Cold chain storage AGV positioning method, system and device and readable storage medium

Abstract

The invention provides a cold chain storage AGV positioning method, system and device and a readable storage medium, and the method comprises the steps: obtaining measurement data, constructing a global map in a storage environment where an AGV is located according to the measurement data, and generating the global positioning coordinates of the AGV; the method comprises the following steps: acquiring environment data, constructing a local map in a storage environment in which an AGV is located according to the environment data, and generating local positioning coordinates of the AGV; and acquiring the global positioning coordinates and the local positioning coordinates, fusing the global positioning coordinates and the local positioning coordinates based on a Kalman filtering algorithm, and obtaining and outputting the fused AGV positioning information. According to the invention, the requirements of high accuracy and high reliability of AGV positioning in a cold chain storage extreme environment are met.

Description

Technical Field

[0001] The present invention relates to the field of cold chain warehouse automation technology, and in particular to a cold chain warehouse AGV positioning method, system, equipment and readable storage medium. Background Art

[0002] With the acceleration of global economic integration and the upgrading of consumer demand, the logistics and warehousing industry is undergoing a profound intelligent transformation. As the core hub of the supply chain for specialty items such as fresh food and pharmaceuticals, cold chain warehousing systems bear the critical mission of ensuring product quality and safety. In ultra-large cold chain centers with daily processing volumes exceeding tens of millions of pieces, traditional manual handling is no longer sufficient for 24 / 7 operations. Automated Guided Vehicles (AGVs), with their efficiency and flexibility, have become the core equipment for achieving cold chain warehousing automation. The positioning accuracy of AGVs not only directly impacts the efficiency of cargo handling and sorting, but also crucially affects the digital precision of inventory management and the development of a safety and protection system for cold chain operations.

[0003] However, cold chain warehouses, operating at an extremely low temperature of -30°C, present a severe challenge to AGV positioning technology. Lithium battery capacity decay in low temperatures, leading to signal transmission delays for electronic components. High humidity can easily cause frost and condensation on sensor surfaces, distorting data from core sensing devices like lidar and visual cameras. The densely packed metal shelves and the frequent movement of goods in and out of the warehouse create a complex electromagnetic reflection environment. This causes multipath effects on wireless positioning signals such as UWB (Ultra-Wideband, a wireless communication technology that uses an extremely wide frequency bandwidth to achieve high-speed data transmission and precise ranging), exacerbating positioning errors. In narrow aisles, AGVs share space with forklifts and mobile racks, and the occlusion caused by dynamic obstacles further exacerbates the risk of positioning system failure. Traditional outdoor positioning technologies based on GPS and Beidou are shielded by metal building structures and cannot be used. Commonly used indoor inertial navigation and visual SLAM (Visual Simultaneous Localization and Mapping) technologies suffer from cumulative error divergence and image feature point loss in low-temperature environments, making them difficult to meet the high positioning accuracy and stability requirements of cold chain warehousing scenarios. Therefore, developing an AGV positioning method that achieves high-precision and high-reliability positioning in cold chain warehousing environments has become a pressing issue for those skilled in the art. Summary of the Invention

[0004] The present invention provides a cold chain warehousing AGV positioning method, system, equipment and readable storage medium, which solves the problem of high-precision and high-reliability requirements for AGV positioning in extreme cold chain warehousing environments.

[0005] According to one aspect of the present invention, a cold chain storage AGV positioning method is provided, comprising: Acquire measurement data, construct a global map of the warehouse environment where the AGV is located based on the measurement data, and generate the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the warehouse environment and a mobile positioning tag carried by the AGV; Acquire environmental data, construct a local map of the storage environment where the AGV is located based on the environmental data, and generate local positioning coordinates of the AGV; the environmental data is collected by an image acquisition device carried by the AGV; The global positioning coordinates and the local positioning coordinates are obtained, and the global positioning coordinates and the local positioning coordinates are fused based on a Kalman filter algorithm to obtain and output the fused AGV positioning information.

[0006] Furthermore, the global positioning coordinate generation step includes: Acquiring the measurement data, performing a time domain deconvolution transform on the measurement data to generate a time-reversed signal sequence; estimating a channel impulse response based on the time-reversed signal sequence, and identifying multipath components including a direct path and a reflected path; Extracting the amplitude, phase, and delay characteristic parameters of the multipath components; separating the direct path signal from the multipath components based on energy thresholds and propagation time characteristics; Based on the prior map information of the storage environment, the reflection path and its characteristics are predicted; an adaptive filter is constructed to remove interference from the measurement signal according to the predicted reflection path and extract the true TOF value; Convert the true TOF value into a distance measurement value, perform trilateral positioning calculation based on the base station coordinates, and obtain the AGV initial position estimate; use the reflection path characteristics to infer the environmental obstacle information and update the obstacle probability distribution of the three-dimensional grid map; According to the initial position estimate, based on Kalman filtering, historical positioning trajectories are integrated to suppress random errors; the matching degree between the current positioning result and the global map is calculated. If the matching degree is lower than a preset threshold, the relocalization process is triggered; if the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output.

[0007] Furthermore, the local positioning coordinate generating step includes: Acquiring the environmental data, performing preprocessing on the environmental data, and extracting features of the preprocessed environmental data based on a multi-scale convolutional neural network to obtain a feature map containing different features; Based on a feature matching algorithm, matching the features included in the feature map with the features of the constructed local map to identify key landmarks in the feature map; Based on the key landmark positions and the AGV motion model, calculating the relative position change of the AGV in the local map; The local map is updated according to the posture change information, and the updated local map and the local positioning coordinates of the AGV are output.

[0008] Furthermore, the step of fusing the global positioning coordinates with the local positioning coordinates based on the Kalman filter algorithm includes: Acquire fused data and preprocess the fused data; the fused data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, the first covariance matrix represents the measurement error of the UWB positioning system, and the second covariance matrix represents the measurement error of the visual positioning system, and the preprocessing includes verifying the data format and abnormal data processing; Initialize the Kalman filter and set the key parameters of the Kalman filter, which include the state variable dimension, state transfer matrix, sampling time, process noise covariance matrix, and observation noise covariance matrix that affect the filtering effect; The Kalman filter triggers fusion calculation according to the sampling time interval to obtain the optimal state estimation after fusion; The optimal state estimate after fusion is obtained, fused positioning coordinate information is extracted based on coordinate transformation, and the fused positioning coordinate information is output to the AGV navigation control module through a communication interface, with an output frequency consistent with the sampling time interval.

[0009] According to a second aspect of the present invention, a cold chain storage AGV positioning system is provided, comprising: The first module is used to obtain measurement data, construct a global map of the warehouse environment where the AGV is located based on the measurement data, and generate the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the warehouse environment and a mobile positioning tag carried by the AGV; The second module is used to obtain environmental data, construct a local map of the storage environment where the AGV is located based on the environmental data, and generate the local positioning coordinates of the AGV; the environmental data is collected by the image acquisition device carried by the AGV; The third module is connected to the first module and the second module respectively through a communication connection, and is used to obtain the global positioning coordinates output by the global positioning module and the local positioning coordinates output by the local positioning module, and fuse the global positioning coordinates and the local positioning coordinates based on the Kalman filter algorithm to obtain and output the fused AGV positioning information.

[0010] Furthermore, the first module includes: a multipath component identification unit connected to a fixed UWB base station and a mobile positioning tag via a communication connection, configured to obtain the measurement data, perform a time domain deconvolution transform on the measurement data to generate a time-reversed signal sequence, estimate a channel impulse response based on the time-reversed signal sequence, and identify multipath components including direct paths and reflected paths; a direct path separation unit, connected to the multipath component identification unit via a communication connection, configured to receive the multipath components output by the multipath component identification unit, extract amplitude, phase, and delay characteristic parameters of the multipath components, and separate the direct path signal from the multipath components based on an energy threshold and propagation time characteristics; a true TOF extraction unit, connected to the direct path separation unit via a communication connection, configured to receive the direct path signal output by the direct path separation unit, predict the reflection path and its characteristics based on prior map information of the storage environment, construct an adaptive filter, perform interference rejection processing on the measurement signal according to the predicted reflection path, and extract the true TOF value; an initial position calculation unit, connected to the real TOF extraction unit via a communication connection, configured to receive the real TOF value output by the real TOF extraction unit, convert the real TOF value into a distance measurement value, perform a trilateration calculation based on the base station coordinates to obtain an initial position estimate of the AGV, infer environmental obstacle information using reflection path features, and update the obstacle probability distribution of the three-dimensional grid map; A global coordinate optimization unit is connected to the initial position calculation unit via a communication connection, and is used to receive the AGV initial position estimate output by the initial position calculation unit. Based on the initial position estimate, the unit fuses the historical positioning trajectory based on the Kalman filter to suppress random errors, and calculates the matching degree between the current positioning result and the global map. If the matching degree is lower than a preset threshold, a relocation process is triggered. If the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output to the coordinate fusion module.

[0011] Furthermore, the second module includes: a feature extraction unit, connected to the image acquisition device carried by the AGV via a communication connection, configured to acquire the environmental data, perform preprocessing on the environmental data, and extract features of the preprocessed environmental data based on a multi-scale convolutional neural network to obtain a feature map containing different features; a key landmark identification unit, connected to the feature extraction unit via a communication connection, configured to receive a feature map output by the feature extraction unit, and match features contained in the feature map with features of the constructed local map based on a feature matching algorithm to identify key landmarks in the feature map; a posture change calculation unit connected to the key landmark identification unit via a communication connection, configured to receive key landmark information output by the key landmark identification unit, and calculate a relative posture change of the AGV in the local map based on the key landmark position and the AGV motion model; The local coordinate output unit is connected to the posture change calculation unit via a communication connection, and is used to receive the relative posture change information output by the posture change calculation unit, update the local map according to the posture change information, and output the updated local map and the local positioning coordinates of the AGV to the coordinate fusion module.

[0012] Furthermore, the third module includes: A data preprocessing unit is connected to the global coordinate optimization unit of the first module and the local coordinate output unit of the second module through a communication connection, and is used to obtain fused data and perform preprocessing; the fused data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, and the preprocessing includes verifying the data format and processing abnormal data; a filter initialization unit, connected to the data preprocessing unit via a communication connection, configured to receive the preprocessed data output by the data preprocessing unit, initialize the Kalman filter, and set key parameters of the Kalman filter, wherein the key parameters include state variable dimension, state transition matrix, sampling time, process noise covariance matrix, and observation noise covariance matrix; a fusion calculation unit connected to the filter initialization unit via a communication connection, configured to receive the initialization parameters output by the filter initialization unit, so that the Kalman filter triggers fusion calculation according to the sampling time interval to obtain a fused optimal state estimate; A result output unit is connected to the fusion calculation unit through a communication connection, and is used to obtain the optimal state estimation after fusion, extract the fused positioning coordinate information based on coordinate transformation, and output the fused positioning coordinate information through a communication interface, with an output frequency consistent with the sampling time interval.

[0013] According to a third aspect of the present invention, a cold chain storage AGV positioning device is provided, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method steps provided in the first aspect are implemented; AGV body, wherein the system provided in the second aspect is integrated into the AGV body.

[0014] According to a fourth aspect of the present invention, a readable storage medium is provided, on which a cold chain warehouse AGV positioning program is stored. When the positioning program is executed by the processor, the method steps provided in the first aspect are implemented.

[0015] The present invention provides a cold chain warehousing AGV positioning method, system, equipment and readable storage medium, which solves the problem of high precision and high reliability of AGV positioning in extreme cold chain warehousing environments. The present invention obtains the measurement data through the UWB base station and the mobile positioning tag to construct the global map to generate the global positioning coordinates, uses the image acquisition device to collect the environmental data to construct the local map to generate the local positioning coordinates, and then fuses the global positioning coordinates with the local positioning coordinates based on the Kalman filter algorithm, thereby suppressing multipath errors and improving positioning accuracy. By updating the three-dimensional grid map in real time and triggering the repositioning mechanism, the dynamic environmental adaptability is enhanced, and the efficiency and reliability are improved with the help of the Kalman filter dynamic weighted optimization algorithm. At the same time, a modular design is adopted to reduce the development difficulty and ensure real-time performance, thereby effectively solving the system instability problem caused by low AGV positioning accuracy and poor environmental adaptability in cold chain warehousing. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0017] Figure 1 This is a flow chart of a cold chain warehouse AGV positioning method according to an embodiment of the present invention;

[0018] Figure 2 is a diagram of steps for generating global positioning coordinates according to an embodiment of the present invention;

[0019] Figure 3 is a diagram of steps for generating local positioning coordinates according to an embodiment of the present invention;

[0020] Figure 4 is a diagram of steps for fusing the global positioning coordinates and the local positioning coordinates based on a Kalman filter algorithm according to an embodiment of the present invention;

[0021] Figure 5 4 is a block diagram of a cold chain warehousing AGV positioning system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments 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 work should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate for the embodiments of the present invention described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, product, or apparatus that includes a series of steps or units is not necessarily limited to the steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatuses.

[0024] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. Example 1

[0025] refer to Figure 1 The cold chain storage AGV positioning method according to an embodiment of the present invention includes:

[0026] S102, obtaining measurement data, constructing a global map of the warehouse environment where the AGV is located based on the measurement data, and generating the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the warehouse environment and a mobile positioning tag carried by the AGV.

[0027] S104, acquiring environmental data, constructing a local map of the storage environment where the AGV is located based on the environmental data, and generating local positioning coordinates of the AGV; the environmental data is acquired by an image acquisition device carried by the AGV.

[0028] S106, obtain the global positioning coordinates and the local positioning coordinates, fuse the global positioning coordinates and the local positioning coordinates based on the Kalman filter algorithm, and obtain and output the fused AGV positioning information. The Kalman filter algorithm is a filtering algorithm that initializes the state variables and covariance matrix, performs prediction calculations based on the state transition matrix at a fixed period to obtain the predicted state and covariance, and then combines the observation matrix with the updated observation noise covariance matrix to complete the update calculation, iteratively obtains the optimal state estimate, and fuses multi-source data and suppresses errors. The Kalman filter algorithm is implemented by the Kalman filter tool, which is a software module that integrates the Kalman filter algorithm and can be directly called to achieve data fusion processing without manually writing algorithm details.

[0029] refer to Figure 2 In step S102, the global positioning coordinate generation step includes:

[0030] S1202, obtain the measurement data, perform time domain deconvolution transform based on the measurement data, and generate a time-reversed signal sequence; estimate the channel impulse response based on the time-reversed signal sequence, and identify the multipath components including the direct path and the reflected path. More specifically, obtain the measurement data obtained by distance measurement between a fixed UWB base station deployed in a warehousing environment and a mobile positioning tag carried by an AGV, the measurement data being the original signal data of the UWB signal propagating between the base station and the tag, including the amplitude, phase, and delay information of the signal. The UWB base station is a device fixedly installed in a warehousing environment, used to transmit and receive UWB signals, and provide a reference point for positioning. The mobile positioning tag is a device carried on the AGV, which can communicate with the UWB base station to achieve distance measurement. There are errors in the measurement data, and the errors mainly come from the following aspects: First, equipment error. The hardware performance difference between the UWB base station and the mobile positioning tag will cause measurement errors. For example, the clock drift of the base station will cause deviations in time measurement, and the general error range is 0.1-1ns (nanoseconds); second, environmental interference error. The metal shelves, refrigeration equipment, etc. in cold chain warehouses will generate electromagnetic interference, resulting in signal amplitude attenuation and phase shift, and the error may reach 1-5ns; third, multipath propagation error. The reflected signal will make the measured data value larger, which will lead to distance measurement error. This error may reach 0.5-3m in complex environments. To address device errors, clock synchronization technology is used to regularly calibrate the clocks of the base station and the tag to control clock drift within 0.05ns. To address environmental interference errors, a filtering circuit is installed at the signal receiving end to filter out interference signals of specific frequencies. UWB devices with strong anti-interference capabilities are also selected to reduce the impact of electromagnetic interference. Multipath propagation errors are addressed using a multipath effect suppression algorithm. A time-domain inverse transform is performed on the measured data. Specifically, the time-domain signal is inverted and convolved to generate a time-reversed signal sequence. The time-domain inverse transform is a signal processing technique that enhances useful signal energy and suppresses noise by inverting and convolving the signal in the time dimension. The time-reversed signal sequence, obtained after the time-domain inverse transform, focuses on the useful signal, facilitating subsequent analysis. Based on this time-reversed signal sequence, an algorithm such as the least squares method is used to estimate the channel impulse response, thereby identifying multipath components, including direct and reflected paths. The channel impulse response is a function that describes the channel's effect on the input signal, reflecting the attenuation and delay characteristics of the signal during propagation. The direct path refers to the path along which the signal travels directly from the base station to the tag. The reflected path refers to the path along which the signal travels after being reflected by obstacles within the warehouse. The multipath components refer to the multiple signal components formed by the signal traveling from the base station to the tag via multiple different paths.The time domain inverse transformation and the estimated channel impulse response are performed to better extract the multipath components from the complex UWB signal, laying the foundation for the subsequent separation of the direct path and the reflected path. Because in cold chain warehousing, shelves, goods, etc. will cause a large number of reflected signals, interfering with positioning accuracy.

[0031] S1204, extract the amplitude, phase and delay characteristic parameters of each multipath component; based on the energy threshold and propagation time characteristics, separate the direct path signal from the multipath components. More specifically, extract the amplitude, phase and delay characteristic parameters of each multipath component. Amplitude is an indicator of signal strength, phase reflects the phase offset of the signal, and delay is the time it takes for the signal to propagate. Based on the energy threshold and propagation time characteristics, separate the direct path signal from the multipath components. The energy threshold is the energy limit used to distinguish between valid signals and noise. Signals exceeding the energy threshold are considered valid signals. The specific value of the energy threshold is determined according to the electromagnetic interference level of the storage environment, usually [-80dBm, -50dBm]. The propagation time characteristic judgment is to calculate the propagation time of each multipath component, and the component with the shortest propagation time is preferentially determined to be the direct path signal. Extracting the characteristic parameters of the multipath components can help identify the signal characteristics of different paths, and separating the direct path signal is because the direct path signal can more accurately reflect the true distance between the base station and the tag. The reflected path signal will cause distance measurement errors, which has a greater impact on subsequent precise positioning.

[0032] S1206, based on the prior map information of the warehouse environment, predict the reflection path and its characteristics; construct an adaptive filter, perform interference removal processing on the measurement signal according to the predicted reflection path, and extract the true TOF value. More specifically, the prior map information includes the location and size of the fixed shelves in the warehouse, the location and thickness of the walls, and the location and diameter data of the columns. The predicted reflection path and its characteristics include the propagation direction, distance, signal attenuation degree, etc. of the reflection path. The order of the filter is determined according to the number of predicted reflection paths. Based on the predicted reflection path, the measurement signal is subjected to interference removal processing, that is, the signal components that are consistent with the predicted reflection path characteristics are filtered out by the filter, and the true TOF (Time of Flight, a technology that calculates distance or speed by measuring the time it takes for particles or waves to propagate in a medium) value is extracted. The TOF value is the time it takes for the signal to propagate from the base station to the tag. In a cold chain warehousing environment, the impact of the multipath effect on the UWB positioning is mainly reflected in the following aspects: First, it causes the distance measurement value to be too large. Due to the existence of the reflected signal, the receiving end will mistake the reflected signal for a direct signal, which will increase the measured TOF value and thus the calculated distance will be greater than the actual distance. For example, when the signal reaches the tag after being reflected by a shelf 10m away, it will propagate 20m more than the direct signal, and the corresponding TOF value will increase by about 66.7ns, and the distance measurement error will increase by about 20m; second, it reduces the stability of positioning accuracy. The intensity and propagation path of the reflected signal will change with the movement of the AGV and the change of goods, resulting in large fluctuations in the measured distance value, making the positioning result unstable, and the error fluctuation range may reach 0.5-5m; third, it causes positioning ambiguity. When the intensity of multiple reflected signals is close to the direct signal, it is difficult for the receiving end to accurately distinguish between the direct signal and the reflected signal, and thus it is impossible to determine the correct propagation path, resulting in large deviations or even errors in the positioning results. The learning rate of the adaptive filter is set at 0.01-0.1. A learning rate that is too high will cause the filter to be unstable, while a learning rate that is too low will slow the convergence speed. Predicting the reflection path and performing interference removal is to eliminate the influence of the reflected signal on the TOF value measurement, because the reflected signal will cause the measured TOF value to be too large, resulting in inaccurate distance calculation. Accurate TOF value is the key to achieving precise positioning.

[0033] S1208: Convert the true TOF value to a distance measurement value and perform trilateration calculations based on the base station coordinates to obtain an initial position estimate for the AGV. Use the reflection path characteristics to infer environmental obstacle information and update the obstacle probability distribution of the three-dimensional grid map. The true TOF value is converted to a distance measurement using the following formula: distance = speed of light x TOF value / 2, where the division by 2 is due to round-trip signal propagation. Trilateration calculations are performed based on the known base station coordinates (pre-measured and calibrated three-dimensional coordinates (x, y, z) within the warehouse environment). The number of base stations should be at least three, and typically 4-8 should be deployed to improve positioning accuracy and reliability. The base stations should be arranged to cover the entire warehouse area to avoid positioning blind spots. Trilateration is a positioning method based on distance measurement. It uses the coordinates of three known points and the distance to the unknown point to determine the unknown point's position through geometric calculations. In this embodiment, trilateration calculations use the distances between three or more base stations and the tag to establish a system of equations to solve the tag's position and obtain an initial position estimate for the AGV. At the same time, the reflection path characteristics are used to infer environmental obstacle information. For example, the position and shape of the obstacle are calculated based on the reverse direction and distance of the reflection path, and the obstacle probability distribution of the three-dimensional grid map is updated. The three-dimensional grid map divides the storage environment into cubic grids with a side length of 0.1-0.5m. The obstacle probability value of each grid is 0 or 1, 0 indicates no obstacle, and 1 indicates an obstacle. The true TOF value is converted into a distance measurement value and trilateral positioning is performed to obtain the initial position estimate of the AGV, which is the basis for positioning. Updating the obstacle probability distribution of the three-dimensional grid map can make the map more accurately reflect environmental changes and provide reliable environmental information for subsequent positioning and path planning.

[0034] S1210: Based on the initial position estimate, the Kalman filter is used to fuse historical positioning trajectories to suppress random errors. The degree of match between the current positioning result and the global map is calculated. If the degree of match is below a preset threshold, a relocalization process is triggered. If the degree of match is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output. Based on the initial position estimate, a Kalman filter tool is invoked to fuse historical positioning trajectories to suppress random errors. The Kalman filter tool is a software module that integrates the Kalman filter algorithm and can be directly invoked to perform data fusion processing without manually programming the algorithm details. The historical positioning trajectory is a sequence of positioning coordinates of the AGV within the past [10 seconds, 30 seconds]. This selected time period reflects movement trends while minimizing processing speed due to excessive data volume. The degree of match is determined by calculating the similarity between the obstacle distribution and surface position of the current positioning result and the corresponding location in the global map. The value range of the degree of match is [0, 1], with 0 indicating a complete mismatch and 1 indicating a complete match. If the matching degree is lower than the preset threshold, the re-positioning process is triggered, that is, the above steps S122 to S128 are re-executed. The preset threshold range is [0.6, 0.8]. If the warehouse environment has high positioning accuracy requirements, such as the AGV needs to shuttle in a narrow channel, the threshold is 0.8; if the environment has slightly lower accuracy requirements, the threshold can be 0.6. The specific value of the preset threshold is set by technicians in this field according to actual conditions. When the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output. .

[0035] refer to Figure 3 In step S104, the local positioning coordinate generation step includes:

[0036] S1302: Acquire the environmental data, perform preprocessing on the environmental data, and extract features from the preprocessed environmental data using a multi-scale convolutional neural network to generate a feature map containing different features. Acquire environmental data acquired by an image acquisition device mounted on the AGV. The image acquisition device is a device installed on the AGV, such as a depth camera or binocular camera, and is used to capture image information of the warehouse environment. The environmental data is a color or grayscale image of the warehouse environment. Perform preprocessing on the environmental data, including image denoising, grayscaling, and distortion correction. Image denoising utilizes a Gaussian filter. The high-low filter is a linear smoothing filter that effectively removes Gaussian noise from images. The Gaussian kernel size and standard deviation determine the filter strength. The Gaussian kernel size can be 3×3 or 5×5. A 3×3 kernel size is suitable for mild noise, while a 5×5 kernel size is suitable for more severe noise. The standard deviation ranges from [0.5, 2] and can be adjusted based on noise intensity, with a larger value being used for high noise levels. Grayscaling converts a color image into a grayscale image by weighted averaging the pixel values ​​of each color channel. Distortion correction modifies the image using a correction formula configured by the image acquisition device based on its internal parameters, including focal length, principal point coordinates, and distortion coefficients. Preprocessing the environmental data eliminates interfering factors such as noise and distortion, improving image quality and providing more reliable data for subsequent feature extraction, as noise and distortion can affect feature accuracy and stability. A multi-scale convolutional neural network (CNN) extracts features from the preprocessed environmental data to generate a feature map containing a detailed hierarchy. The CNN comprises multiple convolutional layers of varying scales. A CNN is a deep learning model that uses convolution kernels of varying sizes to extract features at different scales, comprehensively capturing both local and global information. Convolutional layers are fundamental components of CNNs, extracting features by convolving the kernel with the input image. Convolution kernels are matrices used in convolution operations to extract specific image features, such as edges and textures. The convolution kernel sizes are 3×3, 5×5, and 7×7, respectively. Feature maps of different scales are extracted through the convolutional layers, such as a 3×3 convolution kernel for detail features and a 7×7 convolution kernel for overall contour features. Using this multi-scale convolutional neural network for feature extraction simultaneously captures both detailed and overall image features, improving feature richness and recognition.

[0037] S1304, based on the feature matching algorithm, match the features contained in the feature map with the features of the constructed local map, and identify the key landmarks in the feature map. Based on feature matching, match the features contained in the feature map with the features of the constructed local map, and calculate the Euclidean distance between the features. The Euclidean distance is the straight-line distance between two points in a multidimensional space, which is used to measure the similarity between features. The smaller the distance, the more similar the features. Features with a Euclidean distance less than a preset threshold are considered matching features, and the key landmarks in the feature map are identified. The key landmarks refer to objects or areas with uniqueness and stability in the storage environment, such as specific shelf numbers and goods of unique shapes. The key landmarks serve as positioning reference points for the AGV in the local environment, and the relative position of the AGV is calculated through the positional relationship of the key landmarks.

[0038] S1306, based on the key landmark positions and the AGV motion model, calculate the relative position change of the AGV in the local map. According to the pixel coordinates of the key landmarks in the image, combined with the internal and external parameters of the image acquisition device, the pixel coordinates are converted into three-dimensional coordinates in the local map coordinate system. The external parameters refer to the parameters that describe the relative position and posture relationship between the image acquisition device and the AGV, including the translation vector and the rotation matrix. The AGV motion model adopts a differential drive model, and the model state equation is: ,in 、 is the position coordinate of the AGV in the local map at time k, is the attitude angle of the AGV at time k, is the linear velocity of the AGV at time k-1, is the angular velocity of the AGV at time k-1, By comparing the changes in the three-dimensional coordinates of the key landmarks at adjacent moments and combining the AGV motion model, the linear speed of the AGV is solved using the least squares method. and angular velocity , and then calculate the relative position change of the AGV in the local map 、 、 The relative posture change refers to the change in the position (x, y) and attitude angle (θ) of the AGV in the local map coordinate system between two adjacent moments. The differential drive model is a common AGV motion model that realizes the steering and movement of the AGV by controlling the speed difference between the left and right wheels. Its motion state is determined by the linear speed. and angular velocity The least squares method is a mathematical optimization method that finds the best function match for fitting data by minimizing the sum of squared errors. Calculating the relative pose change enables real-time understanding of the AGV's motion state in the local environment, providing a basis for updating the local map and determining the AGV's real-time local positioning coordinates, thereby achieving precise positioning of the AGV within a local area.

[0039] S1308, update the local map according to the posture change information, and output the updated local map and the local positioning coordinates of the AGV. Take the current posture of the AGV as a reference, combined with the relative posture change 、 、 , calculate the predicted position of the AGV at the new moment. According to the predicted position at the new moment, convert the newly collected environmental features into the local map coordinate system, and add the converted coordinates to the local map. Delete the environmental features in the local map that are beyond the environmental range that the image acquisition device can capture, obtain the updated local map, and determine the final local positioning coordinates of the AGV in the local map based on the updated local map. , and output the updated local map and the local positioning coordinates Updating the local map can make the map reflect changes in the AGV's surrounding environment in real time, ensuring the map's accuracy and effectiveness; outputting the local positioning coordinates can provide location information for the AGV's local navigation and obstacle avoidance, and after fusion with the global positioning coordinates, it can further improve the AGV's overall positioning accuracy.

[0040] refer to Figure 4 The step of fusing the global positioning coordinates with the local positioning coordinates based on the Kalman filter algorithm includes: S1402, obtain fusion data and pre-process the fusion data; the fusion data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, the first covariance matrix represents the measurement error of the UWB positioning system, and the second covariance matrix represents the measurement error of the visual positioning system. The pre-processing includes verifying the data format and abnormal data processing. The global positioning data is obtained from the UWB positioning system through the communication interface. The global positioning data includes the global coordinates. and its corresponding first covariance matrix , obtaining local positioning data from the image acquisition device through a communication interface, the local positioning data including the local coordinates and its corresponding second covariance matrix . Verify the data format of the fused data, including time dimension synchronization, coordinate units unified as meters, attitude angle units as radians, and covariance matrix as a 3x3 symmetric matrix. Mark abnormal data such as those whose coordinates exceed the storage range and replace them with valid data from the previous moment, and store the preprocessed fused data in a data buffer. The fused data acquisition and preprocessing are the basis of fusion. Through the communication interface and the format verification, the fused data input into the Kalman filter is ensured to be valid and consistent. The abnormal data processing and the buffer design improve the fault tolerance of the filter, avoiding the influence of single-frame error data on the overall fusion effect.

[0041] S1404, initialize the Kalman filter and set the key parameters of the Kalman filter. The key parameters include the state variable dimension, state transfer matrix, sampling time, process noise covariance matrix and observation noise covariance matrix that affect the filtering effect. The state transfer matrix is ​​a matrix used to describe the law of change of the system state from the current moment to the next moment. It is generated in combination with the AGV motion model and is the core element of the Kalman filter prediction step. Its accuracy directly affects the reliability of the predicted state. The observation noise covariance matrix is ​​used to measure the uncertainty of the global coordinate data and the local coordinate data. The integrated matrix comprehensively reflects the reliability of the two data and is a key parameter in the Kalman filter update step. Set the state variable dimension , the state transfer matrix , the sampling time , the process noise covariance matrix Initial value, the observation noise covariance matrix Initial weight. The state variables The dimension is 6, among which 、 Indicates location, represents the attitude angle, 、 Indicates the linear speed, In the matrix operation of the Kalman filter, the state variables need to be calculated in the form of column vectors, such as the state transfer matrix Multiplication, the transpose symbol T clarifies the vector dimension of the variable, ensuring that the dimension of the matrix operation matches, for example, the state transfer matrix of 6×6 With 6×1 state variables The result of multiplication is still a 6×1 vector. The sampling time interval range is [0.1 seconds, 0.2 seconds]. The observation noise covariance matrix The initial weight includes the global coordinate data weight and the local coordinate data weight. The global coordinate data weight range is [0.4, 0.6], and the local coordinate data weight range is [0.4, 0.6]. The initial weight is set according to the average accuracy of the two coordinate data in the storage environment. The global coordinate data weight + the local coordinate data weight = 1. Initialize state variables Initialize the covariance by accessing the average value of the global coordinate data and the local coordinate data for the first time is the average value of the first covariance matrix and the second covariance matrix. 、 The parameter value of and stated The adjustment range is 0.5 to 2 times of the initial value to avoid excessive adjustment of the parameters causing filter instability. The adjustment takes effect in real time.

[0042] S1406: The Kalman filter triggers fusion calculation according to the sampling time interval. The latest global positioning coordinate data and the local positioning coordinate data are obtained from the data buffer, and the filter automatically updates the observation noise covariance matrix. Based on state variables With the state transfer matrix Perform prediction calculations to obtain the predicted status and the predicted covariance ; Combined with the observation matrix H k Update calculation and obtain the optimal state estimate through iterative calculation and its corresponding covariance matrix The updated optimal state estimate and the covariance matrix As the input of the next iteration, the prediction and update process is continuously circulated. At the same time, in each iterative calculation process, the filter will dynamically adjust the weights of the global coordinate data and the local coordinate data in the fusion calculation according to the real-time accuracy of the global coordinate data and the local coordinate data, so as to further optimize the accuracy and reliability of the fusion result.

[0043] S1408, obtaining the optimal state estimate after fusion The fused positioning coordinate information is extracted based on coordinate transformation and output to the AGV navigation control module via a communication interface, with the output frequency consistent with the sampling time interval. The coordinate transformation is achieved using preset coordinate transformation parameters, including translation vectors and rotation matrices, to ensure that the fusion results are output in a unified coordinate system.

[0044] Example 2 refer to Figure 5 The cold chain storage AGV positioning system according to an embodiment of the present invention includes: The first module is used to obtain measurement data, build a global map of the storage environment where the AGV is located based on the measurement data, and generate the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the storage environment and the mobile positioning tag carried by the AGV.

[0045] The second module is used to obtain environmental data, build a local map of the storage environment where the AGV is located based on the environmental data, and generate the local positioning coordinates of the AGV; the environmental data is collected by the image acquisition device carried by the AGV.

[0046] The third module is connected to the first module and the second module respectively through a communication connection, and is used to obtain the global positioning coordinates output by the global positioning module and the local positioning coordinates output by the local positioning module, and fuse the global positioning coordinates and the local positioning coordinates based on the Kalman filter algorithm to obtain and output the fused AGV positioning information.

[0047] More specifically, the first module includes: A multipath component identification unit is connected to a fixed UWB base station and a mobile positioning tag via a communication connection, and is used to obtain the measurement data, perform a time domain deconvolution transform on the measurement data to generate a time-reversed signal sequence, estimate the channel impulse response based on the time-reversed signal sequence, and identify multipath components including direct paths and reflected paths.

[0048] A direct path separation unit is connected to the multipath component identification unit via a communication connection, and is used to receive the multipath components output by the multipath component identification unit, extract the amplitude, phase and delay characteristic parameters of the multipath components, and separate the direct path signal from the multipath components based on the energy threshold and propagation time characteristics.

[0049] The true TOF extraction unit is connected to the direct path separation unit via a communication connection, and is used to receive the direct path signal output by the direct path separation unit, predict the reflection path and its characteristics based on the prior map information of the warehouse environment, construct an adaptive filter, perform interference rejection processing on the measurement signal according to the predicted reflection path, and extract the true TOF value.

[0050] An initial position calculation unit is connected to the real TOF extraction unit via a communication connection, and is used to receive the real TOF value output by the real TOF extraction unit, convert the real TOF value into a distance measurement value, perform three-sided positioning calculation based on the base station coordinates to obtain the AGV initial position estimate, use the reflection path characteristics to infer the environmental obstacle information, and update the obstacle probability distribution of the three-dimensional grid map.

[0051] A global coordinate optimization unit is connected to the initial position calculation unit via a communication connection, and is used to receive the AGV initial position estimate output by the initial position calculation unit. Based on the initial position estimate, the unit fuses the historical positioning trajectory based on the Kalman filter to suppress random errors, and calculates the matching degree between the current positioning result and the global map. If the matching degree is lower than a preset threshold, a relocation process is triggered. If the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output to the coordinate fusion module.

[0052] More specifically, the second module includes: The feature extraction unit is connected to the image acquisition device carried by the AGV through a communication connection, and is used to obtain the environmental data, perform preprocessing according to the environmental data, and extract features of the preprocessed environmental data based on a multi-scale convolutional neural network to obtain a feature map containing different features.

[0053] A key landmark identification unit is connected to the feature extraction unit via a communication connection, and is used to receive the feature map output by the feature extraction unit, match the features contained in the feature map with the features of the constructed local map based on a feature matching algorithm, and identify the key landmarks in the feature map.

[0054] The posture change calculation unit is connected to the key landmark identification unit through a communication connection, and is used to receive the key landmark information output by the key landmark identification unit, and calculate the relative posture change of the AGV in the local map based on the key landmark position and the AGV motion model.

[0055] The local coordinate output unit is connected to the posture change calculation unit via a communication connection, and is used to receive the relative posture change information output by the posture change calculation unit, update the local map according to the posture change information, and output the updated local map and the local positioning coordinates of the AGV to the coordinate fusion module.

[0056] More specifically, the third module includes: A data preprocessing unit is connected to the global coordinate optimization unit of the first module and the local coordinate output unit of the second module through a communication connection, and is used to obtain fused data and perform preprocessing; the fused data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, and the preprocessing includes verifying the data format and processing abnormal data.

[0057] A filter initialization unit is connected to the data preprocessing unit through a communication connection, and is used to receive the preprocessed data output by the data preprocessing unit, initialize the Kalman filter, and set the key parameters of the Kalman filter, wherein the key parameters include the state variable dimension, the state transfer matrix, the sampling time, the process noise covariance matrix, and the observation noise covariance matrix.

[0058] A fusion calculation unit is connected to the filter initialization unit through a communication connection, and is used to receive the initialization parameters output by the filter initialization unit, so that the Kalman filter triggers fusion calculation according to the sampling time interval to obtain the optimal state estimation after fusion.

[0059] A result output unit is connected to the fusion calculation unit through a communication connection, and is used to obtain the optimal state estimation after fusion, extract the fused positioning coordinate information based on coordinate transformation, and output the fused positioning coordinate information through a communication interface, with an output frequency consistent with the sampling time interval.

[0060] Compared with the prior art, the above embodiment 1 of the present application achieves the following technical effects:

[0061] At the global positioning level, the time-domain deconvolution transform and channel impulse response estimation are combined with energy thresholds and propagation time characteristics to separate direct path signals. Multipath interference is effectively suppressed through reflection path prediction and adaptive filtering based on a priori maps. Compared to traditional UWB positioning methods, this method significantly reduces multipath error. Furthermore, obstacle information is inferred using reflection path features, improving the obstacle recognition accuracy of the three-dimensional grid map. The global positioning coordinate update frequency is stable in dynamic environments, and relocalization is quickly triggered when the matching degree falls below a preset threshold, ensuring global positioning continuity in extreme low-temperature environments. Regarding local positioning, the multi-scale convolutional neural network significantly improves the success rate of feature extraction from preprocessed images, maintaining a high feature matching rate even in low-temperature and frosted scenes, a significant improvement over traditional visual SLAM algorithms. The combination of key landmark identification and the differential drive model improves the accuracy of relative pose change calculations. Local map updates are timely, addressing positioning drift caused by the lack of visual features in low-temperature environments. During the multi-source fusion phase, the Kalman filter algorithm effectively compensates for the temporal asynchrony of the two types of positioning data through dynamic weight adjustment. The fused positioning information output frequency is stable, with excellent positioning accuracy. This ensures high reliability in continuous positioning in complex scenarios such as areas with dense metal shelves and narrow alleyways. Compared to single-technology solutions, the probability of positioning failure is significantly reduced, fully meeting the cargo handling, path planning, and safe obstacle avoidance requirements of cold chain warehousing AGVs.

[0062] Compared with the prior art, the above-mentioned embodiment 2 of the present application effectively suppresses the multipath effect through the multipath component identification unit, direct path separation unit, true TOF extraction unit, initial position calculation unit, and global coordinate optimization unit in the first module, improves the global positioning accuracy and map update accuracy, and ensures the continuity of global positioning in extreme environments; the feature extraction unit, key landmark identification unit, posture change calculation unit, and local coordinate output unit in the second module enhance the visual feature extraction and matching capabilities in low-temperature scenes, and improve local positioning accuracy and map timeliness; the data preprocessing unit, filter initialization unit, fusion calculation unit, and result output unit in the third module achieve precise fusion of global and local positioning coordinates through the Kalman filter, improve positioning stability and reliability, and fully meet the high-precision positioning operation requirements of cold chain warehousing AGVs.

[0063] It should be noted that, for the sake of simplicity, the aforementioned embodiments are described as a series of action combinations. However, those skilled in the art should be aware that this application is not limited to the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously.

[0064] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course, it can also be implemented by hardware devices. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for enabling a terminal device (AGV body) to execute the method steps described in the embodiments of this application.

[0065] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A cold chain storage AGV positioning method, characterized in that: include: Acquire measurement data, construct a global map of the warehouse environment where the AGV is located based on the measurement data, and generate the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the warehouse environment and a mobile positioning tag carried by the AGV; Acquire environmental data, construct a local map of the storage environment where the AGV is located based on the environmental data, and generate local positioning coordinates of the AGV; the environmental data is collected by an image acquisition device carried by the AGV; The global positioning coordinates and the local positioning coordinates are obtained, and the global positioning coordinates and the local positioning coordinates are fused based on a Kalman filter algorithm to obtain and output the fused AGV positioning information.

2. The method according to claim 1, characterized in that The global positioning coordinate generation step comprises: Acquiring the measurement data, performing a time domain deconvolution transform on the measurement data to generate a time-reversed signal sequence; estimating a channel impulse response based on the time-reversed signal sequence, and identifying multipath components including a direct path and a reflected path; Extracting the amplitude, phase, and delay characteristic parameters of the multipath components; separating the direct path signal from the multipath components based on energy thresholds and propagation time characteristics; Based on the prior map information of the storage environment, the reflection path and its characteristics are predicted; an adaptive filter is constructed to remove interference from the measurement signal according to the predicted reflection path and extract the true TOF value; Convert the true TOF value into a distance measurement value, perform trilateral positioning calculation based on the base station coordinates, and obtain the AGV initial position estimate; use the reflection path characteristics to infer the environmental obstacle information and update the obstacle probability distribution of the three-dimensional grid map; According to the initial position estimate, based on Kalman filtering, historical positioning trajectories are integrated to suppress random errors; the matching degree between the current positioning result and the global map is calculated. If the matching degree is lower than a preset threshold, the relocalization process is triggered; if the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output.

3. The method according to claim 1, wherein the step of generating local positioning coordinates comprises: Acquiring the environmental data, performing preprocessing on the environmental data, and extracting features of the preprocessed environmental data based on a multi-scale convolutional neural network to obtain a feature map containing different features; Based on a feature matching algorithm, matching the features included in the feature map with the features of the constructed local map to identify key landmarks in the feature map; Based on the key landmark positions and the AGV motion model, calculating the relative position change of the AGV in the local map; The local map is updated according to the posture change information, and the updated local map and the local positioning coordinates of the AGV are output.

4. The method according to claim 1, wherein The step of fusing the global positioning coordinates and the local positioning coordinates based on the Kalman filter algorithm includes: Acquire fused data and preprocess the fused data; the fused data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, the first covariance matrix represents the measurement error of the UWB positioning system, and the second covariance matrix represents the measurement error of the visual positioning system, and the preprocessing includes verifying the data format and abnormal data processing; Initialize the Kalman filter and set the key parameters of the Kalman filter, which include the state variable dimension, state transfer matrix, sampling time, process noise covariance matrix, and observation noise covariance matrix that affect the filtering effect; The Kalman filter triggers fusion calculation according to the sampling time interval to obtain the optimal state estimation after fusion; The optimal state estimate after fusion is obtained, fused positioning coordinate information is extracted based on coordinate transformation, and the fused positioning coordinate information is output to the AGV navigation control module through a communication interface, with an output frequency consistent with the sampling time interval.

5. A cold chain warehousing AGV positioning system, characterized in that: include: The first module is used to obtain measurement data, construct a global map of the warehouse environment where the AGV is located based on the measurement data, and generate the global positioning coordinates of the AGV; the measurement data is obtained by measuring the distance between a fixed UWB base station deployed in the warehouse environment and a mobile positioning tag carried by the AGV; The second module is used to obtain environmental data, construct a local map of the storage environment where the AGV is located based on the environmental data, and generate the local positioning coordinates of the AGV; the environmental data is collected by the image acquisition device carried by the AGV; The third module is connected to the first module and the second module respectively through a communication connection, and is used to obtain the global positioning coordinates output by the global positioning module and the local positioning coordinates output by the local positioning module, and fuse the global positioning coordinates and the local positioning coordinates based on the Kalman filter algorithm to obtain and output the fused AGV positioning information.

6. The system according to claim 5, characterized in that The first module includes: a multipath component identification unit connected to a fixed UWB base station and a mobile positioning tag via a communication connection, configured to obtain the measurement data, perform a time domain deconvolution transform on the measurement data to generate a time-reversed signal sequence, estimate a channel impulse response based on the time-reversed signal sequence, and identify multipath components including direct paths and reflected paths; a direct path separation unit, connected to the multipath component identification unit via a communication connection, configured to receive the multipath components output by the multipath component identification unit, extract amplitude, phase, and delay characteristic parameters of the multipath components, and separate the direct path signal from the multipath components based on an energy threshold and propagation time characteristics; a true TOF extraction unit, connected to the direct path separation unit via a communication connection, configured to receive the direct path signal output by the direct path separation unit, predict the reflection path and its characteristics based on prior map information of the storage environment, construct an adaptive filter, perform interference rejection processing on the measurement signal according to the predicted reflection path, and extract the true TOF value; an initial position calculation unit, connected to the real TOF extraction unit via a communication connection, configured to receive the real TOF value output by the real TOF extraction unit, convert the real TOF value into a distance measurement value, perform a trilateration calculation based on the base station coordinates to obtain an initial position estimate of the AGV, infer environmental obstacle information using reflection path features, and update the obstacle probability distribution of the three-dimensional grid map; The global coordinate optimization unit is connected to the initial position calculation unit via a communication connection, and is used to receive the AGV initial position estimate output by the initial position calculation unit. Based on the initial position estimate, the unit fuses the historical positioning trajectory based on the Kalman filter to suppress random errors, and calculates the matching degree between the current positioning result and the global map. If the matching degree is lower than a preset threshold, the relocation process is triggered. If the matching degree is greater than or equal to the preset threshold, the optimized AGV global positioning coordinates are output to the coordinate fusion module.

7. The system according to claim 5, characterized in that The second module includes: a feature extraction unit, connected to the image acquisition device carried by the AGV via a communication connection, configured to acquire the environmental data, perform preprocessing on the environmental data, and extract features of the preprocessed environmental data based on a multi-scale convolutional neural network to obtain a feature map containing different features; a key landmark identification unit, connected to the feature extraction unit via a communication connection, configured to receive a feature map output by the feature extraction unit, and match features contained in the feature map with features of the constructed local map based on a feature matching algorithm to identify key landmarks in the feature map; a posture change calculation unit, connected to the key landmark identification unit via a communication connection, configured to receive key landmark information output by the key landmark identification unit, and calculate a relative posture change of the AGV in the local map based on the key landmark position and the AGV motion model; The local coordinate output unit is connected to the posture change calculation unit via a communication connection, and is used to receive the relative posture change information output by the posture change calculation unit, update the local map according to the posture change information, and output the updated local map and the local positioning coordinates of the AGV to the coordinate fusion module.

8. The system according to claim 5, wherein: The third module includes: A data preprocessing unit is connected to the global coordinate optimization unit of the first module and the local coordinate output unit of the second module through a communication connection, and is used to obtain fused data and perform preprocessing; the fused data includes the global positioning coordinates and their corresponding first covariance matrix, the local positioning coordinates and their corresponding second covariance matrix, and the preprocessing includes verifying the data format and processing abnormal data; a filter initialization unit, connected to the data preprocessing unit via a communication connection, configured to receive the preprocessed data output by the data preprocessing unit, initialize the Kalman filter, and set key parameters of the Kalman filter, wherein the key parameters include state variable dimension, state transition matrix, sampling time, process noise covariance matrix, and observation noise covariance matrix; a fusion calculation unit connected to the filter initialization unit via a communication connection, configured to receive the initialization parameters output by the filter initialization unit, so that the Kalman filter triggers fusion calculation according to the sampling time interval to obtain a fused optimal state estimate; A result output unit is connected to the fusion calculation unit through a communication connection, and is used to obtain the optimal state estimation after fusion, extract the fused positioning coordinate information based on coordinate transformation, and output the fused positioning coordinate information through a communication interface, with an output frequency consistent with the sampling time interval.

9. A cold chain storage AGV positioning device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method steps according to any one of claims 1 to 4 when executing the program; AGV body, wherein the system described in any one of claims 5 to 8 is integrated into the AGV body.

10. A readable storage medium, characterized in that: The readable storage medium stores a cold chain warehouse AGV positioning program, and when the cold chain warehouse AGV positioning program is executed by the processor, the method steps as described in any one of claims 1 to 4 are implemented.

Images