A multi-sensor linkage monitoring method

Abstract

The application discloses a kind of multi-sensor linkage monitoring methods, it is related to logistics management technical field, comprising the following steps:Ⅰ, create sensor state monitoring timer, open independent communication thread, file descriptor is created in the running process of communication thread, and sensor file is checked;Ⅱ, communication thread judges the specified type of each sensor file, and then according to sensor category, data variation frequency and shelf business importance, dynamically adjusts each data acquisition period and processing order;The application can avoid misjudgment under fault-tolerant path, enhance the robustness of monitoring, realize key task priority and efficient use of resources, improve the accuracy of fusion results, and realize fine-grained shelf state recognition, guarantee the real-time performance, stability and fault tolerance of semantic map, improve the controllability of business operation.

Description

Technical Field

[0001] This invention relates to the field of logistics management technology, specifically to a multi-sensor linkage monitoring method. Background Technology

[0002] In modern warehousing and logistics systems, electronic shelving has gradually replaced traditional shelving, becoming an important carrier for intelligent management. With the rapid development of intelligent warehousing and logistics, shelving monitoring systems have become a key link in improving material management efficiency and safety. Traditional single-sensor monitoring methods suffer from limited sensing range, isolated data, and insufficient reliability, making it difficult to meet the demands for real-time performance, accuracy, and intelligence in complex warehousing environments. Multi-sensor linkage monitoring technology, by integrating multimodal information from weight sensors, infrared sensors, and vision modules, can achieve refined identification and dynamic management of shelving status. However, the heterogeneity, spatiotemporal asynchrony, and real-time processing pressure of multi-sensor data bring new challenges to system design and implementation.

[0003] Existing multi-sensor linkage monitoring methods are prone to misjudgment, reducing the robustness of monitoring, failing to guarantee the priority of critical tasks and efficient use of resources, and having low accuracy of fusion results. Furthermore, they cannot achieve fine-grained shelf status recognition, and the real-time performance, stability, and fault tolerance of semantic maps are poor, reducing the controllability of business operations. Therefore, we propose a multi-sensor linkage monitoring method. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies and provide a multi-sensor linkage monitoring method.

[0005] This invention proposes a multi-sensor linkage monitoring method, and the technical solution adopted to solve the technical problem is as follows:

[0006] Ⅰ. Create a sensor status monitoring timer, start an independent communication thread, and the communication thread creates a file descriptor and checks the sensor file during the running process;

[0007] II. The communication thread determines the specified type of each sensor file, and then dynamically adjusts the data acquisition cycle and processing order according to the sensor type, data change frequency and shelf business importance.

[0008] III. During continuous monitoring, data from various sensors are read in real time, and data from different sources are aligned and fused to build a unified semantic map of the shelving.

[0009] IV. Process the fused sensor data in real time through edge nodes, and send a signal containing the sensor position and status to the main thread when the sensor data is confirmed to have changed.

[0010] V. The main thread receives signals, determines the shelf status and outputs coordinates and status information. Based on the determination result, it manages the shelf status and establishes the corresponding virtual shelf model.

[0011] VI. Based on the shelf status, execute the processing strategy and update the shelf map data in the local JSON file, while adjusting the sensor sampling frequency, communication cycle and processing strategy.

[0012] After confirming the readability of the sensor file, this invention first preprocesses the raw metadata of the sensor and converts it into structured features. Simultaneously, a target sensor template set is maintained offline. Real-time features are compared with the templates, and device identification and logging are performed based on similarity scores. The judgment result is returned to the upper-level scheduling module to determine whether to establish a monitoring timer. If the sensor is identified as a target, its historical observation sequence is collected, outlier removal, frame loss processing, and change frequency calculation are performed. A comprehensive priority score is then formed by combining the importance weight of the shelf business, and converted into a sampling period through logistic mapping. Finally, multi-threaded scheduling and anomaly rollback control are performed according to priority. Afterwards, the sensor time field undergoes unified time zone correction and compensation, time offset is aligned using cross-correlation methods, and resampling is performed to a common time grid. Confidence weights are generated by combining uncertainty assessment. Subsequently, multimodal features are extracted, converted into semantic probabilities using a classifier, and weighted fusion is performed to form the posterior probability of the shelf unit. In spatial modeling, a 3D grid and topological units are generated based on the warehouse layout, recording geometric centers and adjacency relationships to construct an initial semantic map. Dynamic updates and conflict rollback are achieved through exponential smoothing and neighborhood consistency constraints. If a "sitting" event is detected, a scanning session is initiated for barcode scanning or manual input verification. If successful, the semantic map and database are updated; if unsuccessful, manual intervention is prompted and a log is recorded. If a "leaving" event is detected, the storage cell is marked as empty and resources are released. Simultaneously, the sampling utility value is calculated and mapped to a target sampling frequency using a sigmoid function. This value is then adjusted based on a preset sampling frequency. Based on the adjusted target sampling frequency, shelf information is collected in real time. This approach avoids misjudgments under fault-tolerant paths, enhances monitoring robustness, prioritizes critical tasks, and ensures efficient resource utilization. It also improves the accuracy of fusion results, enables fine-grained shelf status recognition, guarantees the real-time performance, stability, and fault tolerance of the semantic map, and enhances the controllability of business operations. Attached Figure Description

[0013] Figure 1 This is a framework diagram of a multi-sensor linkage monitoring method. Detailed Implementation

[0014] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Embodiment 1

[0016] The embodiment of the present invention provides a multi-sensor linkage monitoring method. Refer to Figure 1 , Figure 1 which is a framework diagram of a multi-sensor linkage monitoring method provided by the embodiment of the present invention. The method includes the following steps:

[0017] Create a sensor status monitoring timer, start an independent communication thread. The communication thread creates a file descriptor during operation and checks the sensor files.

[0018] Specifically, the communication thread inputs the paths of each sensor as parameters into the system interface call function. If the call is successful, a non-negative integer is returned as the file descriptor. If the call fails, -1 is returned. Then, the system checks the accessibility of each sensor file according to the returned file descriptor, and at the same time checks whether it has read permission. If the sensor file is inaccessible, it enters the error handling process. At the same time, the communication thread converts the error code into text form and outputs it after splicing with the manual prompt. Otherwise, the communication thread takes the corresponding file descriptor as the input, reads the basic information of the corresponding sensor device from the sensor file according to the length limit, and stores the result in a structure or cache. The system calculates the similarity between the target sensor device information and the information actually read by the communication thread. If the similarity is higher than the preset recognition threshold, it is determined as the target sensor file and enters the subsequent monitoring process. Otherwise, the communication thread exits.

[0019] It should be further noted that the specific calculation formula of the file descriptor is as follows: , where represents the file descriptor; represents the system function mapping of the file open operation; represents the path of the sensor device file.

[0020] The specific determination formula of file accessibility is as follows: , where represents the file accessibility result. If , it means the file can be normally accessed; if This indicates that the file is inaccessible. This represents the read permission flag; if read permission is granted... ,otherwise ; This represents the file descriptor validity function, if Returns 1 if the condition is met, otherwise returns 0. This represents the permission validity function, when Returns 1 if the condition is met, otherwise returns 0.

[0021] The communication thread determines the specified type of each sensor file, and then dynamically adjusts the data acquisition cycle and processing order based on the sensor type, data change frequency, and the importance of the shelf business.

[0022] Specifically, after confirming that the sensor file is readable, the communication thread reads the raw metadata from the sensor file in text or binary format, removes leading and trailing whitespace and invisible characters from each raw metadata, standardizes the case of the text data in each raw metadata, performs synonym normalization on the manufacturer name and model, and parses the version number. Then, it fills missing fields in each raw metadata with the successfully read model field or default placeholders to record missing flags. The preprocessed metadata is converted into corresponding structured features through structured mapping. Then, in the offline stage, multiple sets of target sensor templates are obtained based on known target device samples or manually defined by the device specification table. The target sensor template set is periodically maintained and updated in the configuration or database. The real-time generated... The structured features are compared with each set of templates in the target sensor template set, and the similarity score between them is calculated. If the similarity score exceeds the preset strict threshold, the corresponding sensor file is directly determined to be the target sensor type and the template is recorded. If it is higher than the preset tolerance threshold but lower than the preset strict threshold, the fault tolerance path is entered, more fields are read, the reading length is extended, and the verification is performed again. If it is lower than the preset tolerance threshold, it is determined to be a non-target sensor and recorded. If there are multiple templates with similarity scores higher than the preset strict threshold, the target sensor template with the highest similarity score is selected as the final determination result. The determination results of each sensor file are written to the local log and returned to the upper-level scheduling module. The upper-level scheduling module selects to establish a monitoring timer based on the determination results.

[0023] Specifically, once a target sensor is identified, its historical observation sequence is collected, and simultaneously based on 3D... The principle is to remove outliers from historical observation sequences, use interpolation or marking as unavailable to handle frame drops in sensor data collection, and calculate the absolute change of each pair of adjacent readings in the historical observation sequence. Simultaneously, normalize these changes according to the actual time difference to obtain the corresponding rate of change per unit time. Then, perform a moving average on each obtained rate of change per unit time to generate a frequency of change. Establish a shelf business importance weight table, and update the importance weight of each shelf business in the table regularly based on recent inbound / outbound frequencies, stockout alarm counts, and manual priority. Before prioritizing, query the corresponding importance weight in the shelf business importance weight table based on the current sensor's shelf location. If the weight data is outdated, trigger a weight refresh, and recalculate based on the inventory turnover rate of the most recent 24 hours. The corresponding weights are recalculated. Based on the sensor category, change frequency, and importance weight of the shelf business, the comprehensive priority score of each sensor is calculated. The obtained comprehensive priority scores are converted into actual sampling periods through logistic mapping. The sensors are then prioritized according to their comprehensive priority scores. Time slots for sensors with the same priority are allocated according to timestamps or round-robin strategies. After the update is completed, the sampling periods and comprehensive priority scores of each sensor are returned to the local scheduling configuration. In a multi-threaded environment, the communication thread acquires the configuration information lock of the corresponding sensor and releases the lock after the update. If an exception occurs during the update or application process, the previous stable period is retained and the event is recorded, and a rollback is performed. At the same time, after each round of updates, the current update operation is recorded in the audit log in a transactional manner.

[0024] It should be further explained that the original metadata specifically includes the manufacturer string, product ID, hardware version, interface type, device description text, etc.

[0025] The specific formula for calculating the frequency of change is as follows: In the formula, Represents the frequency of change; Represents the total time; Representing the The first numerical observation read; Representing the The first numerical observation read; Represents absolute value operations.

[0026] The specific formula for calculating the actual sampling period is as follows: In the formula, This represents the time interval for the next time the sensor is read. This represents the shortest allowed sampling period; This represents the longest allowed sampling period; The lower limit of the range of values ​​for the sensor's overall priority score; Represents the upper limit of the value range of the comprehensive priority score of the sensor; Represents the comprehensive priority score of the sensor.

[0027] The specific calculation formula for priority sorting is as follows: , where, Represents the sensor The processing priority ranking within the current scheduling period; Represents the sensor The comprehensive priority score of; Represents the priority scores of other sensors in the system ; Represents an indicator function, which is 1 when the condition is satisfied and 0 otherwise.

[0028] Embodiment 2

[0029] The embodiment of the present invention provides a multi-sensor linkage monitoring method. Refer to Figure 1 ,<​​​​​​Specifically, the time field of each sensor data is parsed and verified, and the time data with time zone and daylight saving time biases is converted to a unified time zone. Using a fixed reference time as a benchmark, the time of all sensor data is converted to the standard time unit of this benchmark. Based on the calibration board, known reference points, or manual annotation, a calibration matrix is ​​established for each type of sensor. At the same time, the rigid transformation between each sensor coordinate system and the global coordinate system of the shelf is determined. One or more pairs of overlapping time windows of sensors are selected, and the cross-correlation across time shifts is calculated. The time shift with the largest correlation is selected as the offset of the corresponding sensor. Then, the time series of the offset is calculated and fitted through a sliding window to obtain the corresponding compensation value. Then, the clocks of each sensor are synchronized based on each compensation value. The discrete observations of different sensors at their respective time points are unified into a common time grid through resampling. The uncertainty of each resampled sensor data is evaluated, and the uncertainty of each resampled sensor is back-mapped to generate the confidence weight of each sensor data. Finally, the modal features corresponding to each sensor data are extracted. Then, a lightweight classifier is used to convert each modal feature into a corresponding semantic probability. Using a common time point as the unit, the semantic probabilities or confidence weights from each sensor are weighted and fused, and the posterior probability of each storage cell in the shelf at the current time is output. Based on the warehouse floor plan and shelf layout, each topological cell in the shelf is defined, and the global coordinate system of each shelf is fixed with calibration points. A three-dimensional grid covering the entire area of ​​interest is generated. The geometric center, adjacent cell list and shelf ID of each topological cell are calculated, and the grid metadata is recorded to generate an initial semantic map. The posterior probability of a single storage cell at the current time is written into the semantic map, and the semantic map is iteratively updated between new observations and historical states through exponential smoothing. At the same time, consistency constraints are applied to the adjacent cells of the corresponding storage cells. Then, the difference in probability distribution of the affected grid cells before and after the update is calculated. If the difference exceeds a preset threshold or triggers a violation of business rules, the current update is marked as a conflict and a rollback or degradation is triggered. At the same time, the conflict event is recorded in the audit log.

[0032] The fused sensor data is processed in real time through edge nodes, and when changes in sensor data are confirmed, a signal containing the sensor's position and status is sent to the main thread.

[0033] The main thread receives signals, determines the shelf status, and outputs coordinates and status information. Based on the determination result, it manages the shelf status and establishes the corresponding virtual shelf model.

[0034] Based on the shelf status, the processing strategy is executed, and the shelf map data in the local JSON file is updated. At the same time, the sampling frequency, communication cycle and processing strategy of the sensors are adjusted.

[0035] Specifically, when the main thread receives a "seat" event reported by the sensor, it initiates a scanning session for the corresponding shelf. Simultaneously, the communication thread establishes a session information matching the shelf locally, including the session ID, shelf ID, and user prompt status. It records the session start time and triggers a scanning prompt on the dot matrix screen. It also sets the maximum waiting time and the maximum allowed number of retries for this scan. Within the scanning window, it receives multiple attempt results from the barcode scanner or manual input. The communication thread deduplicates and sorts the received attempt results chronologically, and evaluates the scan quality using statistical indicators. If the scan quality is higher than the preset acceptable threshold, it is considered a "successful shelf scan." If the maximum waiting time is exceeded and the scan quality is lower than the preset acceptable threshold, a re-evaluation is performed. If the scan still fails after reaching the preset maximum number of retries, the session information is marked as "scan failed / pending review," and a timeout message and manual intervention instructions are displayed on the dot matrix screen. Simultaneously, an audit log is written. When the scan is successful and the parsed material ID and shelf location are verified... After verification, the communication thread reads the corresponding entry from the local semantic map based on the material ID and location in the session, and constructs update information including material ID, quantity, timestamp, and source trust level. Through weighted fusion, the update information is merged in memory using a transactional approach, and then written back to disk or embedded database all at once. If the write fails, the memory state is rolled back and the error update information is recorded. After the sensor detects the "away" event, the communication thread immediately marks the shelf as away in the local map, displays an away prompt on the dot matrix screen, marks the shelf as empty and releases the corresponding resources. Based on the shelf's business importance, change risk, and resource cost, the sampling utility value corresponding to the shelf is calculated. The sampling utility value of the shelf is non-linearly transformed using the sigmoid function to convert it to the corresponding target sampling frequency, and it is checked whether the target sampling frequency is between the preset minimum and maximum sampling frequencies. If it exceeds the constraint range, the target sampling frequency is adjusted to the minimum or maximum sampling frequency, and data is collected again based on the latest target sampling frequency.

[0036] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A multi-sensor linkage monitoring method, characterized in that, Includes the following steps: I. Create a sensor status monitoring timer, start a separate communication thread, and during the execution of the communication thread, create a file descriptor and check the sensor file; the specific steps are as follows: S1.1: The communication thread inputs the paths of each sensor as parameters into the system interface calling function. If the call is successful, it returns a non-negative integer as a file descriptor. If the call fails, it returns -1. Then, the system checks the accessibility of each sensor file based on the returned file descriptor and checks whether it has read permissions. S1.2: If the sensor file is inaccessible, the error handling process is initiated. At the same time, the communication thread converts the error code into text form, concatenates it with the manual prompt, and outputs it. Otherwise, the communication thread takes the corresponding file descriptor as input, reads the basic information of the corresponding sensor device from the sensor file according to the length limit, and stores the result in a structure or cache. S1.3: The system calculates the similarity between the target sensor device information and the information actually read by the communication thread. If the similarity is higher than the preset recognition threshold, it is determined to be the target sensor file and enters the subsequent monitoring process; otherwise, the communication thread exits. II. The communication thread determines the specified type of each sensor file, and then dynamically adjusts the data acquisition cycle and processing order according to the sensor type, data change frequency and shelf business importance. III. During continuous monitoring, data from various sensors are read in real time, and data from different sources are aligned and fused to build a unified semantic map of the shelving. IV. Process the fused sensor data in real time through edge nodes, and send a signal containing the sensor position and status to the main thread when the sensor data is confirmed to have changed. V. The main thread receives signals, determines the shelf status and outputs coordinates and status information. Based on the determination result, it manages the shelf status and establishes the corresponding virtual shelf model. VI. Based on the shelf status, execute the processing strategy and update the shelf map data in the local JSON file, while adjusting the sensor sampling frequency, communication cycle and processing strategy.

2. The multi-sensor linkage monitoring method according to claim 1, characterized in that, The specific calculation formula for the file descriptor mentioned in S1.1 is as follows: In the formula, Represents a file descriptor; System function mappings representing file open operations; Represents the file path of the sensor device; The specific formula for determining the accessibility of a file as described in S1.1 is as follows: In the formula, The result represents the file accessibility result, if This indicates that the file can be accessed normally; if This indicates that the file is inaccessible. This represents the read permission flag; if read permission is granted... ,otherwise ; This represents the file descriptor validity function, if Returns 1 if the condition is met, otherwise returns 0. This represents the permission validity function, when Returns 1 if the condition is met, otherwise returns 0.

3. The multi-sensor linkage monitoring method according to claim 2, characterized in that, The specific steps for the communication thread to determine the specified class of each sensor file in step II are as follows: S2.1: After confirming that the sensor file is readable, the communication thread reads the raw metadata from the sensor file in text or binary format, removes leading and trailing whitespace and invisible characters from each raw metadata, unifies the case of the text data in each raw metadata, performs synonym normalization on the manufacturer name and model, parses the version number, and then uses the successfully read same model field or default placeholder to fill the missing fields in each raw metadata and record the missing flag. S2.2: The preprocessed metadata is converted into corresponding structured features through structured mapping. Then, in the offline stage, multiple sets of target sensor templates are obtained by statistical analysis of known target device samples or by manual definition from the device specification table. The target sensor template set is regularly maintained and updated in the configuration or database. S2.3: Compare the real-time generated structured features with each group of templates in the target sensor template set, calculate the similarity score between the two. If the similarity score exceeds the preset strict threshold, the corresponding sensor file is directly determined to be the target sensor type and the template is recorded. If it is higher than the preset tolerance threshold but lower than the preset strict threshold, the fault tolerance path is entered, more fields are read, the reading length is extended, and the verification is performed again. If it is lower than the preset tolerance threshold, it is determined to be a non-target sensor and recorded. S2.4: If multiple templates have similarity scores higher than the preset strict threshold, the target sensor template with the highest similarity score is selected as the final judgment result. The judgment results of each sensor file are written to the local log and returned to the upper-level scheduling module. The upper-level scheduling module selects to establish a monitoring timer based on the judgment results.

4. The multi-sensor linkage monitoring method according to claim 3, characterized in that, The specific steps for dynamically adjusting the data acquisition cycle and processing order described in step II are as follows: S3.1: Once identified as a target sensor, collect the sensor's historical observation sequence, and simultaneously based on 3 The principle is to remove outliers from the historical observation sequence, use interpolation or marking as unavailable to handle frame loss in each sensor acquisition, calculate the absolute change of each pair of adjacent readings in the historical observation sequence, and normalize them according to the actual time difference to obtain the corresponding rate of change per unit time. Then, the obtained rate of change per unit time is processed by moving average to generate the frequency of change. S3.2: Establish a shelf business importance weight table, and adjust each indicator according to the recent inbound and outbound frequency, out-of-stock alarm number, and manual priority. Update the importance weight of each shelf business in the shelf business weight table in a timely manner. Then, before priority calculation, query the corresponding importance weight in the shelf business importance weight table according to the current shelf location of the sensor. If the weight data is too old, trigger a weight refresh and recalculate its corresponding weight according to the inventory turnover rate of the most recent 24 hours. S3.3: Calculate the comprehensive priority score of each sensor based on sensor type, change frequency, and importance weight of shelf business. Convert the obtained comprehensive priority scores into actual sampling periods through logistic mapping. Sort the sensors according to their comprehensive priority scores and then allocate time slots for sensors with the same priority according to timestamp or polling strategy. S3.4: After the update is completed, the sampling period and overall priority score of each sensor are returned to the local scheduling configuration. In a multi-threaded environment, the communication thread acquires the configuration information lock of the corresponding sensor and releases the lock after the update. If an exception occurs during the update or application process, the previous stable period is retained and the event is recorded, and a rollback is performed. At the same time, after each round of update, the current update operation is recorded in the audit log in a transactional manner.

5. The multi-sensor linkage monitoring method according to claim 4, characterized in that, The specific calculation formula for the frequency of change mentioned in S3.1 is as follows: In the formula, Represents the frequency of change; Represents the total time; Representing the The first numerical observation read; Representing the The first numerical observation read; Represents absolute value operation; The specific calculation formula for the actual sampling period mentioned in S3.3 is as follows: In the formula, This represents the time interval for the next time the sensor is read. This represents the shortest allowed sampling period; This represents the longest allowed sampling period; The lower limit of the range of values ​​for the sensor's overall priority score; This represents the upper limit of the range of values ​​for the sensor's overall priority score; The overall priority score representing the sensor; The specific calculation formula for the priority sorting described in S3.3 is as follows: In the formula, Representative sensor Ranking of processing priorities within the current scheduling cycle; Representative sensor The overall priority score; Represents other sensors in the system Priority score; This represents an indicator function; it is 1 if the condition is true, and 0 otherwise.

6. The multi-sensor linkage monitoring method according to claim 5, characterized in that, The specific steps for constructing a unified shelf semantic map as described in step III are as follows: S4.1: Analyze and verify the time field of each sensor data, and convert the time with time zone and daylight saving time deviation to a unified time zone. Use a fixed reference time as a benchmark, convert the time of all sensor data to the standard time unit of this benchmark, and establish a calibration matrix for each type of sensor based on the calibration board, known reference points or manual annotations. At the same time, determine the rigid transformation between each sensor coordinate system and the global coordinate system of the shelf. S4.2: Select one or more pairs of overlapping time windows of sensors, calculate the cross-correlation across time shifts, and select the time shift with the largest correlation as the offset of the corresponding sensor. Then, calculate the time series of the offset through a sliding window and fit it to obtain the corresponding compensation value. Then, synchronize the clocks of each sensor based on each compensation value. S4.3: Discrete observations from different sensors at their respective time points are unified into a common time grid through resampling. The uncertainty of each resampled sensor data is evaluated, and the uncertainty of each resampled sensor is reverse-mapped to generate confidence weights for each sensor data. Then, the modal features corresponding to each sensor data are extracted. Afterward, each modal feature is converted into a corresponding semantic probability through a lightweight classifier. Using the common time point as the unit, the semantic probabilities or confidence weights from each sensor are weighted and fused, and the posterior probability of each shelf unit in the shelf at the current time is output. S4.4: Based on the warehouse floor plan and shelf layout, define each topological unit in the shelf, fix the global coordinate system of each shelf with calibration points, generate a three-dimensional grid covering the entire area of ​​interest, calculate the geometric center, adjacent unit list and shelf ID of each topological unit, and record grid metadata to generate an initial semantic map. S4.5: Write the posterior probability of a single cell at the current moment into the semantic map, and use exponential smoothing to iteratively update the semantic map between new observations and historical states. At the same time, apply consistency constraints to adjacent cells of the corresponding cell. Then calculate the difference in probability distribution of affected cells before and after the update. If the difference exceeds a preset threshold or triggers a violation of business rules, mark the current update as a conflict and trigger rollback or downgrade processing. At the same time, record the conflict event in the audit log.

7. The multi-sensor linkage monitoring method according to claim 1, characterized in that, The specific steps in step VI, which involve executing a processing strategy based on the shelf status and updating the shelf map data in the local JSON file, are as follows: S5.1: When the main thread receives the "sitting" event reported by the sensor, it starts a scanning session for the corresponding shelf. At the same time, the communication thread establishes session information matching the shelf locally, including session ID, shelf ID and user prompt status, records the session start time and triggers a scanning prompt on the dot matrix screen, and sets the maximum waiting time and the maximum number of retries allowed for this scan. S5.2: Within the scanning window, multiple attempt results are received from the barcode scanner or manual input. The communication thread performs deduplication and time-series sorting on each received attempt result, and evaluates the scanning quality through statistical indicators. If the scanning quality is higher than the preset qualified threshold, it is determined as "successful rack entry scanning". S5.3: If the maximum waiting time is exceeded and the scan quality is lower than the preset qualified threshold, a new judgment will be made. If the scan still fails after reaching the preset maximum number of retries, the session information will be marked as "scan failed / pending review", and a timeout message and manual intervention guidance will be displayed on the dot matrix screen, and the information will be written to the audit log. S5.4: When the scan is successful and the material ID and shelf location are parsed and verified, the communication thread reads the corresponding entry in the local semantic map based on the material ID and location in the session, and constructs update information containing material ID, quantity, timestamp and source trust level. The update information is merged in memory in a transactional manner through weighted fusion, and then written back to disk or embedded database at once. If the write fails, the memory state is rolled back and the error update information is recorded. S5.4: After the sensor detects the "leaving the seat" event, the communication thread immediately marks the shelf as leaving the shelf in the local map, displays the leaving the shelf prompt on the dot matrix screen, marks the shelf as empty and releases the corresponding resources, and calculates the sampling utility value corresponding to the shelf based on the shelf's business importance, change risk and resource cost. S5.5: The sampling utility value of the shelf is nonlinearly transformed by the sigmoid function to the corresponding target sampling frequency, and it is checked whether the target sampling frequency is between the preset minimum and maximum sampling frequencies. If it exceeds the constraint range, the target sampling frequency is adjusted to the minimum or maximum sampling frequency, and data is collected again based on the latest target sampling frequency.

Images