Intelligent chain store patrol method and system based on multi-source video stream

Abstract

The invention discloses a multi-source video stream-based chain store intelligent tour inspection method and system, and belongs to the technical field of chain store tour inspection. The method comprises the following steps: extracting a target area from a multi-source video stream, calibrating the target area as a reference area, judging an image state to distinguish a steady-state time period, and classifying cameras to calibrate an inspection camera and a blind compensation camera; constructing an inspection video list based on the steady-state time period; performing behavior feature analysis on the inspection video list to identify abnormal behavior features; and after the abnormal behavior characteristics are confirmed to be true and effective through key checking, outputting an alarm item. According to the method, the abnormal behavior characteristics can be accurately identified, false alarms are effectively reduced, and the intelligent level of shop patrol, the accuracy of risk identification and the reliability of alarm items are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of chain store inspection technology, specifically relating to a chain store intelligent inspection method and system based on multi-source video streams. Background Technology

[0002] As an efficient and standardized business operation model, chain stores have occupied a pivotal position in many industries such as retail, catering, and services. Their large-scale and branded characteristics make them a key carrier connecting production and consumption and enhancing market vitality. In order to ensure the standardization of chain store operations, the uniformity of service quality, and the consistency of brand image, store inspection, as a core management activity, has irreplaceable strategic value for ensuring the normal operation of stores, optimizing customer experience, and enhancing overall competitiveness.

[0003] Current technology for store inspections relies on manual review and analysis of surveillance videos over extended periods. This requires personnel to maintain continuous concentration, leading to visual fatigue and mental fatigue, resulting in low efficiency. Furthermore, the involvement of subjective human factors makes it difficult to guarantee the accuracy and consistency of inspection results. For subtle, hidden anomalies, such as improper merchandise placement, unsanitary areas, or unusual customer behavior, omissions or misjudgments are common, impacting daily store operations and potentially causing safety hazards or damage to brand reputation. In addition, the coverage and frequency of manual inspections are limited by labor costs and time, making it difficult to achieve comprehensive, real-time monitoring of all stores at all times in a large, widely distributed chain system.

[0004] To address the aforementioned issues, this invention provides a method and system for intelligent store inspection of chain stores based on multi-source video streams. Summary of the Invention

[0005] The purpose of this invention is to provide a smart store inspection method or system for chain stores based on multi-source video streams, in order to solve the problem that under the existing inspection methods, staff may overlook some abnormalities in details due to fatigue or negligence.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for intelligent store inspection of chain stores based on multi-source video streams, comprising the following steps: Behavioral feature analysis includes: using comparison algorithms to analyze changes in the image content in the inspection video list; statistically analyzing behavioral information in the inspection images, including pedestrian flow, activity paths, and areas of residence, to obtain behavioral fluctuation values; and comparing the behavioral fluctuation values ​​with preset allowable ranges to identify abnormal behavioral features. When behavioral feature analysis identifies abnormal behavioral characteristics in the inspection video list that deviate from the preset tolerance range, the following actions are taken: Perform key verification on abnormal behavior characteristics; and after confirming that the abnormal behavior characteristics are real and valid through key verification, output an alarm item.

[0007] Preferably, prior to behavioral feature analysis, the process further includes constructing a list of inspection videos. The steps for constructing the list of inspection videos include: Extract the target region from multi-source video streams and label it as the reference region; The image status of the reference area is judged to distinguish the steady-state period; Based on the steady-state time periods of the cameras identified as inspection cameras, a time-ordered list of inspection videos is constructed.

[0008] Preferably, the step of judging the image state of the reference area to distinguish the steady-state time period includes: Extract edge feature information from the current image; Extract historical video content whose time difference from the current video time point is less than a preset time threshold; When the overlap between edge feature information and historical image content is higher than a preset overlap threshold, the current image state is judged as a steady-state period.

[0009] Preferably, the method further includes classifying the cameras, which includes: Determine the correspondence between multiple cameras corresponding to multi-source video streams; If the correspondence is consistent, then the camera will be designated as an inspection camera; If the correspondence is inconsistent, the camera will be labeled as a blind spot camera.

[0010] Preferably, it also includes dynamic adjustment of the camera, the dynamic adjustment of the camera including: When the video content of the inspection camera is detected to be obstructed to a degree greater than the preset obstruction threshold or the signal strength is less than the preset signal strength threshold, the inspection camera will be marked as a camera to be replaced. Preferably, the dynamic adjustment of the camera also includes: automatically searching for cameras in other coverage areas as alternatives to reconstruct the monitoring area.

[0011] Preferably, the key verification of abnormal behavioral characteristics includes: Anomaly reports are generated based on abnormal behavioral characteristics obtained from behavioral feature analysis; Perform secondary verification on the time periods and regions mentioned in the anomaly report to confirm that the abnormal behavior characteristics are genuine and valid.

[0012] This invention also discloses a smart store inspection system for chain stores based on multi-source video streams, comprising: The store status monitoring module is used to extract the target area from multi-source video streams and mark it as the reference area, and to judge the image status of the reference area to distinguish between steady-state time periods and offset time periods. The behavioral risk analysis module is used to build a list of inspection videos in response to the steady-state period detected by the store status monitoring module, and to perform behavioral feature analysis on the inspection video list to identify abnormal behavioral features caused by the deviation of behavioral fluctuation values ​​from the preset allowable range. The risk verification and alarm module is used to perform key verification on the abnormal behavior characteristics identified by the response behavior risk analysis module, and output alarm items after confirming that the abnormal behavior characteristics are real and valid.

[0013] Preferably, the store status monitoring module is also used for: The correspondence between multiple cameras corresponding to multi-source video streams is determined so that the cameras can be identified as inspection cameras or blind spot cameras. When the video content of the inspection camera meets the preset replacement conditions, the inspection camera is marked as a replacement camera to trigger the search for alternative cameras and the reconstruction of the monitoring area.

[0014] Preferably, the behavioral risk analysis module is used for: A comparison algorithm is used to analyze changes in the image content of the inspection video list; By statistically analyzing behavioral information in the inspection images, including pedestrian flow, activity paths, and areas of residence, behavioral fluctuation values ​​are obtained. These behavioral fluctuation values ​​are then compared with preset allowable ranges to identify abnormal behavioral characteristics.

[0015] Beneficial effects 1. This invention enables intelligent store patrol for chain stores by acquiring multi-source video streams, calibrating target areas, judging image status, constructing inspection video lists, and analyzing behavioral features. Specifically, it uses image recognition technology to extract target areas from the camera's shooting area and calibrate them as reference areas, distinguishing between steady-state and offset time periods. Simultaneously, it determines the correspondence between cameras, calibrates inspection cameras and blind spot cameras, and then counts the steady-state time periods of inspection cameras as inspection time points. Based on this, it constructs an inspection video list, thereby enabling comprehensive and real-time monitoring of chain stores, ensuring the continuity and effectiveness of the inspection process, and significantly improving the intelligence level and coverage of store patrols.

[0016] 2. This invention analyzes the behavioral characteristics and assesses risks in a list of inspection videos, enabling precise identification of abnormal behavioral features. Specifically, it employs a comparison algorithm to analyze changes in image content across different time periods within the inspection video list. By statistically analyzing behavioral information such as pedestrian traffic, activity paths, and areas of residence, it obtains behavioral fluctuation values ​​and compares them with preset safety thresholds to determine whether the behavioral features are abnormal. This effectively avoids the subjectivity and lag inherent in manual inspections, enabling early warning and precise location of store anomalies. It significantly improves the accuracy and timeliness of risk identification, providing strong support for store safety management.

[0017] 3. By performing key verification and alarm item output, this invention can effectively reduce false alarms and improve the reliability of alarm items. It performs secondary confirmation on all abnormal items obtained from behavioral feature analysis, generates an abnormality report to be verified, and performs specific verification processing on the time period and area involved. If the key verification confirms that the abnormality is real and valid, an alarm item is generated and a prompt is output. This can filter out false alarms caused by occasional events or system errors, ensuring that only real abnormalities that have been strictly confirmed will be output as alarm items. This avoids unnecessary waste of resources and interference, and significantly improves the accuracy of alarm items and the reliability of the system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the intelligent store inspection method for chain stores based on multi-source video streams according to the present invention; Figure 2 This is a module diagram of the intelligent store inspection system for chain stores based on multi-source video streams according to the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0020] Example 1 See Figure 1 This embodiment provides a smart store inspection method for chain stores based on multi-source video streams, applicable to real-time automatic inspection and risk identification of different functional areas or sensitive areas in chain stores. The multi-source video streams refer to a collection of real-time video data simultaneously collected and transmitted by multiple cameras at different locations and from different angles, which together constitute comprehensive coverage of the monitored area.

[0021] Specifically, the method includes the following steps: The system acquires real-time multi-source video streams from multiple cameras deployed in different locations within the store. For each video stream, image content analysis is performed to identify and extract key monitoring locations in each image. The identification results are marked as target areas, and these target areas are uniformly designated as baseline areas, serving as reference standards for subsequent inspections, image status judgments, and behavioral feature analysis.

[0022] When extracting target areas, the system prioritizes assessing the match between the camera's shooting angle and the store's designated key areas. Specifically, for critical areas such as the checkout counter and high-value merchandise display areas, cameras providing clear, complete, and unobstructed shooting angles are selected as the baseline to ensure monitoring accuracy. Furthermore, when identifying target areas, the system automatically recognizes and processes the shape boundaries of adjacent areas. That is, when two adjacent shelf areas are identified as independent target areas, they are merged according to preset merging rules such as distance between areas and functional relevance. This optimizes recognition accuracy, reduces redundant information, and improves the accuracy of subsequent behavioral feature analysis, ensuring monitoring covers all key locations and laying the foundation for subsequent image status analysis and camera selection. The preset merging rules specifically refer to the criteria the system uses when identifying adjacent target areas to determine whether to merge these independently identified areas into a larger logical area based on pre-defined conditions such as distance between areas and functional relevance.

[0023] Determine the image state of the currently displayed screen in the reference area. If the screen content remains visually stable, it is marked as a steady-state period, which is a continuous period of time in which the video screen content in the reference area remains visually stable and there are no obvious changes in structure or lighting conditions. If the screen structure or lighting conditions change significantly, it is marked as an offset period, which is a continuous period of time in which the video screen content in the reference area changes significantly in structure or lighting conditions.

[0024] When determining whether an image is in a steady state, edge feature information in the current image is extracted as a reference. Specifically, the contour information of the image is extracted by analyzing the gradient changes of pixel gray values. Edge feature information is the feature extracted from the region in the image where the pixel gray value or color value changes significantly. It is usually used to describe the contour, boundary and texture of an object and can effectively reflect the stability of the structure and layout of objects in the image. The aforementioned pixel gray value gradient change refers to the rate and direction of gray value change between adjacent pixels in the image. Through calculation, the edge and texture information of the image can be revealed. Simultaneously, historical image content with a time difference less than a preset time threshold from the current image is extracted, and overlap matching is performed. Overlap matching can be achieved by calculating the structural similarity index or pixel-level difference between two images to quantify the degree of consistency of their content. The preset time threshold refers to a pre-set time value used by the system to limit the maximum time difference between historical images and the current image when comparing historical image content.

[0025] If the overlap is higher than the preset overlap threshold, the image presentation is classified into a steady state period, indicating that the content of the image remains basically unchanged. The preset overlap threshold is a pre-set similarity value limit used to determine whether the content of two images remains basically unchanged when performing the overlap matching. If the overlap is lower than a preset overlap threshold, the image is categorized into an offset time period, indicating a significant change in the image content. A correlation determination is made based on the relative position of the camera and the target area in the image and the coverage of the shooting angle. Specifically, it's determined whether the camera's shooting angle covers the target area. If they match, the camera is defined as an inspection camera, providing a clear, complete, and unobstructed shooting angle, primarily responsible for routine inspections of the target area. If they don't match, but the target area is still partially covered or supplemented in the image, the corresponding camera is defined as a blind spot camera, used to fill in monitoring blind spots and ensure monitoring continuity. If, during the inspection, an obstruction greater than a preset obstruction threshold or a signal strength lower than a preset signal strength threshold is found in the video content... The phenomenon is that the inspection camera will be marked as waiting to be replaced. At this time, other cameras covering the area will be automatically searched for as replacements, and the monitoring area will be reconstructed to ensure the continuity of monitoring. Specifically, if the main inspection camera fails due to malfunction or obstruction, the adjacent backup camera will be immediately activated to take over the monitoring task of the area. If it is confirmed that an area cannot be effectively covered by the inspection camera for a long time, such as due to store layout adjustments or camera installation location restrictions, resulting in a persistent monitoring blind spot in a specific area, then the area will be designated as the blind spot camera monitoring area, and the blind spot camera will be assigned to monitor it specifically. If it is verified that all cameras to be replaced can restore normal monitoring, such as after the faulty camera is repaired or the obstruction is removed, its video stream quality will return to normal, then there is no need to introduce the blind spot camera, thereby optimizing resource allocation.

[0026] The preset occlusion threshold mentioned above refers to a pre-set numerical limit used to determine whether the degree of occlusion in the video content reaches the level that requires processing or alarm, while the preset signal strength threshold refers to a pre-set numerical limit used to determine whether the video stream signal quality meets the normal working requirements.

[0027] The system statistically analyzes the continuous steady-state states of inspection cameras, identifying key nodes within these periods as inspection time points. Specifically, based on the statistical analysis of the continuous steady-state states of the inspection cameras, key time points are selected within these periods for extracting video frames as reference samples. Several video frames are then extracted before and after these corresponding inspection time points as reference samples. These samples are then compared with image samples from other inspection cameras to construct the first inspection data with a clear time order. The specific steps for constructing the inspection video list include: Starting with the camera area marked by the inspection cameras, all cameras containing the target area are sequentially searched and sorted within the shooting range to form a sorted list. Specifically, based on the geographical location and coverage of the cameras on the store floor plan, the cameras are sorted in order from left to right or from front to back. The video content of the corresponding time period in the sorted list is extracted as the input data for the inspection task to ensure the completeness and logic of the inspection video list. The above operation not only considers the continuous images of a single camera, but also incorporates the shooting perspective information of multiple adjacent cameras on the same area for comparison and integration.

[0028] For the first inspection data, image content analysis is used to identify the activity path, stopping area, and movement speed of people in the image content frame by frame, forming behavioral information. That is, the data such as the activity path, stopping area, and movement speed of people identified from video frames through image content analysis are used to quantify and describe the numerical values ​​of behavioral characteristics. Behavioral feature analysis includes frame-by-frame comparison of video frames within a selected time period, performing offset analysis on any two frames (i.e., calculating the motion vectors of pixels to capture minute movements of objects in the frame and analyzing the degree of change in the frame content), where the pixel motion vector refers to the direction and magnitude of the movement of each pixel in the image content between consecutive frames, used to describe the trajectory and speed of the object in the frame; extracting comparison differences at set time intervals, which reflect the degree of change in the frame content; and generating numerical indicators reflecting behavioral changes, such as average pixel change rate and motion vector amplitude, based on the extracted results, and marking areas with significant fluctuations for subsequent behavioral feature analysis. The average pixel change rate refers to the average rate of change of the grayscale or color value of pixels in the video frame within a certain time period, used to measure the intensity of dynamic changes in the frame; the motion vector amplitude refers to the magnitude of the motion vector of the pixel, reflecting the speed of the object's movement. A set of preset thresholds is set to determine whether the behavior is abnormal. The identified behavior information is compared with the preset thresholds to determine whether it belongs to the activity path under normal monitoring, as follows: If the behavioral information is within the preset allowable range, it is considered a normal behavioral parameter; if the behavioral characteristics include situations such as excessive stay time, abnormally short path, or repeated abnormal activity area, and the behavioral information deviates from the preset threshold from the preset allowable range, it is considered an abnormal parameter, and the abnormal time node and location are recorded.

[0029] The preset allowable range is the normal fluctuation range set by the behavioral information.

[0030] For all parameters identified as abnormal, they are aggregated into objects to be verified and automatically redirected to the secondary verification process. The secondary verification process includes re-extracting the original video of the time period corresponding to the abnormal node, calculating the offset change value, and comparing it with the image content under the historical normal state. The offset change value is a numerical value of the degree of change in the image content calculated in the secondary verification process by comparing the original video of the time period corresponding to the abnormal node with the image content under the historical normal state.

[0031] Before generating the final alarm item, a secondary verification operation needs to be performed on the time period and corresponding area marked as pending verification in the anomaly report content. This is to confirm the accuracy and necessity of the anomaly. The secondary verification operation includes re-analyzing the video frames within the time period and cross-verifying them with other sensor data such as access control records and POS transaction data to generate verification images and verification parameters. The verification image is formed by stitching or overlaying multiple key frames during the period when the abnormal behavior occurred, and the verification parameters include the duration of the abnormal behavior, the number of people involved, and the intensity of the behavior. The verification image and verification parameters are used to further confirm the accuracy and necessity of the anomaly and avoid false alarms. If the offset change value is significantly higher than the set baseline, and abnormal nodes appear continuously or are densely distributed, the abnormality is confirmed to have alarm significance. Finally, an alarm item is generated and an alarm signal and detailed report are sent to the store or management center, including the alarm area, abnormal time, abnormal parameter type, and suggested handling method.

[0032] In the actual implementation process, a verification mechanism was also set up to mitigate false alarms: If the anomaly is initially determined to be invalid and fails to meet the effective standard during the secondary verification, or is suspected to be caused by factors such as changes in ambient lighting or customer obstruction, the anomaly will be judged as an invalid alarm and no alarm signal will be output, thereby improving the overall reliability and practicality.

[0033] The method described in this embodiment can be deployed and executed for chain stores of different sizes and different monitoring equipment platforms. It can automatically allocate monitoring area resources, dynamically extract monitoring indicators, and reasonably identify real abnormal behavior characteristics, significantly improving the efficiency and security of intelligent store patrols for chain stores.

[0034] Alarm significance refers to the severity and authenticity of an abnormal event that meets the standard for sending an alarm signal to the store or management center. Alarm item refers to the alarm record that is finally generated after confirming that the abnormality has the aforementioned alarm significance, and includes detailed information such as alarm area, abnormal time, abnormal parameter type, and suggested handling method.

[0035] The above description is a detailed description of the embodiments of the present invention and should not be construed as a limitation on the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope defined in the claims. Various modifications and variations made by those skilled in the art without departing from the concept and scope of the present invention are all within the scope of protection of the present invention.

[0036] Example 2 To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0037] See Figure 2 This embodiment provides a chain store intelligent inspection system based on multi-source video streams. The system is used to execute the chain store intelligent inspection method based on multi-source video streams described above. It can automatically identify and verify abnormal behaviors in the store, realize efficient and accurate remote inspection, reduce manual inspection costs and reduce false alarms.

[0038] In its implementation, this system can be deployed on cloud servers, edge computing devices, or local server clusters, and communicates with multiple cameras deployed in various chain stores via a network. Logically, the system can be divided into the following collaborative modules: The store status monitoring module is configured to preprocess and assess the status of multi-source video streams from multiple cameras within the chain stores, providing a stable and reliable data foundation for subsequent behavioral feature analysis.

[0039] In a specific execution process, this module extracts pre-defined key areas from multi-source video streams, specifically store entrances, cash registers, main shelf areas, etc., and marks these areas as the target areas. To ensure the consistency of the analysis benchmark, the image features of the target areas in the initial or standard state are further marked as the benchmark areas.

[0040] The image state of the reference area is continuously judged to distinguish between the steady-state period and the offset period. Specifically, the edge feature information of the current image is extracted, and historical image content with a time difference of less than a preset time threshold of preferably 5 seconds from the current image time point is retrieved. By calculating the overlap between the current edge feature information and the historical image content, when the overlap is higher than a preset overlap threshold, it indicates that the basic elements such as the camera viewpoint and scene layout have not changed significantly, and the current image state is judged as a steady-state period. Conversely, if the overlap is lower than the threshold, it is judged as an offset period. The video data of this period may not be suitable for conventional behavior analysis due to reasons such as camera movement or drastic scene changes.

[0041] Furthermore, this module is also responsible for classifying and dynamically managing cameras, and determining the correspondence between multiple cameras corresponding to multi-source video streams. Specifically, by analyzing their field of view overlap, geometric position relationship, etc., if multiple cameras can stably cover the same scene or related scenes, their correspondence is determined to be consistent, and these cameras are marked as inspection cameras, serving as the main data source for behavioral feature analysis. If the coverage area of ​​a camera is not stably associated with other cameras or is mainly used to cover blind spots, their correspondence is determined to be inconsistent, and they are marked as blind spot filler cameras.

[0042] During operation, the module also performs dynamic adjustments to the cameras and continuously monitors the video content of the inspection cameras. When it detects that the video screen is obstructed and the degree of obstruction is greater than a preset obstruction threshold, or when the video signal strength is less than a preset signal strength threshold due to network fluctuations or other reasons, these situations meet the preset replacement conditions. At this time, the inspection camera will be marked as a replacement and the search mechanism will be automatically triggered to try to call up blind spot cameras covering similar areas or other available cameras as replacements to reconstruct the monitoring area and ensure the continuity and integrity of data collection.

[0043] The core task of the behavioral risk analysis module is to conduct in-depth analysis of video content after the store status monitoring module has confirmed the steady-state period in order to identify potential abnormal behavioral characteristics.

[0044] Specifically, in response to the steady-state period detected by the store status monitoring module, a list of inspection videos arranged in chronological order is constructed based on the video streams identified as inspection cameras within the steady-state period. This list of inspection videos constitutes the basic dataset for subsequent behavioral feature analysis.

[0045] Behavioral feature analysis is performed on the inspection video list. One or more comparison algorithms are used to analyze the changes in the image content in the list in order to capture dynamic information in the scene. Specifically, through target detection and tracking technology, the behavioral information in the inspection images, including the flow of people representing the number of people passing through a specific area per unit time, the activity path representing the movement trajectory of customers or employees in the store, and the residence area defined as the area where customers or employees stay for more than a certain period of time, is continuously statistically analyzed and quantified to obtain real-time behavioral fluctuation change values ​​that reflect the dynamics of the store.

[0046] The calculated behavioral fluctuation value is compared with a preset tolerance range. This preset tolerance range is a normal behavior pattern range defined based on historical data, store operation standards, or specific inspection rules (such as zero foot traffic during non-business hours). When the detected change value deviates from this preset tolerance range, that is, non-zero foot traffic is detected during late night hours, or an abnormally long period of stay is detected in a specific area, this situation is identified as an abnormal behavior feature and passed to the next module for processing.

[0047] The risk verification and alarm module, as the system's decision-making and output end, is responsible for the final confirmation of the abnormal behavioral characteristics identified by the behavioral risk analysis module, and only triggers an alarm after confirming that they are real and valid, thereby effectively filtering out false alarms caused by factors such as algorithms and environment.

[0048] In response to the abnormal behavioral characteristics identified by the behavioral risk analysis module, a structured anomaly report will be generated based on the detailed information of the characteristics, such as the time, location, and specific behavioral description.

[0049] A key verification procedure was initiated for this anomaly report. The core of this procedure is to perform a secondary verification operation to confirm the authenticity and validity of the abnormal behavior characteristics. The secondary verification operation includes: calling an analysis model with higher computational complexity but also higher accuracy to re-analyze the time period and area involved in the anomaly report; or, retrieving videos of related areas taken by the blind spot cameras within the same time period for cross-comparison to eliminate possible misjudgments caused by a single camera or a single algorithm. The aim is to ensure the authenticity of the anomaly through multi-dimensional and multi-method verification.

[0050] After verifying the abnormal behavior characteristics through key checks and confirming that they are genuine and valid, an alarm item is finally output. This alarm item can be presented in various forms according to the preset configuration, such as sending a push notification containing an abnormal snapshot and video clip to the mobile device of the store manager, highlighting the abnormal event on the large screen of the central monitoring system, or automatically generating a detailed inspection abnormality log for subsequent auditing and review.

[0051] Through the collaborative work of the aforementioned store status monitoring module, behavioral risk analysis module, and risk verification and alarm module, the system described in this embodiment can construct a complete closed loop from data preprocessing and behavioral feature analysis to risk verification. It can automatically and around the clock conduct intelligent store inspections of chain stores, not only promptly detecting abnormal situations but also ensuring the accuracy of alarms through a rigorous verification mechanism. This greatly improves the management efficiency and security level of chain stores and is suitable for the retail industry with a large number of branches that require standardized remote management.

[0052] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-source video stream-based intelligent store inspection method for chain stores, characterized in that, Comprising: behavior feature analysis, comprising: using a comparison algorithm to perform change analysis on image content in the inspection video list; counting behavior information including crowd flow, activity path and residence area in the inspection image to obtain a behavior fluctuation change value; comparing the behavior fluctuation change value with a preset allowable range to identify abnormal behavior characteristics; When the behavior fluctuation change value in the inspection video list deviates from the preset allowable range to form an abnormal behavior characteristic through the behavior feature analysis, performing: performing key review on the abnormal behavior characteristic; and outputting an alarm item after confirming that the abnormal behavior characteristic is real and effective through the key review. 2.The multi-source video stream based intelligent store inspection method for chain stores according to claim 1, wherein, Before the behavior feature analysis, it further comprises constructing the inspection video list, and the steps of constructing the inspection video list comprise: extracting a target area from the multi-source video stream and labeling it as a reference area; judging the image state of the reference area to distinguish a steady state period; based on the steady state period labeled as the inspection camera, constructing a time-sequenced inspection video list. 3.The multi-source video stream based intelligent store tour method for chain stores according to claim 2, wherein, The step of judging the image state of the reference area to distinguish a steady state period comprises: extracting edge feature information of the current image frame; extracting historical frame content with a time difference from the current frame time point less than a preset time threshold; when the coincidence degree of the edge feature information and the historical frame content is higher than a preset coincidence degree threshold, the current image state is judged as a steady state period. 4.The multi-source video stream based intelligent store inspection method for chain stores of claim 2, wherein, It further comprises classifying the cameras, and classifying the cameras comprises: judging the corresponding relationship of the plurality of cameras corresponding to the multi-source video stream; if the corresponding relationship is consistent, the camera is labeled as an inspection camera; if the corresponding relationship is inconsistent, the camera is labeled as a blind filling camera.

5. The method of claim 4, wherein, It further comprises dynamic adjustment of the camera, and the dynamic adjustment of the camera comprises: when it is monitored that the video content of the inspection camera appears a phenomenon that the blocking degree is greater than a preset blocking threshold or the signal strength is less than a preset signal strength threshold, the inspection camera is labeled as a to-be-replaced state.

6. The method of claim 1, wherein, The dynamic adjustment of the camera further comprises: automatically searching for other cameras covering the area as a replacement to reconstruct the monitoring area.

7. The method of claim 1, wherein, The key review on the abnormal behavior characteristic comprises: forming an abnormal report from the abnormal behavior characteristic obtained by the behavior feature analysis; performing a secondary verification operation on the time period and the area involved in the abnormal report to confirm that the abnormal behavior characteristic is real and effective.

8. A multi-source video stream-based intelligent store inspection system for chain stores, characterized in that, Comprising: a store state monitoring module for extracting a target area from the multi-source video stream and labeling it as a reference area, and judging the image state of the reference area to distinguish a steady state period and a deviation period; a behavior risk analysis module for, in response to the steady state period monitored by the store state monitoring module, constructing an inspection video list, and performing behavior feature analysis on the inspection video list to identify abnormal behavior characteristics formed by a behavior fluctuation change value deviating from a preset allowable range; a risk review and alarm module for, in response to the abnormal behavior characteristics identified by the behavior risk analysis module, performing key review, and outputting an alarm item after confirming that the abnormal behavior characteristic is real and effective. 9.The multi-source video stream based intelligent store inspection system for chain stores of claim 8, wherein, The store state monitoring module is further used for: The corresponding relationship of multiple cameras corresponding to multiple video streams is judged to calibrate the cameras as inspection cameras or blind-spot supplement cameras; When it is monitored that the video content of the inspection camera meets the preset condition to be replaced, the inspection camera is calibrated as a state to be replaced to trigger the search of a replacement camera and the reconstruction of a monitoring area. 10.The multi-source video stream based intelligent store inspection system for chain stores of claim 8, wherein, The behavior risk analysis module is configured to: adopt a comparison algorithm to analyze the changes in image content in the inspection video list; statistically analyze the behavior information including the pedestrian flow, activity path and residence area in the inspection image to obtain a behavior fluctuation change value, and compare the behavior fluctuation change value with a preset allowable range to identify abnormal behavior characteristics.

Images