density warehousing environments.

[0091] Specifically: First, detect the distance between the AGV forklift and the target object: A laser ranging sensor or an ultrasonic transmitting unit 120 is used to detect the distance between the AGV forklift and the target object in real time.

[0092] When it is detected that the distance is less than a preset threshold (a first preset condition), the image acquisition module is triggered.

[0093] Next, a two-dimensional digital image of the target area is acquired: The two-dimensional digital image of the target area is collected by the camera installed on the AGV forklift.

[0094] The image acquisition module is only started when the distance meets the first preset condition to avoid indiscriminate continuous operation.

[0095] Then, identify whether the target pallet exists in the target area: Use image processing algorithms (such as edge detection, template matching, etc.) to process the collected two-dimensional digital image to identify whether the target pallet exists in the target area.

[0096] If the target pallet is identified, proceed to the next step; otherwise, continue to monitor the distance.

[0097] Next, locate the coordinates on both sides of the front of the target pallet: The coordinates of the front and sides of the target pallet are located through image processing algorithms to determine the boundary position of the pallet.

[0098] Additionally, calculate the width dimension of the target pallet: Calculate the width of the pallet based on the coordinates of the two sides of the front of the located target pallet.

[0099] Finally, trigger an early warning and stop the current workflow of the AGV forklift: If the calculated pallet width exceeds the preset threshold (the second preset condition), an early warning signal is triggered and the current workflow of the AGV forklift is terminated to prevent collisions or jams caused by excessive pallet width.

[0100] Technical Effects Reduced redundant computational load: By capturing images only when the spacing meets the first preset condition, redundant information processing caused by indiscriminate continuous work is avoided, significantly reducing the computational load of the image processing system.

[0101] Improved recognition reliability: Optimized detection logic and data collection strategies reduce the impact of ambient light changes and background interference on recognition results, improving recognition reliability in complex dynamic scenarios.

[0102] Energy saving and hardware protection: Reducing the working hours of the camera and image processing module reduces device power consumption and hardware loss, and extends the service life of the device.

[0103] Improved operational efficiency: Accurate pallet width detection and early warning mechanisms avoid false warnings and workflow interruptions caused by misidentification, improving the operational efficiency of AGV forklifts.

[0104] Through the implementation of the above embodiments, the present invention significantly reduces the redundant computing load of the image processing system while ensuring the accuracy of AGV forklift pallet overwidth detection, while improving the recognition reliability in complex dynamic scenes, and meeting the comprehensive requirements of AGV equipment for efficiency, stability and energy saving in high-density warehousing environments.

[0105] In a preferred embodiment, the AGV forklift pallet over-width detection and early warning method includes the following steps: First, the distance between the AGV forklift and the target object is detected. The AGV forklift is a forklift handling robot equipped with automatic navigation. The target object is a standardized pallet to be moved in a warehouse. A laser ranging module uses a pulsed sensor with a wavelength of 905nm. Installed at the front of the AGV fork, the module measures straight-line distance at a frequency of 10Hz. A detection value of less than or equal to 1.5 meters triggers the subsequent process. This distance threshold can be dynamically adjusted based on the AGV model.

[0106] Secondly, a two-dimensional digital image of the target area is captured in a triggered manner. When the distance meets the first preset condition, the industrial camera is activated to capture the target area image. The industrial camera is a global shutter CMOS sensor with a resolution of 1920×1080, a frame rate of 30fps, and a field of view covering a 2m×1.5m area in front of the forks. Using regional dynamic exposure technology, the exposure time is automatically adjusted from 50 microseconds to 10 milliseconds based on ambient light intensity. A ROI image of the target area is captured and transmitted to the image processing unit. The target area is defined as a detection range of 2 meters wide with the fork centerline as the axis of symmetry to eliminate background interference.

[0107] Next, multi-feature fusion is used to identify the target pallet. The image processing unit performs the following operations. In the preprocessing stage, contour features are extracted using Gaussian filtering and Canny edge detection. The Gaussian filter kernel is 3×3, and the dual thresholds for Canny edge detection are 30 and 150, respectively. In the feature recognition stage, the pallet's texture features are analyzed based on a pretrained CNN classifier. At least three raised features that meet the foot size threshold are detected, and the projected area must be greater than or equal to 100 square centimeters. The topology verification stage verifies whether the spatial arrangement of the raised features conforms to the topological constraints of the tripod pallet. The maximum spacing deviation between the three points must be less than or equal to 15%.

[0108] Sub-pixel coordinate positioning and conversion are then performed. After confirming the presence of the target pallet, a sub-pixel corner detection algorithm is used to locate the coordinates of both sides of the pallet's front face, with a positioning accuracy of 0.1 pixel. Using a pre-calibrated camera intrinsic parameter matrix and an inverse perspective transformation model, pixel coordinates are converted to 3D coordinates in the world coordinate system, eliminating perspective distortion errors. The focal length of the camera intrinsic parameter matrix is ​​8 mm, and the pixel size is 3.45 microns.

[0109] Then, the dynamic error compensation width is calculated. The actual width of the pallet is calculated based on the converted coordinates. The basic calculation formula is The dynamic compensation formula is: , where ΔD is 2 cm, is the absolute value of the maximum error of ultrasonic ranging, D ref The material compensation coefficient α is determined by looking up the table. The α of wooden pallets is 0.15, the α of plastic pallets is 0.25, and the α of metal pallets is 0.30.

[0110] Finally, the graded response control. When the calibration width W cal When the safety threshold of 1.2 meters ± 50 mm is exceeded, an audible and visual alarm is activated at a frequency of 2kHz and a sound intensity of 90dB or higher. The AGV's power supply is cut off within 5 milliseconds via the CAN FD bus. Event data is uploaded to a central management system, which automatically plans an obstacle avoidance path.

[0111] The following technical effects can be achieved through the method of the embodiment of the present invention. In terms of computational efficiency, 80% of invalid images are filtered out through the distance threshold, and the single-frame processing time is reduced from 120 milliseconds to 50 milliseconds. The test platform is an Intel i5 processor. In terms of detection accuracy, the width measurement error is less than or equal to ±2 mm, and the false recognition rate is less than 2%, based on 1,000 test verifications. In terms of response performance, the response time from judgment to termination of the action is less than or equal to 200 milliseconds, and the collision accident rate is less than 0.1%. In terms of environmental adaptability, in a low-light environment, when the illumination is less than or equal to 50 lux, the detection accuracy is greater than or equal to 98.5%.

[0112] In an extended implementation, an adaptive threshold adjustment module can be integrated. Regarding dynamic parameter optimization, the distance threshold in high-density shelf areas is reduced from 1.5 meters to 1.2 meters. Regarding environmental adaptation, HDR imaging mode is activated in bright light scenes and exposure is extended to 8 milliseconds. Regarding the self-calibration mechanism, the camera's intrinsic parameter matrix is ​​recalibrated after three consecutive false alarms. This module dynamically adjusts detection parameters by analyzing historical data and sensor feedback, including lighting, temperature, and humidity, and collaborates with the AGV scheduling platform to achieve closed-loop control.

[0113] In a preferred embodiment, based on the two-dimensional digital image of the target area, it is identified whether the target tray exists in the target area, see Figure 4 , Figure 4 This is a flowchart of a method for an AGV forklift to identify the presence of a target pallet in a target area, as provided by an example in an embodiment of the present application, specifically including: Perform the following processing on the two-dimensional digital image: S401, using the Canny edge detection algorithm to extract edge gradient feature parameters; S402, extracting geometric feature parameters through ORB feature matching algorithm; S403, fusing edge gradient feature parameters and geometric feature parameters to generate fused feature parameters; S404: Compare the fused feature parameters with the preset pallet feature database: At least three convex feature parameters meeting the foot size threshold are detected; And / or, the spatial arrangement of the protrusion feature parameters complies with the topological constraints of the tripod tray.

[0114] S405: If the condition in step S404 is met, it is determined that the target tray exists.

[0115] This embodiment further defines a specific method for identifying a target pallet based on a two-dimensional digital image based on the above embodiment. The combination of the Canny edge detection algorithm and the ORB feature matching algorithm has the following advantages: 1. Improve feature extraction accuracy: The Canny edge detection algorithm can accurately extract the contour features of the target pallet, while the ORB feature matching algorithm can effectively extract geometric features. The combination of the two can significantly improve the comprehensiveness and accuracy of feature extraction.

[0116] 2. Enhanced recognition robustness: By fusing edge gradient feature parameters with geometric feature parameters, it can effectively cope with complex environmental factors such as lighting changes and background interference, thereby improving the robustness of the recognition algorithm.

[0117] 3. Optimize computational efficiency: The ORB feature matching algorithm has high computational efficiency. Combined with the Canny edge detection algorithm, it can reduce computational complexity while ensuring accuracy.

[0118] 4. Meet practical application needs: By detecting foot size thresholds and topological constraints, it can accurately identify standard three-legged pallets, meeting the actual needs of pallet identification in high-density storage environments.

[0119] Specifically: 1. Perform processing on a two-dimensional digital image: The Canny edge detection algorithm is used to extract edge gradient feature parameters. First, the image is Gaussian filtered to remove noise, then the image gradient amplitude and direction are calculated, and finally the edge features are extracted through double threshold detection and edge tracking.

[0120] Extract geometric feature parameters through ORB feature matching algorithm. Detect key points in the image and calculate their ORB descriptors to generate geometric feature parameters.

[0121] The edge gradient feature parameters and geometric feature parameters are fused to generate fused feature parameters. The edge gradient information is combined with the geometric feature information to form a feature vector that comprehensively describes the target pallet.

[0122] 2. Compare and merge feature parameters with the preset pallet feature database: The fused feature parameters are compared with the preset pallet feature database to determine whether the target pallet exists.

[0123] If at least three raised feature parameters that meet the foot size threshold are detected, and the spatial arrangement of these raised feature parameters meets the topological constraints of the tripod pallet (for example, the maximum spacing deviation between the three points does not exceed 15%), then it is determined that the target pallet exists.

[0124] The solution of this embodiment can significantly improve the accuracy, algorithm robustness and computational efficiency of target pallet identification. By fusing edge gradient feature parameters with geometric feature parameters, combining the Canny edge detection algorithm and the ORB feature matching algorithm, it can effectively cope with lighting changes and background interference in complex environments, and significantly reduce the misrecognition rate. At the same time, the high efficiency of the ORB feature matching algorithm is combined with the accuracy of the Canny edge detection algorithm to optimize the computational efficiency of image processing and meet real-time requirements. In addition, by detecting the foot size threshold and topological constraints, it can accurately identify standard tripod pallets to meet the actual needs of AGV forklift intelligent operations in high-density warehousing environments. Based on the above embodiment, this embodiment further optimizes the target pallet identification method, and adopts the method of multi-feature fusion and topological constraint verification to provide a more accurate and efficient technical guarantee for the practical application of the AGV forklift pallet over-width detection and warning system.

[0125] Example 4 An embodiment of the present invention further provides a computer-readable storage medium storing at least one program code, which is loaded and executed by a processor to implement the AGV forklift pallet overwidth detection and warning method described in Example 1 and Example 2 above. The computer-readable storage medium stores a computer program, which is executed by a processor.

[0126] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0127] Example 5 An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements some or all of the steps of the AGV forklift pallet overwidth detection and warning method described in Examples 1 and 2.

[0128] The above description is only a preferred embodiment of the present application and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An AGV forklift pallet over-width detection and warning system, characterized in that: include: A distance measurement and condition judgment module is used to detect the distance between the AGV forklift and the target object and determine whether the distance meets a first preset condition; an image collector, connected to the distance measurement and condition judgment module, for collecting a two-dimensional digital image of the target area when the distance meets a first preset condition; an image recognition and positioning module, connected to the image collector and the distance measurement and condition judgment module respectively, for identifying whether a target pallet exists in the target area based on the two-dimensional digital image and the distance, and locating the coordinates of both sides of the front of the target pallet; A calculation module, connected to the image recognition and positioning module, for calculating the width of the target pallet based on the coordinates on both sides of the front side; The judgment execution module is connected to the calculation module and is used to judge whether the width dimension meets the second preset condition. If so, an early warning operation is performed; otherwise, the current working process of the AGV forklift is terminated.

2. The system according to claim 1, wherein The distance measurement and condition judgment module includes: Positioning unit, used to detect whether the AGV forklift has entered the target area; An ultrasonic transmitting unit, connected to the positioning unit, is used to transmit an ultrasonic pulse signal when detecting that the AGV forklift enters the target area; an ultrasonic receiving unit, connected to the ultrasonic transmitting unit, and configured to receive a reflected signal from a target object; The distance calculation unit is used to generate real-time distance data based on the time difference between the transmitted signal and the reflected signal.

3. The system according to claim 1, wherein The distance measurement and condition judgment module also includes: Cycle counting unit, used to record the number of consecutive measurements; A condition judgment unit is connected to the cycle counting unit and is used to confirm that the distance meets the first preset condition when the distance measurement values ​​are all lower than the preset distance threshold for three consecutive times.

4. The system according to claim 1, wherein: The image recognition and positioning module includes: A dual-modal feature extraction unit, configured to extract edge gradient feature parameters from the two-dimensional digital image using a Canny edge detection algorithm and extract geometric feature parameters using an ORB feature matching algorithm; a feature fusion unit connected to the dual-modal feature extraction unit, configured to fuse the edge gradient feature parameter with the geometric feature parameter to generate a fused feature parameter; A topology matching unit is connected to the feature fusion unit and is used to determine the existence of the target pallet based on the matching results of the fused feature parameters and the preset database. The judgment conditions include detecting at least three raised features that meet the foot size threshold and the spatial arrangement of the raised feature parameters meeting the tripod pallet topology constraint.

5. The system according to claim 4, wherein: The bimodal feature extraction unit comprises: FAST corner detection subunit, used to detect corners in the image in real time when performing ORB feature matching; The BRIEF descriptor generation subunit is used to generate binary descriptors of corner points when performing ORB feature matching.

6. The system according to claim 4, wherein: The image recognition and positioning module also includes: The Hough transform processing unit is used to establish a coordinate system based on the center point of the tripod tray's foot and fit the boundary line using the parametric equation ρ = xcosθ + ysinθ; where ρ represents the vertical distance from the line to the coordinate origin (polar diameter), and θ represents the angle between the line and the positive direction of the x-axis (polar angle); The edge merging unit is used to determine that two lines are the same physical edge when the polar diameter difference between them is ≤ 5 mm. ρ1 and ρ2 represent the vertical distances (polar diameters) between the two lines and the coordinate origin, respectively, and δ = 5 mm is the pallet thickness tolerance.

7. The system according to claim 1, wherein: The calculation module includes: The original width calculation unit is used to calculate the original width value W based on the coordinates on both sides raw ; The error compensation unit is used to perform dynamic compensation calculations, and its formula is: , where ΔD=2cm is the preset absolute value of the ranging error, D ref is the real-time ranging threshold, α is the material compensation coefficient and α∈[0.1,0.3].

8. A method for detecting and warning of excessive width of an AGV forklift pallet, characterized in that: The steps include: Detect the distance between the AGV forklift and the target object; If the distance satisfies a first preset condition, acquiring a two-dimensional digital image of the target area; Based on the two-dimensional digital image of the target area, identifying whether a target pallet exists in the target area; If it is identified that a target pallet exists in the target area, locating the coordinates of both sides of the front side of the target pallet; Calculating the width of the target pallet based on the coordinates of both sides of the front side of the target pallet; If the width of the target pallet meets the second preset condition, an early warning is triggered, and the current working process of the AGV forklift is terminated.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the AGV forklift pallet overwidth detection and early warning method according to any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the AGV forklift pallet overwidth detection and early warning method according to any one of claims 1 to 7 are implemented.

Images