A low-power consumption camera interframe differential compression storage method and system

Abstract

The application provides a low-power consumption camera interframe differential compression storage method and system, which is applied to the field of data processing; the application establishes a reference frame cache by associating a monitoring video frame with a key frame anchor point, and identifies a picture change degree by using an interframe differential mechanism; only representative frame anchor points and index data are retained in a long-time static monitoring scene, repeated storage of a large number of similar video frames is avoided, local storage space occupation is effectively reduced, meanwhile, a key frame collection period is dynamically adjusted and a change event feature set is generated, so that the system can adaptively adjust a data collection strategy according to a scene change condition, and abnormal event traceability is ensured while invalid data transmission is reduced.

Description

Technical Field

[0001] This invention relates to the field of data processing, and in particular to an inter-frame differential compression storage method and system for a low-power camera. Background Technology

[0002] With the development of smart security and IoT technologies, low-power cameras are widely used in scenarios requiring long-term monitoring, such as home security, community surveillance, park management, and field monitoring. To extend the device's battery life, these cameras typically employ low-power operating modes, such as AOV (Always-On Video) mode, enabling the device to reduce power consumption while maintaining continuous monitoring capabilities.

[0003] However, surveillance videos exhibit distinct scene characteristics: most of the time, the monitored footage remains static or changes minimally, only showing significant changes when people, vehicles, or other targets enter the monitored area. Existing video storage solutions typically store video data continuously at a fixed frame rate or compress continuous video streams using video coding standards such as H.264 and H.265. While these encoding methods can reduce the video bitrate, for long-term static scenes, a large number of essentially identical video frames still need to be continuously stored, making it difficult to completely eliminate redundant data by relying solely on encoding prediction. Summary of the Invention

[0004] This invention aims to solve the problems of how to reduce the storage and transmission of redundant video data in static scenes, improve storage utilization, reduce device energy consumption and cloud storage costs, and provides a low-power camera inter-frame differential compression storage method and system.

[0005] The present invention employs the following technical means to solve the technical problem: This invention provides an inter-frame differential compression storage method for a low-power camera, comprising: Based on the surveillance video frames collected by the camera within a preset period, the surveillance video frames are associated with pre-stored key frame anchor points to establish a corresponding reference frame cache, and the inter-frame difference value between the surveillance video frames and the reference frame cache is calculated. Determine whether the inter-frame difference value is greater than a preset change threshold; If not, sparse sampling is performed on the still image according to a preset time interval, frame anchor representative points are extracted from the monitoring video frame to form corresponding monitoring index data, and the acquisition cycle of the next key frame is dynamically adjusted according to the changing trend of the inter-frame difference value within several consecutive sampling periods. Through several consecutive changing frames, a corresponding set of change event features is generated. The monitoring index data specifically includes timestamps, device status information and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction and target dwell time. Determine whether the set of change event features meets the preset abnormal event triggering conditions; If the conditions are met, cached video data within a preset time range before the change event is triggered is collected. The cached video data is then spliced ​​with the real-time video stream after the change event is triggered to obtain a complete event video. Based on the event data of the complete event video, the complete event video is compressed and managed using a keyframe indexing method. According to the target motion trajectory formed by the complete event video during the abnormal event, the preset inter-frame difference judgment model is dynamically updated. The event data specifically includes importance level, storage capacity, and network status.

[0006] Furthermore, before the step of sparsely sampling still images according to a preset time interval, extracting frame anchor representative points from the monitoring video frames, and forming corresponding monitoring index data, the method further includes: Based on the image difference results acquired in several consecutive acquisition cycles, the stable parameters within the corresponding time window are calculated, and the corresponding scene stable sequence is generated by combining the continuous stable parameters between adjacent time windows. Determine whether the stable sequence of the scenario meets the preset stability maintenance conditions; If not, a transitional cache set is established, and the current collected data is temporarily stored in the transitional cache set. Continuous consistency analysis is performed on the transitional cache set to calculate the degree of local consistency between each collection moment. Based on the degree of local consistency, the corresponding stability correction coefficient is obtained. The scene stability sequence is corrected based on the stability correction coefficient to obtain the corrected stability sequence. The corresponding stable evolution trajectory is generated through the corrected stability sequence.

[0007] Furthermore, the step of dynamically adjusting the acquisition period of the next key frame based on the changing trend of the inter-frame difference value within several consecutive sampling periods, and generating a corresponding set of change event features through several consecutive changing frames, further includes: Based on the preset associated data of the continuously changing frames, a corresponding change association chain is established, wherein the associated data specifically includes the order of occurrence, duration, spatial distribution and location migration relationship; Determine whether the changed association chain satisfies the preset continuous association condition; If not, the set of change event features is split into several independent change units, the abnormal change information is divided into several independent change units, the persistence influence coefficient of the independent change units is calculated, the processing priority of the independent change units is rearranged according to the persistence influence coefficient, the corresponding candidate event queue is constructed according to the processing priority, the candidate event queue is stored in the set to be verified, new image information is dynamically received during subsequent acquisition, incremental matching is performed on the candidate event queue, the candidate event queue is expanded and updated through the continuous association relationship formed by the new image information and the candidate event queue, and the corresponding set of change event features is regenerated. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

[0008] Furthermore, the step of collecting cached video data within a preset time range before the triggering of the change event, and then splicing the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, further includes: Based on the end node of the cached video and the start node of the real-time video, the association features between the end node and the start node are identified, and video connection parameters are generated according to the association features. Determine whether the video stitching parameters meet the preset continuous association conditions; If not, the process reverses the search to find the end position of the cached video and the start position of the real-time video, searches for the associated video segment with the start frame of the real-time video in the cache area, divides the corresponding splicing boundaries according to the search results, performs transition correction on the video data between the splicing boundaries, dynamically adjusts the consistency of motion trajectories between adjacent video frames, and recombines the corrected video data with the real-time video stream to generate a continuous event video.

[0009] Furthermore, the step of determining whether the inter-frame difference value is greater than a preset change threshold also includes: Based on the difference data between the current video frame and the reference video frame, generate inter-frame difference description parameters; Determine whether the difference description parameter exceeds a preset change limit; If not, low-amplitude change information in the inter-frame difference description parameters is collected, persistent background change features are extracted from the low-amplitude change information, the background change features are matched with environmental change features saved in the historical monitoring period, and the environmental change attribution of the monitoring video frame is dynamically classified according to the matching result, corresponding environmental adaptation parameters are generated, and the update period of the reference video frame is adjusted according to the environmental adaptation parameters. The environmental change attribution specifically includes periodic environmental changes and known environmental changes.

[0010] Furthermore, the step of determining whether the set of change event features meets the preset abnormal event triggering conditions also includes: Based on the event behavior description parameters corresponding to the set of changed event features, the event behavior description parameters are matched with the historical behavior data corresponding to the current monitoring area to obtain the degree of difference between the current changed behavior and the normal behavior pattern. Based on the degree of difference, the event credibility evaluation value is calculated. Determine whether the credibility evaluation value of the event reaches the preset credibility level; If so, then generate event confirmation information corresponding to the event credibility evaluation value, and based on the event confirmation information, call the pre-stored video cache data, filter video segments related to the current change area from the cache time window before the event occurred, filter the cached videos for validity based on the target activity trajectory, and dynamically remove video content unrelated to the event. The event confirmation information specifically includes the scope of the event's impact, the target activity area, and the direction of the event's development.

[0011] Furthermore, the step of associating the surveillance video frames acquired by the camera within a preset period with pre-stored keyframe anchor points further includes: Based on the preset acquisition conditions of the camera in the current scene, scene description features corresponding to the monitoring video frame are generated, wherein the acquisition conditions specifically include spatial layout information, target distribution information and scene composition information; Determine whether the scene description features can match pre-stored keyframe anchor points; If possible, an image region mapping relationship is established based on the feature correspondence between the monitoring video frame and the reference anchor point, the monitoring video frame is divided into several associated regions, and the anchor point source information corresponding to each associated region is recorded respectively.

[0012] This invention also provides an inter-frame differential compression storage system for a low-power camera, comprising: The calculation module is used to associate the monitoring video frames collected by the camera within a preset period with pre-stored key frame anchor points, establish a corresponding reference frame cache, and calculate the inter-frame difference value between the monitoring video frames and the reference frame cache. The judgment module is used to determine whether the inter-frame difference value is greater than a preset change threshold; The execution module is used to perform sparse sampling of still images according to a preset time interval if no, extract frame anchor representative points from the monitoring video frames to form corresponding monitoring index data, dynamically adjust the acquisition cycle of the next key frame based on the changing trend of inter-frame difference values ​​within several consecutive sampling periods, and generate a corresponding set of change event features through several consecutive changing frames. The monitoring index data specifically includes timestamps, device status information, and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction, and target dwell time. The second judgment module is used to determine whether the set of change event features meets the preset abnormal event triggering conditions; The second execution module is used to, if satisfied, collect cached video data within a preset time range before the triggering of the change event, splice the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, compress and manage the complete event video using a keyframe indexing method based on the event data of the complete event video, and dynamically update a preset inter-frame difference judgment model according to the target motion trajectory formed by the complete event video during the abnormal event. The event data specifically includes importance level, storage capacity and network status.

[0013] Furthermore, it also includes: The generation module is used to calculate the stable parameters within the corresponding time window based on the image difference results acquired in several consecutive acquisition cycles, and to generate the corresponding scene stable sequence by combining the continuous stable parameters between adjacent time windows. The third judgment module is used to determine whether the scene stable sequence meets the preset stability maintenance conditions; The third execution module is used to establish a transitional cache set if not, temporarily store the current collected data in the transitional cache set, perform continuous consistency analysis on the transitional cache set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficient based on the local consistency degree, correct the scene stability sequence based on the stability correction coefficient, obtain the corrected stability sequence, and generate the corresponding stable evolution trajectory through the corrected stability sequence.

[0014] Furthermore, the execution module also includes: A submodule is established to establish a corresponding change association chain based on the preset association data of the continuously changing frames. The association data specifically includes the order of occurrence, duration, spatial distribution, and location migration relationship. The judgment submodule is used to determine whether the changed association chain meets the preset continuous association condition; The execution submodule is used to, if not, split the set of change event features, divide the abnormal change information into several independent change units, calculate the persistence influence coefficient corresponding to the independent change unit, rearrange the processing priority of the independent change units according to the persistence influence coefficient, construct the corresponding candidate event queue according to the processing priority, store the candidate event queue in the set to be verified, dynamically receive new image information during subsequent acquisition, perform incremental matching on the candidate event queue, expand and update the candidate event queue through the continuous association relationship formed by the new image information and the candidate event queue, and regenerate the corresponding set of change event features. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

[0015] This invention provides an inter-frame differential compression storage method and system for low-power cameras, which has the following advantages: This invention establishes a reference frame cache by associating monitoring video frames with key frame anchor points and using an inter-frame differential mechanism to identify the degree of image change. In long-term static monitoring scenarios, only representative frame anchor points and index data are retained, avoiding the repeated storage of a large number of similar video frames and effectively reducing local storage space usage. At the same time, by dynamically adjusting the key frame acquisition cycle and generating a set of change event features, the system can adaptively adjust the data acquisition strategy according to scene changes, ensuring the traceability of abnormal events while reducing invalid data transmission. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the inter-frame differential compression and storage method for a low-power camera according to the present invention. Figure 2 This is a structural block diagram of an embodiment of the inter-frame differential compression storage system for a low-power camera according to the present invention. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. The realization of the purpose, functional features, and advantages of the invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings.

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Appendix Figure 1The present invention provides a method for inter-frame differential compression storage of a low-power camera according to an embodiment of the present invention, comprising: S1: Based on the monitoring video frames collected by the camera within a preset period, associate the monitoring video frames with the pre-stored key frame anchor points, establish a corresponding reference frame cache, and calculate the inter-frame difference value between the monitoring video frames and the reference frame cache. S2: Determine whether the inter-frame difference value is greater than a preset change threshold; S3: If not, sparse sampling is performed on the still image according to the preset time interval, frame anchor representative points are extracted from the monitoring video frame to form corresponding monitoring index data, and the acquisition cycle of the next key frame is dynamically adjusted according to the changing trend of the inter-frame difference value within several consecutive sampling periods. Through several consecutive changing frames, a corresponding set of change event features is generated. The monitoring index data specifically includes timestamps, device status information and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction and target dwell time. S4: Determine whether the set of change event features meets the preset abnormal event triggering conditions; S5: If satisfied, collect cached video data within a preset time range before the change event is triggered, and stitch the cached video data with the real-time video stream after the change event is triggered to obtain a complete event video. Based on the event data of the complete event video, compress and manage the complete event video using a keyframe indexing method. According to the target motion trajectory formed by the complete event video during the abnormal event, dynamically update the preset inter-frame difference judgment model. The event data specifically includes importance level, storage capacity, and network status.

[0020] In this embodiment, the system associates the monitored video frames captured by the camera within a preset period with pre-stored key frame anchors to establish a corresponding reference frame buffer. It then calculates the inter-frame difference between the monitored video frame and the reference frame buffer. The system then determines whether this inter-frame difference is greater than a preset change threshold to execute corresponding steps. For example, if the system determines that the inter-frame difference between the monitored video frame and the reference frame buffer is greater than the preset change threshold, the system considers the current image to no longer be in a low-change static state and needs to switch from low-power monitoring mode to change analysis mode. The system will then extract the relevant data from the current monitored video frame and the reference frame buffer. The reference frame buffer contains regions of change. Based on the location, size, continuity, and direction of these changes, the regions are regionalized and labeled, generating corresponding change region information. Simultaneously, a preset number of consecutive video frames are collected, and the change regions within each consecutive frame are correlated to determine if they originate from the same target or process. For example, if the system determines that the inter-frame difference between the monitored video frame and the reference frame buffer is not greater than a preset change threshold, the system considers the current monitored image to have not significantly changed compared to the historical reference image corresponding to the key frame anchor point. The system will then proceed according to the preset... The system performs sparse sampling of still images at set time intervals, extracts frame anchor points (representative keyframe anchor points) from the monitored video frames, and forms corresponding monitoring index data. The monitoring index data specifically includes timestamps, device status information, and environmental parameter information. Based on the changing trend of inter-frame difference values ​​over several consecutive sampling periods, the system dynamically adjusts the acquisition period of the next keyframe. Through several continuously changing frames, a corresponding set of change event features is generated, specifically including the duration of change, the diffusion speed of the change area, the direction of motion, and the target dwell time. The system then uses the timestamps, device status information, and environmental parameter information corresponding to the keyframe anchor points... The system generates monitoring index data, enabling rapid location of corresponding monitoring periods without relying on continuous and complete video streams. It can correlate and manage monitoring data by combining equipment operating status and environmental change information, improving video data retrieval efficiency and management capabilities in static scenes. Furthermore, by dynamically adjusting the next keyframe acquisition cycle based on the changing trend of inter-frame difference values ​​within a continuous sampling period, the camera can adaptively change its data acquisition frequency according to the degree of change in the actual scene, further reducing energy consumption during image acquisition, processing, and storage while ensuring monitoring continuity. The system then determines whether these sets of changing event features meet preset abnormal event triggering conditions to execute corresponding steps.For example, when the system determines that the set of change event features does not meet the preset abnormal event triggering conditions, the system will consider that although there has been a certain degree of image change in the current monitored area, it may belong to ordinary environmental disturbance, normal personnel activity, lighting changes, or other low-risk changes. The system will extract key descriptive information that reflects the current scene change status and associate the key descriptive information with the corresponding timestamp, device operating status, and environmental parameters to generate a low-priority change record. At the same time, based on the location of the current change area, the duration of the change, and the change trend in subsequent acquisition cycles, the system will adaptively adjust the current keyframe anchor point so that the updated keyframe anchor point can more accurately reflect the current monitoring environment status, serving as a reference for subsequent inter-frame difference calculations. For example, when the system determines that the set of change event features meets the preset abnormal event triggering conditions, the system will consider that the current change is more likely to be an abnormal disturbance and needs to immediately enter the abnormal event recording and storage process. The system will collect data within a preset time range before the change event is triggered. The system stitches cached video data with real-time video streams triggered by events to obtain complete event video. Based on this complete event video, which includes event data such as importance level, storage capacity, and network status, the system compresses and manages the complete event video using keyframe indexing. It dynamically updates a pre-defined inter-frame difference judgment model based on the target motion trajectory formed during the abnormal event. By stitching cached video data with real-time video streams triggered by events, and combining event importance level, storage capacity, and network status for keyframe indexing compression management of the complete event video, the system can select an appropriate data storage method based on actual resource conditions. This ensures the complete preservation of key processes during abnormal events while reducing the occupation of invalid video data. Furthermore, using keyframe indexing instead of traditional continuous video storage reduces video data write volume and transmission pressure, decreases local storage space consumption and cloud storage costs, and improves data management efficiency of low-power cameras in long-term monitoring environments.

[0021] It should be noted that sparse sampling of still images is performed according to a preset time interval. Frame anchor points are extracted from the monitored video frames to form corresponding monitoring index data. Based on the changing trend of inter-frame difference values ​​over several consecutive sampling periods, the acquisition period of the next key frame is dynamically adjusted. Through several continuously changing frames, a corresponding set of change event features is generated, specifically: The system first performs sparse sampling of the currently static monitoring footage according to a preset time interval. For example, in normal monitoring mode, the camera may collect 30 frames of video data per second, but after inter-frame difference judgment, it is found that the current area has not changed significantly for a long time. At this time, the system does not need to save all 30 frames, but instead collects one frame as a representative image according to a preset period, such as every 10 seconds, 30 seconds or 1 minute. The collected video frames are not directly used as complete video data storage, but frame anchor representative points that can reflect the current monitoring scene status are extracted from them. The frame anchor representative points can be understood as key reference images in the static phase, such as the background of a room where no one enters, a road scene that remains unchanged for a long time, or a fixed layout image inside a warehouse. Subsequently, the system establishes corresponding monitoring index data based on the extracted frame anchor points. This monitoring index data does not contain complete video content, but rather a set of information describing the monitoring status at that specific time point. Specifically, it may include the timestamp corresponding to the current frame, camera operating status information, and environmental parameter information. For example, in a warehouse monitoring scenario, if the system captures a still image at 1:00 AM, it only saves the time information "01:00:00," the camera's current battery level, network connection status, ambient brightness, temperature, and other parameters, without needing to save the large number of repetitive video frames between 00:59:00 and 01:00:00. When it is necessary to query the monitoring status for that time period later, the system can quickly locate the corresponding time point based on the index information without needing to retrieve the complete video file. Furthermore, the system dynamically adjusts the next keyframe acquisition cycle based on the changing trend of the inter-frame difference value over several consecutive sampling periods. For example, if the inter-frame difference value calculated by the system is close to zero over several consecutive sampling periods, it indicates that the current area has remained stationary for a long time. In this case, the next keyframe acquisition cycle can be appropriately extended, for example, from once every 10 seconds to once every 60 seconds, to reduce camera acquisition, processing, and storage operations. Conversely, if the system detects that the inter-frame difference value gradually increases over several sampling periods, such as a slow change in the position of objects in the warehouse, the keyframe acquisition cycle is shortened, for example, from once every 60 seconds to once every 5 seconds, to improve the ability to record scene changes. When continuous changes occur in the monitored footage, the system will further acquire several consecutive changing frames and analyze these frames to generate a corresponding set of change event features. For example, in a community entrance monitoring scenario, if the system finds that multiple video frames within 5 consecutive seconds show changes in the same area, it will extract information such as the duration of the change, the speed of the change area's spread, the direction of movement, and the target's dwell time. Among these, the duration of the change is used to determine whether the change is short-lived, the speed of the change area's spread is used to determine whether the target is moving or expanding its influence range, the direction of movement is used to describe the target's movement trend, and the target's dwell time is used to determine whether the target remains in the monitored area for a long time. For example, when a camera monitors an unattended office for an extended period, the system uses sparse sampling to save only key frame anchor points at fixed times each day, and gradually extends the sampling period based on environmental stability. When someone enters the office at night, there are significant differences between the continuously acquired video frames and historical key frames. The system automatically shortens the acquisition period, continuously acquires the changing video frames, and extracts a set of event features such as "person's entry time, direction of movement, and area of ​​stay," providing a data foundation for subsequent abnormal event judgment and complete event video storage. Through the above methods, the system can reduce the frequency and amount of video data acquisition in static scenes, while automatically increasing the data acquisition capability when effective changes occur, achieving a dynamic monitoring effect of "low power consumption when static and high completeness when changing".

[0022] It should be added that, cached video data within a preset time range before the triggering of the change event is collected, and the cached video data is spliced ​​with the real-time video stream after the triggering of the change event to obtain a complete event video. Based on the event data of the complete event video, the complete event video is compressed and managed using a keyframe indexing method. According to the target motion trajectory formed by the complete event video during the abnormal event, the preset inter-frame difference judgment model is dynamically updated, specifically as follows: The system calls the video data that the camera has pre-cached before the abnormal event is triggered. The cached video data is used to save the original state of the monitored area for a period of time before the event occurs. For example, when the monitored area detects that a person has entered the restricted area, the system not only obtains the real-time video stream after the person enters, but also simultaneously retrieves the cached video within 30 seconds before the person enters, so as to retain the complete process of the person approaching, entering and the abnormal behavior before and after the event, and avoid the loss of key images due to the delay in the event triggering. Subsequently, the system correlates the cached video data with the real-time video stream triggered by the change event. Based on the time information, frame order, and continuity of the two video segments, it stitches them together to obtain a complete event video. Specifically, the system uses the trigger time of the abnormal event as the connection node, taking the cached video segment before the trigger as the pre-event stage and the real-time video stream after the trigger as the event development stage. It also corrects the time interval between the two video segments to ensure that the final generated video continuously reflects the occurrence of the abnormal event. For example, in a warehouse monitoring scenario, when the system detects personnel entering the warehouse, it automatically extracts the environmental footage before entry, the personnel's entry process, and the activity footage after entry, forming a complete event recording for easy subsequent tracing and analysis. Furthermore, the system performs keyframe index compression management on the complete event video based on the event data corresponding to the complete event video. The event data includes information such as event importance level, current storage capacity, and network status. The system determines the video saving priority based on the event importance level, the video compression ratio based on the storage capacity, and the local storage or cloud transmission method based on the network status. During video compression management, the system extracts key nodes from the complete event video that represent the event's development process, such as the target's first appearance location, the location where the target's movement direction changes, the location where the target stays for a relatively long time, and the event's end location, and establishes keyframe index information. For a large number of repeated frames in a continuous process, the system reduces redundant storage by using index references. For example, in a 5-minute abnormal event video, if the target only exhibits a few key action changes, the system does not need to save all repeated frames, but only saves the keyframes and their corresponding index relationships, thus reducing storage space usage while ensuring the integrity of the event process. Finally, the system dynamically updates the preset inter-frame differential judgment model based on the target motion trajectory formed during the abnormal event in the complete event video. Specifically, the system analyzes the target's movement range, direction of movement, dwell area, and image change patterns during the abnormal event, and feeds the analysis results back to the inter-frame differential judgment model, enabling the model to adjust subsequent judgment parameters according to the target change characteristics in the actual monitoring environment. For example, in parking area monitoring, the system discovers through multiple abnormal event records that vehicles typically exhibit slow entry, dwelling, and departure trajectories. In subsequent detection processes, the differential judgment method can be optimized based on this motion pattern to improve the ability to identify abnormal vehicle dwellings or movements. In personnel monitoring areas, the system can also adjust change detection strategies based on the range of personnel activity and movement characteristics to reduce false triggering.

[0023] In this embodiment, before step S3, which involves sparsely sampling still images according to a preset time interval, extracting frame anchor representative points from the monitored video frames, and forming corresponding monitoring index data, the method further includes: S301: Based on the image difference results acquired in several consecutive acquisition cycles, calculate the stable parameters within the corresponding time window, and combine the continuous stable parameters between adjacent time windows to generate the corresponding scene stable sequence. S302: Determine whether the stable sequence of the scenario meets the preset stability maintenance condition; S303: If not, then establish a transitional cache set, temporarily store the current collected data in the transitional cache set, perform continuous consistency analysis on the transitional cache set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficient based on the local consistency degree, correct the scene stability sequence based on the stability correction coefficient, obtain the corrected stability sequence, and generate the corresponding stable evolution trajectory through the corrected stability sequence.

[0024] In this embodiment, the system calculates stable parameters within the corresponding time window based on the image difference results acquired over several consecutive acquisition cycles (stable parameters are quantitative indicators used to characterize the stable state of the current monitoring scene, calculated based on the degree of difference between continuously acquired images within a single time window). Combining this with continuous stable parameters between adjacent time windows (continuous stable parameters are continuous stable state description information formed by combining the changing relationships of stable parameters within multiple adjacent time windows), the system generates a corresponding scene stability sequence (a scene stability sequence is a data sequence reflecting the trend of the monitoring scene changing over time, formed by arranging continuous stable parameters corresponding to multiple consecutive time windows in chronological order). The system then determines whether these scene stability sequences meet pre-set stability maintenance conditions and executes corresponding steps accordingly. For example, when the system determines that these scene stability sequences meet the pre-set stability maintenance conditions, the system considers that the current monitoring area has not experienced significant image changes over several consecutive acquisition cycles, and the scene is in a continuously stable state. Based on the stability level corresponding to the scene stability sequence, the system reduces the processing frequency of the current monitoring video frames, extracts only representative image frames as stable state reference frames according to a preset sampling period, and associates these reference frames with corresponding time information, device operating status, and environmental parameters to generate a static scene index. Data is introduced; for example, when the system determines that these scene stable sequences cannot meet the pre-set stability maintenance conditions, the system will assume that there are no consecutive similar images in the current monitoring video, and a high proportion of duplicate data will not be generated. Therefore, there is no need to enter the static scene low-redundancy processing mode. The system will establish a transitional buffer set, temporarily store the current collected data in the transitional buffer set, perform continuous consistency analysis on the transitional buffer set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficients based on different local consistency degrees, and correct the scene stable sequences based on these stability correction coefficients to obtain the corrected stable sequences. Through the corrected stable sequences, data is generated... The system generates a corresponding stable evolution trajectory. By performing consistency analysis on continuously collected data in the transition buffer set, the system calculates the degree of local consistency between different collection times and obtains the corresponding stability correction coefficient based on different degrees of local consistency. This enables dynamic correction for monitoring scenarios with different degrees of change, avoiding the problem of abnormal decline in stable sequences caused by non-critical factors such as changes in lighting, slight occlusion, and equipment vibration. At the same time, by generating a stable evolution trajectory based on the corrected stable sequence, the system can record the continuous evolution process of the monitoring scenario from a stable state to a changing state, providing a more complete data basis for subsequent judgment on whether to enter the change detection, event analysis, or data augmentation collection stage.

[0025] It should be noted that a transitional buffer set is established to temporarily store the current collected data. A continuous consistency analysis is performed on the transitional buffer set to calculate the local consistency degree between different collection times. Based on the local consistency degree, a corresponding stability correction coefficient is obtained. The scene's stable sequence is then corrected using this stability correction coefficient to obtain a corrected stable sequence. Finally, a corresponding stable evolution trajectory is generated using this corrected stable sequence. Specifically: The system stores video frames within a continuous acquisition period into a transitional buffer set in chronological order. The transitional buffer set is used to store intermediate data during the transition from a stable state to a changing state. For example, in a community road monitoring scenario, the system initially detects that no one has passed through the road area for a long time and that it is in a stable state. However, if a slight change is detected in the image within a certain period of time, the system will not immediately identify it as an abnormal event. Instead, it will temporarily store the video frames within a certain period of time before and after the change into the transitional buffer set and determine whether the change persists through subsequent analysis. Subsequently, the system performs consistency analysis on the continuous video frames in the transition buffer set, compares the local image preservation between video frames corresponding to different acquisition times, and calculates the local consistency degree between each acquisition time; the local consistency degree is used to describe the proportion of different areas in the monitoring screen that maintain a similar state. For example, it can analyze whether fixed background areas, equipment areas, road areas and other locations maintain the same state in continuous time. For example, in a warehouse monitoring scenario, most areas inside the warehouse remain unchanged across multiple consecutive data collection times, with only the area near the entrance showing slight differences due to changes in lighting. In this case, the system can calculate that most areas have a high degree of local consistency, while the consistency of the changed areas is low. In this way, it is possible to avoid mistaking local environmental changes for significant changes in the entire monitoring scenario. Furthermore, the system obtains a corresponding stability correction coefficient based on the degree of local consistency obtained between different acquisition times. The stability correction coefficient is used to adjust the original scene stability sequence so that it can more accurately reflect the true stability state of the current monitoring environment. For example, when the degree of local consistency is high within multiple consecutive time windows, it indicates that the current changes are mainly concentrated in a few areas, and the system can increase the stability correction coefficient to keep the scene stability at a high level. When inconsistencies occur continuously in multiple areas, it indicates that the scene is undergoing continuous changes, and the system decreases the stability correction coefficient so that the stability sequence can more sensitively reflect the trend of change. Subsequently, the system corrects the original scene stable sequence based on the stability correction coefficient to obtain the corrected stable sequence. Compared with the stable sequence generated directly based on the difference results of a single image, the corrected stable sequence takes into account the local changes within a continuous time range, which can reduce the impact of transient interference factors on the judgment of the stable state. For example, when the brightness of the image changes due to the brief illumination of vehicle lights in a nighttime monitoring scene, the original stable sequence may show a significant decrease. However, after correction by local consistency analysis, the system can identify that the change only affects a local area and lasts for a short period of time, thus maintaining a higher stable state. Finally, the system generates a corresponding stable evolution trajectory based on the corrected stable sequence to describe the development process of the monitored scene over time. The stable evolution trajectory can reflect the continuous trend of the scene from a stable state, a transitional state to a changing state. For example, in office area monitoring, the system can use the stable evolution trajectory to find that frequent human activity during the day leads to a decrease in stability, while the stability gradually recovers at night when no one is around. This provides a basis for subsequent dynamic adjustment of key frame acquisition cycle, video storage strategy, and abnormal event judgment.

[0026] In this embodiment, step S3, which involves dynamically adjusting the acquisition period of the next key frame based on the changing trend of the inter-frame difference values ​​over several consecutive sampling periods, and generating a corresponding set of change event features through several consecutive changing frames, further includes: S31: Based on the preset associated data of the continuously changing frames, establish a corresponding change association chain, wherein the associated data specifically includes the order of occurrence, duration, spatial distribution and location migration relationship; S32: Determine whether the changed association chain satisfies the preset continuous association condition; S33: If not, the set of change event features is split into several independent change units, the abnormal change information is divided into several independent change units, the persistence influence coefficient of the independent change units is calculated, the processing priority of the independent change units is rearranged according to the persistence influence coefficient, the corresponding candidate event queue is constructed according to the processing priority, the candidate event queue is stored in the set to be verified, new image information is dynamically received during subsequent acquisition, incremental matching is performed on the candidate event queue, the candidate event queue is expanded and updated through the continuous association relationship formed by the new image information and the candidate event queue, and the corresponding set of change event features is regenerated. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

[0027] In this embodiment, the system establishes a corresponding change association chain based on pre-set association data of continuously changing frames. This association data specifically includes the order of occurrence, duration, spatial distribution, and location migration relationships. The system then determines whether the change association chain meets pre-set continuous association conditions to execute corresponding steps. For example, when the system determines that the change association chain meets the pre-set continuous association conditions, the system considers that the current change may be continuously generated by the same target movement, the same behavioral process, or the same environmental state change. The system then performs continuous association based on the order of occurrence, duration, spatial distribution, and location migration relationships in the change association chain. The system integrates and analyzes the changing frames to determine the relationships between the changing nodes, extracts the corresponding change development information, and generates corresponding change trajectory description data through the change association chain. It records the target's movement direction, activity range, and location during the change process. For example, if the system determines that the change association chain cannot meet the pre-set continuous association conditions, it will consider the change in the current monitoring screen to be caused by a single isolated frame change or transient noise. The system will then split the feature set of the change event, dividing the abnormal change information into several independent change units. The abnormal change information specifically includes excessively long time intervals and spatial spans. For excessively large changes and inconsistent evolution directions, the system calculates the persistent impact coefficients of these independent change units. Based on these coefficients, the processing priorities of these units are rearranged, and candidate event queues are constructed according to these priorities. These queues are then stored in a set to be verified. During subsequent data acquisition, new image information is dynamically received, and incremental matching is performed on these queues. By establishing continuous associations between new image information and the queues, the system expands and updates these queues, regenerating the corresponding change event feature set. By calculating the persistent impact coefficients of each independent change unit and rearranging processing priorities based on these coefficients, the system can allocate resources according to the potential impact of each change unit on the subsequent monitoring status. Changes with the potential for continued development are prioritized for retention and analysis, while low-impact, short-lived changes are downgraded. Furthermore, by dynamically receiving new image information and incrementally matching it with the queues, the system can continuously track weak change processes that do not initially meet the conditions for continuous association. When subsequent change information establishes a continuous association with existing queues, the system automatically expands and updates the queues, regenerating a complete change event feature set.

[0028] It should be noted that the set of change event features is split into several independent change units, the abnormal change information is divided into several independent change units, the persistence influence coefficient of each independent change unit is calculated, the processing priority of each independent change unit is rearranged according to the persistence influence coefficient, a corresponding candidate event queue is constructed according to the processing priority, the candidate event queue is stored in the set to be verified, and new image information is dynamically received during subsequent acquisition. Incremental matching is performed on the candidate event queue, and the candidate event queue is expanded and updated through the continuous association relationship formed between the new image information and the candidate event queue, and the corresponding set of change event features is regenerated. Specifically: The system categorizes and organizes each change feature based on the time information, spatial location information, and change development information recorded in the change event feature set. Changes with excessively long time intervals, large spatial spans, or inconsistent evolution directions are separated. For example, in a park monitoring scenario, the system detects personnel movement in the entrance area at 10:00:05, vehicle changes in the parking area at 10:00:30, and light changes at the road edge at 10:01:00. Since these three changes occur at different times, in different spatial locations, and have no correlation in their trends, the system will not treat them as the same abnormal event. Instead, it will classify them into personnel change unit, vehicle change unit, and environmental change unit, respectively. After the feature set of the change event is decomposed, the system further calculates the persistence influence coefficient corresponding to each independent change unit. The persistence influence coefficient is used to represent the degree to which a certain change unit may affect the subsequent monitoring status. It can be determined comprehensively based on factors such as the duration of the change, the range of the affected area, the number of times the change occurs repeatedly, and the trend of the change's spread. For example, a brightness change that lasts only two frames and disappears quickly has a low impact on the monitoring results due to its short duration and small range of influence, and its persistence influence coefficient is small. On the other hand, a change unit in which a person enters the monitoring area and moves and stays there continuously has a high persistence influence coefficient because of its longer duration and gradually expanding range of influence. Subsequently, the system rearranges the processing priorities of each change unit based on the persistence impact coefficient corresponding to different independent change units. Specifically, for change units with high persistence impact coefficients, the system increases their processing priority, prioritizes the allocation of image analysis resources, and continuously acquires subsequent change information; for change units with low persistence impact coefficients, the processing priority is reduced, and only basic change description information is retained. For example, in a warehouse monitoring scenario, local changes caused by flashing equipment indicator lights may not affect the overall monitoring status, so the processing priority is low; while change units formed by the continuous movement of personnel inside the warehouse have higher priority and need to be continuously tracked. Furthermore, the system constructs a corresponding candidate event queue according to the rearranged processing priority, and stores the candidate event queue in the set to be verified; the candidate event queue is used to save change process information that has not yet formed a complete continuous event, but has further observation value; for example, when the camera first detects a person briefly appearing at the edge of the monitored area, since the current duration is insufficient and the movement trajectory is incomplete, the system will not immediately trigger abnormal event recording, but will establish the change information as a candidate event queue and wait for subsequent data collection for verification; During subsequent monitoring, the system continuously and dynamically receives new image information and performs incremental matching between the new image information and the candidate event queue in the set to be verified. Specifically, the system analyzes the location, time of appearance, and direction of change in the new image and determines whether it is related to the change unit in the existing candidate event queue. For example, if a person only stays at the edge of the screen when they first appear, the system establishes a corresponding candidate event queue. If the person continues to move into the monitored area in a subsequent new image, the system matches the new change information with the existing candidate event queue based on the positional migration relationship in the new image. Once a continuous association is established between newly added image information and the candidate event queue, the system expands and updates the corresponding candidate event queue, adding the newly added time information, spatial location information, and change trend information to the original queue, so that the originally scattered change units gradually form a complete change process; for example, the candidate event queue that originally only recorded "person appearance" will be further expanded to form a complete change record containing "entry time, movement path, and dwelling location" after continuously receiving subsequent movement, dwelling and other image information; Finally, the system regenerates the set of change event features based on the expanded and updated candidate event queue, and uses the updated change duration, change area range, movement direction, and target dwell state as the basis for subsequent abnormal event judgment. In this way, the system can continuously observe changes that are not obvious in the initial stage and have insufficient correlation, so that some progressive abnormal behaviors can be re-identified in the subsequent development process without directly increasing the acquisition frequency or starting the full recording process in the initial stage.

[0029] In this embodiment, step S5, which involves collecting cached video data within a preset time range before the triggering of a change event and splicing the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, further includes: S51: Based on the end node of the cached video and the start node of the real-time video, identify the association features between the end node and the start node, and generate video connection parameters according to the association features; S52: Determine whether the video connection parameters meet the preset continuous association conditions; S53: If not, then reverse the search for the end position of the cached video and the start position of the real-time video, search again for the associated video segment with the start frame of the real-time video in the cache area, divide the corresponding splicing boundary according to the search results, perform transition correction on the video data between the splicing boundaries, dynamically adjust the consistency of motion trajectory between adjacent video frames, and recombine the corrected video data with the real-time video stream to generate continuous event video.

[0030] In this embodiment, the system identifies the correlation features between the end node of the cached video and the start node of the real-time video based on the end node and start node. Based on these correlation features, video splicing parameters are generated. The system then determines whether these video splicing parameters meet pre-set continuous correlation conditions to execute corresponding steps. For example, if the system determines that these video splicing parameters meet the pre-set continuous correlation conditions, it considers the cached video data and the real-time video stream to belong to the same event development process and can directly perform video fusion processing to form a complete and continuous event record. The system determines the optimal connection position between the end node of the cached video and the start node of the real-time video based on the video splicing parameters, arranges and merges the two parts of video data in chronological order, and adjusts for possible time deviations, inter-frame intervals, and differences in image changes between the cached video and the real-time video using the video splicing parameters to ensure continuous playback of the spliced ​​video data. Conversely, if the system determines that these video splicing parameters do not meet the pre-set continuous correlation conditions, the system considers the two video data segments to be... Unable to achieve natural transitions in time sequence and content changes, exhibiting obvious video gaps, abrupt scene changes, or sudden changes in target state, the system reversely searches the end position of the cached video and the start position of the real-time video. It then re-searches the cached area for the associated video segment (the segment with the highest degree of correlation) with the starting frame of the real-time video. Based on different search results, corresponding splicing boundaries are defined, and the video data between these splicing boundaries undergoes transitional correction. This dynamically adjusts the consistency of motion trajectories between adjacent video frames, and the corrected video data is then recombined with the real-time video stream to generate continuous event video. By defining corresponding splicing boundaries based on different search results and performing transitional correction on the video data between these boundaries, the system can dynamically repair issues such as inconsistent time intervals, significant changes in scene state, and discontinuous target motion. Furthermore, by recombining the corrected video data with the real-time video stream to generate continuous event video, it can still restore a relatively complete event process even when the original video segments have splicing defects, improving the usability and traceability of abnormal event recordings.

[0031] It should be noted that the process involves reverse retrieval of the end position of the cached video and the start position of the real-time video, retrieving the associated video segment from the cache area that corresponds to the start frame of the real-time video, defining the corresponding splicing boundaries based on the retrieval results, performing transition correction on the video data between the splicing boundaries, dynamically adjusting the consistency of motion trajectories between adjacent video frames, and then recombining the corrected video data with the real-time video stream to generate a continuous event video. Specifically: The system first uses the start frame information of the real-time video to reverse-search the end position of the cached video and the video data within the corresponding time range to find the video segment with the highest correlation to the start frame of the real-time video. The correlation can be comprehensively judged based on information such as target position, screen structure, background features, target motion state, and image change trend. For example, in a scenario where a person abnormally enters the monitored area, the last frame of the cached video shows the person outside the entrance, while the first frame of the real-time video shows the person has entered the room, and there is a clear positional difference between the two. In this case, the system will not directly connect the two frames, but will search backward in the cache area to find a video segment where the person is gradually approaching the entrance and the position is closer to the start frame of the real-time video, as a new connection reference. Furthermore, the system determines the corresponding splicing boundary based on the matching results between the video segments obtained from the reverse retrieval and the starting frame of the real-time video. The splicing boundary is used to divide the video areas that need to be retained and the video areas that need to be adjusted in the cached video, so that the system can avoid repeating the processing of the entire cached video. For example, when the system finds that the frame between the 20th and 25th seconds of the cached video has a high correlation with the starting state of the real-time video, the video nodes within this time range are used as splicing reference boundaries, and the relevant video data before and after them are processed as the pre-event stage and the event development stage, respectively. Subsequently, the system performs transition correction on the video data between the splicing boundaries to eliminate discontinuities between two video segments caused by differences in acquisition time, changes in target movement speed, and changes in image state. Specifically, the system analyzes the target position change relationship between consecutive video frames within the splicing area and adjusts the inter-frame relationship with obvious jumps to maintain a smoother transition process between adjacent video frames. For example, in a vehicle monitoring scenario, if a vehicle has just entered the right side of the frame in the cached video, while the vehicle has already appeared in the center of the road when the real-time video starts, a direct connection would result in the vehicle moving instantaneously. The system adjusts the position change relationship between intermediate transition frames to make the vehicle's movement trajectory conform to the actual driving process. Furthermore, the system dynamically adjusts the consistency of motion trajectories between adjacent video frames based on the target's motion state during the correction process. Specifically, the system extracts the target's direction of movement, speed of movement, and range of position changes in consecutive video frames, and determines whether adjacent frames conform to normal motion patterns. When abnormal displacement is detected between certain video frames, the system re-associates the motion trajectory of that part based on the target's state relationship in the preceding and following video frames, ensuring the continuity of the target's movement in the video. For example, in a personnel monitoring scenario, if a person experiences a significant positional change between two adjacent video frames, the system will combine the person's direction and speed of movement in the preceding and following frames to adjust the consistency of the intermediate change process, ensuring that the final video continuously displays the person's walking process. Finally, the system recombines the corrected video data after overcorrection and motion trajectory adjustment with the real-time video stream triggered by the change event to generate a continuous event video. The generated continuous event video not only contains the real-time process after the abnormal event occurs, but also retains information on the appearance, movement and state changes of the target before the event occurs, avoiding the loss of key processes due to abnormal video splicing.

[0032] In this embodiment, step S2, which determines whether the inter-frame difference value is greater than a preset change threshold, further includes: S21: Generate inter-frame difference description parameters based on the difference data between the current video frame and the reference video frame; S22: Determine whether the difference description parameter exceeds a preset change limit; S23: If not, collect low-amplitude change information from the inter-frame difference description parameters, extract persistent background change features from the low-amplitude change information, match the background change features with the environmental change features saved in the historical monitoring period, dynamically classify the environmental change attribution of the monitoring video frame according to the matching result, generate corresponding environmental adaptation parameters, and adjust the update period of the reference video frame according to the environmental adaptation parameters. The environmental change attribution specifically includes periodic environmental changes and known environmental changes.

[0033] In this embodiment, the system generates inter-frame difference description parameters based on the difference data between the current video frame and the reference video frame. The system then determines whether these difference description parameters exceed a preset change limit to execute corresponding steps. For example, if the system determines that the inter-frame difference description parameters have exceeded the preset change limit, the system considers the current change not to be caused by minor noise, device jitter, or normal image fluctuations, and therefore has some effective information value. The system will determine the change area in the current video frame based on the difference description parameters, extract the location, range, and degree of change of the change area, and continuously track and analyze the current change area. Instead of immediately triggering event processing based on the single-frame change result, the system continues to collect several subsequent video frames to determine whether the change persists and whether it has a continuous change pattern. Conversely, if the system determines that the inter-frame difference description parameters have not exceeded the preset change limit, the system considers the current video frame to have not undergone significant content changes compared to the reference video frame, and the current monitoring image remains in its original stable state. The system will collect these inter-frame differences... The system extracts persistent background change features from low-amplitude change information in the description parameters, matches different background change features with environmental change features saved in historical monitoring periods, and dynamically classifies the environmental change attribution of monitoring video frames based on the matching results. The environmental change attribution specifically includes periodic environmental changes and known environmental changes, generating corresponding environmental adaptation parameters. Based on the environmental adaptation parameters, the update cycle of reference video frames is adjusted. By matching different background change features with environmental change features saved in historical monitoring periods, the system can analyze the current scene status in conjunction with historical environmental change patterns. For example, it can identify changes in day and night illumination, weather changes, changes in brightness in fixed areas, and periodic environmental disturbances. Based on the matching results, the system dynamically classifies the environmental change attribution of monitoring video frames. At the same time, by adjusting the update cycle of reference video frames based on environmental adaptation parameters, the system can dynamically maintain reference video frames according to the current environmental stability. When environmental changes are small and remain stable for a long time, the update cycle of reference video frames is extended to reduce invalid update operations and data processing overhead.

[0034] It should be noted that the process involves collecting low-amplitude change information from the inter-frame difference description parameters, extracting persistent background change features from this low-amplitude change information, matching these background change features with environmental change features stored within historical monitoring periods, dynamically classifying the environmental change attribution of the monitoring video frames based on the matching results, generating corresponding environmental adaptation parameters, and adjusting the update period of the reference video frames according to these environmental adaptation parameters. Specifically: The system first extracts low-amplitude change information from the inter-frame difference description parameters. This low-amplitude change information is used to indicate subtle differences between the current video frame and the reference video frame that do not reach a significant change threshold, such as changes in screen brightness, changes in background texture, and changes in the surface state of fixed objects. Based on the low-amplitude change information in multiple consecutive acquisition cycles, the system analyzes whether different change areas persist and extracts background change features with long-term stability from them. For example, in an indoor office area monitoring scenario, the brightness of the area near the window will gradually change during the day as the position of the sun changes, but the main structure of the office, such as desks, chairs, and walls, will not change. In this case, the system will not judge the brightness change as an abnormal event, but will extract the continuous background change features such as the brightness change of the window area and the range of shadow movement from the continuous video frames and record them as current environmental change information. Furthermore, the system matches the extracted background change features with the environmental change features saved within the historical monitoring period to determine whether the current low-amplitude change belongs to a normal change in the historical environmental change pattern. The environmental change features saved within the historical monitoring period may include information such as brightness change trends in different time periods, fixed area change patterns, and periodic background change states. For example, if the system has recorded that the window area experiences gradually increasing light changes from 8:00 AM to 10:00 AM every day in the past several periods, then when a similar change is detected now, it can be determined that the change belongs to a normal environmental change, rather than an abnormal change in the monitored area. Based on the matching results, the system further dynamically classifies the environmental changes corresponding to the current monitoring video frame. Specifically, the system can classify the current change as a periodic environmental change, random environmental disturbance, or unknown change state based on the similarity between the current background change characteristics and historical environmental change characteristics. For example, when the current change highly matches the historical lighting change pattern, the system classifies it as a normal environmental change; when the change characteristics differ significantly from historical data but the impact range is small, the system can classify it as a temporary environmental disturbance; when the change area continues to expand and cannot match historical environmental characteristics, it is retained as an unknown change state to be further analyzed. Subsequently, the system generates corresponding environmental adaptation parameters based on the environmental change attribution results. These environmental adaptation parameters are used to describe the degree of influence of the current monitoring environment on the reference video frame maintenance strategy. For example, for environmental changes that are stable over a long period and have obvious periodic patterns, the system generates higher environmental adaptation parameters, indicating that the current environmental changes are predictable and that frequent updates to the reference state are not required. For environments that change rapidly or have unclear patterns, lower environmental adaptation parameters are generated to ensure that the system maintains a higher reference update frequency. Finally, the system adjusts the update cycle of the reference video frame based on the environmental adaptation parameters, enabling the reference video frame to adapt to changes in the current monitoring environment. For example, in a warehouse unattended at night, if the system detects minimal background changes over several consecutive hours, it extends the update cycle of the reference video frame, updating the reference status only periodically to reduce video processing and storage operations. In contrast, in the shopping mall entrance area, where pedestrian traffic and ambient lighting changes frequently, the system shortens the update cycle of the reference video frame based on the environmental adaptation parameters, allowing the reference image to reflect the current environmental status in a timely manner and improving the accuracy of subsequent inter-frame difference judgments.

[0035] In this embodiment, step S4, which determines whether the set of change event features meets the preset abnormal event triggering conditions, further includes: S41: Based on the event behavior description parameters corresponding to the set of changed event features, match the event behavior description parameters with the historical behavior data corresponding to the current monitoring area to obtain the degree of difference between the current changed behavior and the normal behavior pattern, and calculate the event credibility evaluation value based on the degree of difference. S42: Determine whether the event credibility evaluation value reaches the preset credibility level; S43: If so, generate event confirmation information corresponding to the event credibility evaluation value, call the pre-stored video cache data according to the event confirmation information, filter video segments related to the current change area from the cache time window before the event occurred, filter the cached video for validity according to the target activity trajectory, and dynamically remove video content unrelated to the event. The event confirmation information specifically includes the scope of event impact, target activity area and event development direction.

[0036] In this embodiment, the system matches the event behavior description parameters corresponding to the set of changed event features with the historical behavior data corresponding to the current monitoring area to obtain the degree of difference between the current changed behavior and the normal behavior pattern. Based on different degrees of difference, the system calculates the event credibility evaluation value. Then, the system determines whether the event credibility evaluation value reaches the preset credibility level to execute the corresponding steps. For example, when the system determines that the event credibility evaluation value has not reached the preset credibility level, the system will consider that situations such as people staying briefly in the monitoring area, objects being temporarily moved and then returning to their original position, and changes in the target outline caused by changes in lighting may all form certain change characteristics, but their behavior patterns are relatively close to historical normal behavior. Therefore, they are not identified as abnormal events. After determining that the event credibility evaluation value has not reached the credibility level, the system will not immediately trigger the complete event recording and storage process. Instead, it will reduce the processing level of the current event, mark the changed behavior as an event to be observed, retain the corresponding event behavior description parameters, and simultaneously perform delayed tracking of the video change process corresponding to the event to be observed. In subsequent acquisition cycles, it will continue to acquire new monitoring video frames and analyze the lines in the new video frames. To determine whether the change deviates further from the historical normal behavior pattern; for example, when the system determines that the credibility evaluation value of the event has reached a pre-set credibility level, the system will consider the current change event to have caused a certain degree of image change and belong to a real abnormal event. The system will generate event confirmation information corresponding to the event credibility evaluation value. The event confirmation information specifically includes the scope of the event's impact, the target activity area, and the event's development direction. Based on this event confirmation information, the system calls the pre-stored video cache data, filters video segments related to the current change area from the cache time window before the event occurred, and performs validity filtering of the cached video according to the target activity trajectory, dynamically removing video content irrelevant to the event. By calling the pre-stored video cache data according to the event confirmation information and filtering video segments related to the current change area from the cache time window before the event occurred, the system can effectively retain the key process before the abnormal event occurred, avoiding the loss of important pre-event images due to delays in event triggering. At the same time, by dynamically removing video content irrelevant to the event according to the target activity trajectory, the system can reduce the size of video data while ensuring the complete recording of abnormal events, reducing subsequent storage space occupation and network transmission pressure.

[0037] It should be noted that the generation of event confirmation information corresponding to the event credibility evaluation value, based on the event confirmation information, involves calling pre-stored video cache data, filtering video segments related to the currently changing area from the cache time window before the event occurred, and performing validity filtering on the cached videos based on the target activity trajectory, dynamically removing video content irrelevant to the event. Specifically: The system generates event confirmation information based on the event behavior description parameters corresponding to the current change event. This event confirmation information describes the specific impact of the abnormal event, including the event's impact range, target activity area, and event development direction. Specifically, the event impact range indicates the size of the spatial area involved in the current change, the target activity area determines the main location of the abnormal target, and the event development direction describes the target's movement trend within the monitored area. For example, in a warehouse monitoring scenario, after the system detects personnel entering a restricted area, it confirms through event credibility evaluation that this behavior is a high-credibility abnormal event. At this point, event confirmation information is generated, recording the area the personnel entered, their activity range, and the direction of movement from the entrance to the equipment area, providing a basis for subsequent video screening. Subsequently, the system retrieves pre-stored video cache data based on the event confirmation information, and filters video clips related to the currently changing area from the cache time window before the event occurred. Since low-power cameras typically do not continuously save complete high-definition video, but instead use caching to temporarily store recent video data, the system can utilize a cache time window before the event is triggered to obtain the key process before the abnormal event occurs. For example, when the system detects personnel approaching an important equipment area at 10:05, it will retrieve cached video from 10:00 to 10:05, and based on the target activity area in the event confirmation information, it will only filter video clips related to personnel entering the passage and the area around the equipment, without processing the video content of the entire monitored area. Furthermore, the system evaluates the validity of the selected cached videos based on the target's activity trajectory, analyzing the target's position changes, movement paths, and dwell status within the cached videos to determine whether each video segment is related to the current abnormal event. For example, in a parking lot monitoring scenario, after the system detects an abnormal vehicle dwelling, it analyzes the video data prior to the event based on the vehicle's movement trajectory. If a cached video segment shows a vehicle entering from the entrance and gradually approaching the abnormal dwelling area, that video segment is retained. If another video segment only records vehicles passing by on a distant road and is not related to the current vehicle's trajectory, it is determined to be irrelevant video content. After completing the validity screening, the system dynamically removes video content unrelated to the event, retaining only video clips that reflect the complete process before and after the event. Specifically, the system can retain or delete different clips in the cached video based on the continuity of the target activity trajectory, regional correlation, and temporal proximity, so that the final retained video data revolves around the abnormal event. For example, in a community monitoring scenario, when the system confirms that a person enters a restricted area at an abnormal time, it will retain the video of the person approaching, entering, and staying in the area, while deleting video content of other unrelated persons' activities in other areas during the same period.

[0038] In this embodiment, step S1, which associates the surveillance video frames acquired by the camera within a preset period with pre-stored keyframe anchor points, further includes: S11: Based on the preset acquisition conditions of the camera in the current scene, generate scene description features corresponding to the monitoring video frame, wherein the acquisition conditions specifically include spatial layout information, target distribution information and scene composition information; S12: Determine whether the scene description features can match the pre-stored keyframe anchor points; S13: If possible, then based on the feature correspondence between the monitoring video frame and the reference anchor point, establish an image region mapping relationship, divide the monitoring video frame into several associated regions, and record the anchor point source information corresponding to each associated region.

[0039] In this embodiment, the system generates scene description features corresponding to the monitoring video frame based on the pre-set acquisition conditions of the low-power camera in the current frame. The acquisition conditions specifically include spatial layout information, target distribution information, and scene composition information. Then, the system determines whether these video description features can match pre-stored keyframe anchors to execute corresponding steps. For example, when the system determines that the scene description features corresponding to the monitoring video frame cannot match the pre-stored keyframe anchors, the system will consider that the current scene description features cannot establish a valid correspondence with the existing keyframe anchors, and the original reference state can no longer accurately describe the current monitoring environment. The system will mark the current monitoring video frame as a scene state to be updated, further analyze the unmatched scene description features to determine the reason for the matching failure, and at the same time, re-establish a temporary reference state based on the currently unmatched scene description features, and extract new scene feature information from the currently acquired video frame to form candidate keyframe anchors. For example, when the system determines that the scene description features corresponding to a monitored video frame can match pre-stored keyframe anchor points, the system assumes that there is no significant difference between the currently acquired video frame and the historically saved key reference states. Based on the feature correspondence between the monitored video frame and the reference anchor points, the system establishes an image region mapping relationship, dividing the monitored video frame into several associated regions and recording the anchor point source information corresponding to each associated region. By establishing an image region mapping relationship based on the feature correspondence between the monitored video frame and the reference anchor points, the system can associate different regions in the current monitored image with corresponding regions in the historical reference states, thereby achieving regionalized management of the monitored scene. Simultaneously, by dividing the monitored video frame into several associated regions and recording the anchor point source information corresponding to each associated region, the system can retain reference data for different regions, providing a more accurate data foundation for subsequent inter-frame difference calculation, change area location, and event analysis.

[0040] It should be noted that, based on the feature correspondence between the monitored video frames and reference anchor points, an image region mapping relationship is established, dividing the monitored video frames into several associated regions, and recording the anchor point source information corresponding to each associated region, specifically: The system first extracts scene feature information from the surveillance video frames and matches this feature information with the corresponding features in the keyframe anchor points to determine the positional correspondence between each area in the current video frame and the historical reference image. The feature correspondence can be established based on information such as spatial location, regional structure, target distribution, and background composition. For example, in a warehouse monitoring scenario, the historical keyframe anchor points record the fixed layout inside the warehouse, including the shelving area, equipment area, aisle area, and entrance / exit area. When the system acquires a new surveillance video frame, it will associate the corresponding area in the current image with the same area in the historical keyframe based on stable features such as wall boundaries, shelving outlines, and equipment positions. Subsequently, based on the established feature correspondence, the system divides the complete surveillance video frame into several associated regions. The associated regions are used to represent local image ranges with the same reference source or similar change patterns, and each associated region can independently participate in subsequent change detection. For example, in a community entrance monitoring scenario, the system can divide the current video frame into a vehicle passage area, a pedestrian activity area, an access control area, and a background environment area. Among them, the vehicle passage area corresponds to changes in vehicle movement, the pedestrian activity area corresponds to changes in people entering or leaving, and the background environment area is mainly used to determine environmental changes such as lighting and weather. Furthermore, the system records the anchor point source information corresponding to each associated region to clarify the reference position and reference status of each region in the keyframe anchor point. The anchor point source information may include the historical reference frame number corresponding to the region, the spatial position of the region, the feature association relationship, and the reference status information, etc. For example, the device region in the current monitoring video frame may correspond to the third reference region in the historical keyframe anchor point. After the system records the source information of this region, in the subsequent detection process, when the device region changes, the corresponding anchor point information can be directly called for difference analysis without having to retrace the entire keyframe. For example, in a parking lot monitoring scenario, the system stores key frame anchor points for the parking lot's idle state, including the entrance area, parking space area, vehicle lane area, and lighting area. When a new monitoring video frame is acquired, the system matches it with historical key frames using features such as lane boundaries, parking space lines, and the location of fixed facilities to establish a regional mapping relationship and divide the current video frame into multiple associated regions. If a vehicle subsequently enters a parking space area, the system can use the anchor point source information corresponding to that area to perform difference calculations by only calling the reference state of that parking space area, without needing to analyze the entire parking lot screen.

[0041] Reference Appendix Figure 2 An inter-frame differential compression and storage system for a low-power camera according to one embodiment of the present invention includes: The calculation module 10 is used to associate the monitoring video frames with pre-stored key frame anchor points based on the monitoring video frames collected by the camera within a preset period, establish a corresponding reference frame cache, and calculate the inter-frame difference value between the monitoring video frames and the reference frame cache. The judgment module 20 is used to determine whether the inter-frame difference value is greater than a preset change threshold; The execution module 30 is used to perform sparse sampling of the still image according to a preset time interval if no, extract frame anchor representative points from the monitoring video frame to form corresponding monitoring index data, dynamically adjust the acquisition cycle of the next key frame according to the changing trend of the inter-frame difference value within several consecutive sampling periods, and generate a corresponding set of change event features through several consecutive changing frames. The monitoring index data specifically includes timestamps, device status information and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction and target dwell time. The second judgment module 40 is used to determine whether the set of change event features meets the preset abnormal event triggering conditions; The second execution module 50 is used to, if satisfied, collect cached video data within a preset time range before the triggering of the change event, splice the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, compress and manage the complete event video using a keyframe indexing method based on the event data of the complete event video, and dynamically update a preset inter-frame difference judgment model according to the target motion trajectory formed by the complete event video during the abnormal event. The event data specifically includes importance level, storage capacity and network status.

[0042] In this embodiment, the calculation module 10 associates the monitoring video frames acquired by the camera within a preset period with pre-stored key frame anchors to establish a corresponding reference frame buffer, calculates the inter-frame difference value between the monitoring video frame and the reference frame buffer, and then the judgment module 20 determines whether the inter-frame difference value is greater than a preset change threshold to execute the corresponding steps. For example, when the system determines that the inter-frame difference value between the monitoring video frame and the reference frame buffer is greater than the preset change threshold, the system will consider that the current scene is no longer in a low-change static state and needs to switch from low-power monitoring mode to change analysis mode. The system will extract the current monitoring video... For changes in a frame that differ from the reference frame buffer, the changes are regionalized based on their location, size, continuity, and direction of change, generating corresponding change area information. Simultaneously, a preset number of consecutive video frames are collected, and the change areas in each consecutive video frame are correlated to determine whether the change areas are generated by the same target or the same change process. For example, if the system determines that the inter-frame difference between the monitored video frame and the reference frame buffer is not greater than a preset change threshold, the execution module 30 will consider that the current monitored image has not changed significantly compared to the historical reference image corresponding to the key frame anchor point, and the system will... Sparse sampling of still images is performed at pre-set time intervals. Frame anchor points (representative keyframe anchor points) are extracted from the monitored video frames to form corresponding monitoring index data. This monitoring index data specifically includes timestamps, device status information, and environmental parameter information. Based on the changing trend of inter-frame difference values ​​over several consecutive sampling periods, the acquisition period of the next keyframe is dynamically adjusted. Through several continuously changing frames, a corresponding set of change event features is generated. This set of change event features specifically includes the duration of change, the diffusion speed of the change area, the direction of motion, and the target dwell time. The system uses the timestamps, device status information, and environmental parameter information corresponding to the keyframe anchor points to form... The monitoring index data can quickly locate the corresponding monitoring period without relying on a continuous and complete video stream. It can combine the device operating status and environmental change information to perform correlation management of monitoring data, improve the efficiency and management capability of video data retrieval in static scenes, and dynamically adjust the next key frame acquisition cycle based on the change trend of inter-frame difference value within a continuous sampling period. This allows the camera to adaptively change the data acquisition frequency according to the degree of change in the actual scene, further reducing energy consumption in image acquisition, processing and storage while ensuring monitoring continuity. Then, the second judgment module 40 judges whether these change event feature sets meet the preset abnormal event triggering conditions to execute the corresponding steps.For example, when the system determines that the set of change event features cannot meet the preset abnormal event triggering conditions, the system will consider that although there has been a certain degree of image change in the current monitored area, it may belong to ordinary environmental disturbance, normal personnel activity, lighting changes, or other low-risk changes. The system will extract key descriptive information that can reflect the current scene change status and associate the key descriptive information with the corresponding timestamp, equipment operating status, and environmental parameters to generate a low-priority change record. At the same time, based on the location of the current change area, the duration of the change, and the change trend in subsequent acquisition cycles, the system will adaptively adjust the current keyframe anchor point so that the updated keyframe anchor point can more accurately reflect the current monitoring environment status, serving as a reference for subsequent inter-frame difference calculations. For example, when the system determines that the set of change event features can meet the preset abnormal event triggering conditions, the second execution module 50 will consider that the current change is more likely to be an abnormal disturbance and needs to immediately enter the abnormal event recording and storage process. The system will collect the data before the change event is triggered. The system collects cached video data within a specified range and stitches it together with the real-time video stream triggered by the event to obtain a complete event video. Based on the event data of this complete event video, which includes importance level, storage capacity, and network status, the system compresses and manages the complete event video using keyframe indexing. It dynamically updates a pre-set inter-frame difference judgment model based on the target motion trajectory formed during the abnormal event. By stitching cached video data with the real-time video stream triggered by the event, and combining the event importance level, storage capacity, and network status for keyframe indexing compression management of the complete event video, the system can select an appropriate data storage method based on actual resource conditions. This ensures the complete preservation of the key processes of the abnormal event while reducing the occupation of invalid video data. Furthermore, by using keyframe indexing instead of traditional continuous video storage, the system reduces video data writing volume and transmission pressure, decreases local storage space consumption and cloud storage costs, and improves the data management efficiency of low-power camera equipment in long-term monitoring environments.

[0043] In this embodiment, it also includes: The generation module is used to calculate the stable parameters within the corresponding time window based on the image difference results acquired in several consecutive acquisition cycles, and to generate the corresponding scene stable sequence by combining the continuous stable parameters between adjacent time windows. The third judgment module is used to determine whether the scene stable sequence meets the preset stability maintenance conditions; The third execution module is used to establish a transitional cache set if not, temporarily store the current collected data in the transitional cache set, perform continuous consistency analysis on the transitional cache set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficient based on the local consistency degree, correct the scene stability sequence based on the stability correction coefficient, obtain the corrected stability sequence, and generate the corresponding stable evolution trajectory through the corrected stability sequence.

[0044] In this embodiment, the system calculates stable parameters within the corresponding time window based on the image difference results acquired over several consecutive acquisition cycles (stable parameters are quantitative indicators used to characterize the stable state of the current monitoring scene, calculated based on the degree of difference between continuously acquired images within a single time window). Combining this with continuous stable parameters between adjacent time windows (continuous stable parameters are continuous stable state description information formed by combining the changing relationships of stable parameters within multiple adjacent time windows), the system generates a corresponding scene stability sequence (a scene stability sequence is a data sequence reflecting the trend of the monitoring scene changing over time, formed by arranging continuous stable parameters corresponding to multiple consecutive time windows in chronological order). The system then determines whether these scene stability sequences meet pre-set stability maintenance conditions and executes corresponding steps accordingly. For example, when the system determines that these scene stability sequences meet the pre-set stability maintenance conditions, the system considers that the current monitoring area has not experienced significant image changes over several consecutive acquisition cycles, and the scene is in a continuously stable state. Based on the stability level corresponding to the scene stability sequence, the system reduces the processing frequency of the current monitoring video frames, extracts only representative image frames as stable state reference frames according to a preset sampling period, and associates these reference frames with corresponding time information, device operating status, and environmental parameters to generate a static scene index. Data is introduced; for example, when the system determines that these scene stable sequences cannot meet the pre-set stability maintenance conditions, the system will assume that there are no consecutive similar images in the current monitoring video, and a high proportion of duplicate data will not be generated. Therefore, there is no need to enter the static scene low-redundancy processing mode. The system will establish a transitional buffer set, temporarily store the current collected data in the transitional buffer set, perform continuous consistency analysis on the transitional buffer set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficients based on different local consistency degrees, and correct the scene stable sequences based on these stability correction coefficients to obtain the corrected stable sequences. Through the corrected stable sequences, data is generated... The system generates a corresponding stable evolution trajectory. By performing consistency analysis on continuously collected data in the transition buffer set, the system calculates the degree of local consistency between different collection times and obtains the corresponding stability correction coefficient based on different degrees of local consistency. This enables dynamic correction for monitoring scenarios with different degrees of change, avoiding the problem of abnormal decline in stable sequences caused by non-critical factors such as changes in lighting, slight occlusion, and equipment vibration. At the same time, by generating a stable evolution trajectory based on the corrected stable sequence, the system can record the continuous evolution process of the monitoring scenario from a stable state to a changing state, providing a more complete data basis for subsequent judgment on whether to enter the change detection, event analysis, or data augmentation collection stage.

[0045] In this embodiment, the execution module further includes: A submodule is established to establish a corresponding change association chain based on the preset association data of the continuously changing frames. The association data specifically includes the order of occurrence, duration, spatial distribution, and location migration relationship. The judgment submodule is used to determine whether the changed association chain meets the preset continuous association condition; The execution submodule is used to, if not, split the set of change event features, divide the abnormal change information into several independent change units, calculate the persistence influence coefficient corresponding to the independent change unit, rearrange the processing priority of the independent change units according to the persistence influence coefficient, construct the corresponding candidate event queue according to the processing priority, store the candidate event queue in the set to be verified, dynamically receive new image information during subsequent acquisition, perform incremental matching on the candidate event queue, expand and update the candidate event queue through the continuous association relationship formed by the new image information and the candidate event queue, and regenerate the corresponding set of change event features. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

[0046] In this embodiment, the system establishes a corresponding change association chain based on pre-set association data of continuously changing frames. This association data specifically includes the order of occurrence, duration, spatial distribution, and location migration relationships. The system then determines whether the change association chain meets pre-set continuous association conditions to execute corresponding steps. For example, when the system determines that the change association chain meets the pre-set continuous association conditions, the system considers that the current change may be continuously generated by the same target movement, the same behavioral process, or the same environmental state change. The system then performs continuous association based on the order of occurrence, duration, spatial distribution, and location migration relationships in the change association chain. The system integrates and analyzes the changing frames to determine the relationships between the changing nodes, extracts the corresponding change development information, and generates corresponding change trajectory description data through the change association chain. It records the target's movement direction, activity range, and location during the change process. For example, if the system determines that the change association chain cannot meet the pre-set continuous association conditions, it will consider the change in the current monitoring screen to be caused by a single isolated frame change or transient noise. The system will then split the feature set of the change event, dividing the abnormal change information into several independent change units. The abnormal change information specifically includes excessively long time intervals and spatial spans. For excessively large changes and inconsistent evolution directions, the system calculates the persistent impact coefficients of these independent change units. Based on these coefficients, the processing priorities of these units are rearranged, and candidate event queues are constructed according to these priorities. These queues are then stored in a set to be verified. During subsequent data acquisition, new image information is dynamically received, and incremental matching is performed on these queues. By establishing continuous associations between new image information and the queues, the system expands and updates these queues, regenerating the corresponding change event feature set. By calculating the persistent impact coefficients of each independent change unit and rearranging processing priorities based on these coefficients, the system can allocate resources according to the potential impact of each change unit on the subsequent monitoring status. Changes with the potential for continued development are prioritized for retention and analysis, while low-impact, short-lived changes are downgraded. Furthermore, by dynamically receiving new image information and incrementally matching it with the queues, the system can continuously track weak change processes that do not initially meet the conditions for continuous association. When subsequent change information establishes a continuous association with existing queues, the system automatically expands and updates the queues, regenerating a complete change event feature set.

[0047] In this embodiment, the second execution module further includes: The generation submodule is used to identify the association features between the end node of the cached video and the start node of the real-time video, and generate video stitching parameters based on the association features; The second judgment submodule is used to determine whether the video connection parameters meet the preset continuous association conditions; The second execution submodule is used to, if not, reversely retrieve the end position of the cached video and the start position of the real-time video, re-find the associated video segment with the start frame of the real-time video from the cache area, divide the corresponding splicing boundary according to the retrieval result, perform transition correction on the video data between the splicing boundary, dynamically adjust the consistency of motion trajectory between adjacent video frames, and recombine the corrected video data with the real-time video stream to generate a continuous event video.

[0048] In this embodiment, the system identifies the correlation features between the end node of the cached video and the start node of the real-time video based on the end node and start node. Based on these correlation features, video splicing parameters are generated. The system then determines whether these video splicing parameters meet pre-set continuous correlation conditions to execute corresponding steps. For example, if the system determines that these video splicing parameters meet the pre-set continuous correlation conditions, it considers the cached video data and the real-time video stream to belong to the same event development process and can directly perform video fusion processing to form a complete and continuous event record. The system determines the optimal connection position between the end node of the cached video and the start node of the real-time video based on the video splicing parameters, arranges and merges the two parts of video data in chronological order, and adjusts for possible time deviations, inter-frame intervals, and differences in image changes between the cached video and the real-time video using the video splicing parameters to ensure continuous playback of the spliced ​​video data. Conversely, if the system determines that these video splicing parameters do not meet the pre-set continuous correlation conditions, the system considers the two video data segments to be... Unable to achieve natural transitions in time sequence and content changes, exhibiting obvious video gaps, abrupt scene changes, or sudden changes in target state, the system reversely searches the end position of the cached video and the start position of the real-time video. It then re-searches the cached area for the associated video segment (the segment with the highest degree of correlation) with the starting frame of the real-time video. Based on different search results, corresponding splicing boundaries are defined, and the video data between these splicing boundaries undergoes transitional correction. This dynamically adjusts the consistency of motion trajectories between adjacent video frames, and the corrected video data is then recombined with the real-time video stream to generate continuous event video. By defining corresponding splicing boundaries based on different search results and performing transitional correction on the video data between these boundaries, the system can dynamically repair issues such as inconsistent time intervals, significant changes in scene state, and discontinuous target motion. Furthermore, by recombining the corrected video data with the real-time video stream to generate continuous event video, it can still restore a relatively complete event process even when the original video segments have splicing defects, improving the usability and traceability of abnormal event recordings.

[0049] In this embodiment, the determination module further includes: The second generation submodule is used to generate inter-frame difference description parameters based on the difference data between the current video frame and the reference video frame. The third judgment submodule is used to determine whether the difference description parameter exceeds a preset change limit; The third execution submodule is used to, if not, collect low-amplitude change information in the inter-frame difference description parameters, extract persistent background change features from the low-amplitude change information, match the background change features with environmental change features saved in the historical monitoring period, dynamically classify the environmental change attribution of the monitoring video frame according to the matching result, generate corresponding environmental adaptation parameters, and adjust the update period of the reference video frame according to the environmental adaptation parameters. The environmental change attribution specifically includes periodic environmental changes and known environmental changes.

[0050] In this embodiment, the system generates inter-frame difference description parameters based on the difference data between the current video frame and the reference video frame. The system then determines whether these difference description parameters exceed a preset change limit to execute corresponding steps. For example, if the system determines that the inter-frame difference description parameters have exceeded the preset change limit, the system considers the current change not to be caused by minor noise, device jitter, or normal image fluctuations, and therefore has some effective information value. The system will determine the change area in the current video frame based on the difference description parameters, extract the location, range, and degree of change of the change area, and continuously track and analyze the current change area. Instead of immediately triggering event processing based on the single-frame change result, the system continues to collect several subsequent video frames to determine whether the change persists and whether it has a continuous change pattern. Conversely, if the system determines that the inter-frame difference description parameters have not exceeded the preset change limit, the system considers the current video frame to have not undergone significant content changes compared to the reference video frame, and the current monitoring image remains in its original stable state. The system will collect these inter-frame differences... The system extracts persistent background change features from low-amplitude change information in the description parameters, matches different background change features with environmental change features saved in historical monitoring periods, and dynamically classifies the environmental change attribution of monitoring video frames based on the matching results. The environmental change attribution specifically includes periodic environmental changes and known environmental changes, generating corresponding environmental adaptation parameters. Based on the environmental adaptation parameters, the update cycle of reference video frames is adjusted. By matching different background change features with environmental change features saved in historical monitoring periods, the system can analyze the current scene status in conjunction with historical environmental change patterns. For example, it can identify changes in day and night illumination, weather changes, changes in brightness in fixed areas, and periodic environmental disturbances. Based on the matching results, the system dynamically classifies the environmental change attribution of monitoring video frames. At the same time, by adjusting the update cycle of reference video frames based on environmental adaptation parameters, the system can dynamically maintain reference video frames according to the current environmental stability. When environmental changes are small and remain stable for a long time, the update cycle of reference video frames is extended to reduce invalid update operations and data processing overhead.

[0051] In this embodiment, the second determination module further includes: The calculation submodule is used to match the event behavior description parameters corresponding to the set of changed event features with the historical behavior data corresponding to the current monitoring area, obtain the degree of difference between the current changed behavior and the normal behavior pattern, and calculate the event credibility evaluation value based on the degree of difference. The fourth judgment submodule is used to determine whether the event credibility evaluation value reaches the preset credibility level; The fourth execution submodule is used to generate event confirmation information corresponding to the event credibility evaluation value if the event is true, and to call pre-stored video cache data according to the event confirmation information, to filter video segments related to the current change area from the cache time window before the event occurred, to perform validity filtering on the cached video according to the target activity trajectory, and to dynamically remove video content that is irrelevant to the event. The event confirmation information specifically includes the scope of the event's impact, the target activity area, and the direction of the event's development.

[0052] In this embodiment, the system matches the event behavior description parameters corresponding to the set of changed event features with the historical behavior data corresponding to the current monitoring area to obtain the degree of difference between the current changed behavior and the normal behavior pattern. Based on different degrees of difference, the system calculates the event credibility evaluation value. Then, the system determines whether the event credibility evaluation value reaches the preset credibility level to execute the corresponding steps. For example, when the system determines that the event credibility evaluation value has not reached the preset credibility level, the system will consider that situations such as people staying briefly in the monitoring area, objects being temporarily moved and then returning to their original position, and changes in the target outline caused by changes in lighting may all form certain change characteristics, but their behavior patterns are relatively close to historical normal behavior. Therefore, they are not identified as abnormal events. After determining that the event credibility evaluation value has not reached the credibility level, the system will not immediately trigger the complete event recording and storage process. Instead, it will reduce the processing level of the current event, mark the changed behavior as an event to be observed, retain the corresponding event behavior description parameters, and simultaneously perform delayed tracking of the video change process corresponding to the event to be observed. In subsequent acquisition cycles, it will continue to acquire new monitoring video frames and analyze the lines in the new video frames. To determine whether the change deviates further from the historical normal behavior pattern; for example, when the system determines that the credibility evaluation value of the event has reached a pre-set credibility level, the system will consider the current change event to have caused a certain degree of image change and belong to a real abnormal event. The system will generate event confirmation information corresponding to the event credibility evaluation value. The event confirmation information specifically includes the scope of the event's impact, the target activity area, and the event's development direction. Based on this event confirmation information, the system calls the pre-stored video cache data, filters video segments related to the current change area from the cache time window before the event occurred, and performs validity filtering of the cached video according to the target activity trajectory, dynamically removing video content irrelevant to the event. By calling the pre-stored video cache data according to the event confirmation information and filtering video segments related to the current change area from the cache time window before the event occurred, the system can effectively retain the key process before the abnormal event occurred, avoiding the loss of important pre-event images due to delays in event triggering. At the same time, by dynamically removing video content irrelevant to the event according to the target activity trajectory, the system can reduce the size of video data while ensuring the complete recording of abnormal events, reducing subsequent storage space occupation and network transmission pressure.

[0053] In this embodiment, the calculation module further includes: The third generation submodule is used to generate scene description features corresponding to the monitoring video frame based on the preset acquisition conditions of the camera in the current scene. The acquisition conditions specifically include spatial layout information, target distribution information and scene composition information. The fifth judgment submodule is used to determine whether the scene description features can match the pre-stored keyframe anchor points; The fifth execution submodule is used to, if possible, establish an image region mapping relationship based on the feature correspondence between the monitoring video frame and the reference anchor point, divide the monitoring video frame into several associated regions, and record the anchor point source information corresponding to each associated region.

[0054] In this embodiment, the system generates scene description features corresponding to the monitoring video frame based on the pre-set acquisition conditions of the low-power camera in the current frame. The acquisition conditions specifically include spatial layout information, target distribution information, and scene composition information. Then, the system determines whether these video description features can match pre-stored keyframe anchors to execute corresponding steps. For example, when the system determines that the scene description features corresponding to the monitoring video frame cannot match the pre-stored keyframe anchors, the system will consider that the current scene description features cannot establish a valid correspondence with the existing keyframe anchors, and the original reference state can no longer accurately describe the current monitoring environment. The system will mark the current monitoring video frame as a scene state to be updated, further analyze the unmatched scene description features to determine the reason for the matching failure, and at the same time, re-establish a temporary reference state based on the currently unmatched scene description features, and extract new scene feature information from the currently acquired video frame to form candidate keyframe anchors. For example, when the system determines that the scene description features corresponding to a monitored video frame can match pre-stored keyframe anchor points, the system assumes that there is no significant difference between the currently acquired video frame and the historically saved key reference states. Based on the feature correspondence between the monitored video frame and the reference anchor points, the system establishes an image region mapping relationship, dividing the monitored video frame into several associated regions and recording the anchor point source information corresponding to each associated region. By establishing an image region mapping relationship based on the feature correspondence between the monitored video frame and the reference anchor points, the system can associate different regions in the current monitored image with corresponding regions in the historical reference states, thereby achieving regionalized management of the monitored scene. Simultaneously, by dividing the monitored video frame into several associated regions and recording the anchor point source information corresponding to each associated region, the system can retain reference data for different regions, providing a more accurate data foundation for subsequent inter-frame difference calculation, change area location, and event analysis.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for inter-frame differential compression storage of a low-power camera, characterized in that, Includes the following steps: Based on the surveillance video frames collected by the camera within a preset period, the surveillance video frames are associated with pre-stored key frame anchor points to establish a corresponding reference frame cache, and the inter-frame difference value between the surveillance video frames and the reference frame cache is calculated. Determine whether the inter-frame difference value is greater than a preset change threshold; If not, sparse sampling is performed on the still image according to a preset time interval, frame anchor representative points are extracted from the monitoring video frame to form corresponding monitoring index data, and the acquisition cycle of the next key frame is dynamically adjusted according to the changing trend of the inter-frame difference value within several consecutive sampling periods. Through several consecutive changing frames, a corresponding set of change event features is generated. The monitoring index data specifically includes timestamps, device status information and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction and target dwell time. Determine whether the set of change event features meets the preset abnormal event triggering conditions; If the conditions are met, cached video data within a preset time range before the change event is triggered is collected. The cached video data is then spliced ​​with the real-time video stream after the change event is triggered to obtain a complete event video. Based on the event data of the complete event video, the complete event video is compressed and managed using a keyframe indexing method. According to the target motion trajectory formed by the complete event video during the abnormal event, the preset inter-frame difference judgment model is dynamically updated. The event data specifically includes importance level, storage capacity, and network status.

2. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, Before the step of sparsely sampling still images according to a preset time interval, extracting frame anchor representative points from the monitoring video frames, and forming corresponding monitoring index data, the method further includes: Based on the image difference results acquired in several consecutive acquisition cycles, the stable parameters within the corresponding time window are calculated, and the corresponding scene stable sequence is generated by combining the continuous stable parameters between adjacent time windows. Determine whether the stable sequence of the scenario meets the preset stability maintenance conditions; If not, a transitional cache set is established, and the current collected data is temporarily stored in the transitional cache set. Continuous consistency analysis is performed on the transitional cache set to calculate the degree of local consistency between each collection moment. Based on the degree of local consistency, the corresponding stability correction coefficient is obtained. The scene stability sequence is corrected based on the stability correction coefficient to obtain the corrected stability sequence. The corresponding stable evolution trajectory is generated through the corrected stability sequence.

3. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, The step of dynamically adjusting the acquisition period of the next key frame based on the changing trend of inter-frame difference values ​​within a series of consecutive sampling periods, and generating a corresponding set of change event features through a series of consecutive changing frames, further includes: Based on the preset associated data of the continuously changing frames, a corresponding change association chain is established, wherein the associated data specifically includes the order of occurrence, duration, spatial distribution and location migration relationship; Determine whether the changed association chain satisfies the preset continuous association condition; If not, the set of change event features is split into several independent change units, the abnormal change information is divided into several independent change units, the persistence influence coefficient of the independent change units is calculated, the processing priority of the independent change units is rearranged according to the persistence influence coefficient, the corresponding candidate event queue is constructed according to the processing priority, the candidate event queue is stored in the set to be verified, new image information is dynamically received during subsequent acquisition, incremental matching is performed on the candidate event queue, the candidate event queue is expanded and updated through the continuous association relationship formed by the new image information and the candidate event queue, and the corresponding set of change event features is regenerated. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

4. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, The step of collecting cached video data within a preset time range before the triggering of the change event, and then splicing the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, further includes: Based on the end node of the cached video and the start node of the real-time video, the association features between the end node and the start node are identified, and video connection parameters are generated according to the association features. Determine whether the video stitching parameters meet the preset continuous association conditions; If not, the process reverses the search to find the end position of the cached video and the start position of the real-time video, searches for the associated video segment with the start frame of the real-time video in the cache area, divides the corresponding splicing boundaries according to the search results, performs transition correction on the video data between the splicing boundaries, dynamically adjusts the consistency of motion trajectories between adjacent video frames, and recombines the corrected video data with the real-time video stream to generate a continuous event video.

5. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, The step of determining whether the inter-frame difference value is greater than a preset change threshold further includes: Based on the difference data between the current video frame and the reference video frame, generate inter-frame difference description parameters; Determine whether the difference description parameter exceeds a preset change limit; If not, low-amplitude change information in the inter-frame difference description parameters is collected, persistent background change features are extracted from the low-amplitude change information, the background change features are matched with environmental change features saved in the historical monitoring period, and the environmental change attribution of the monitoring video frame is dynamically classified according to the matching result, corresponding environmental adaptation parameters are generated, and the update period of the reference video frame is adjusted according to the environmental adaptation parameters. The environmental change attribution specifically includes periodic environmental changes and known environmental changes.

6. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, The step of determining whether the set of change event features meets the preset abnormal event triggering conditions further includes: Based on the event behavior description parameters corresponding to the set of changed event features, the event behavior description parameters are matched with the historical behavior data corresponding to the current monitoring area to obtain the degree of difference between the current changed behavior and the normal behavior pattern. Based on the degree of difference, the event credibility evaluation value is calculated. Determine whether the credibility evaluation value of the event reaches the preset credibility level; If so, then generate event confirmation information corresponding to the event credibility evaluation value, and based on the event confirmation information, call the pre-stored video cache data, filter video segments related to the current change area from the cache time window before the event occurred, filter the cached videos for validity based on the target activity trajectory, and dynamically remove video content unrelated to the event. The event confirmation information specifically includes the scope of the event's impact, the target activity area, and the direction of the event's development.

7. The inter-frame differential compression storage method for a low-power camera according to claim 1, characterized in that, The step of associating the surveillance video frames acquired by the camera within a preset period with pre-stored keyframe anchor points further includes: Based on the preset acquisition conditions of the camera in the current scene, scene description features corresponding to the monitoring video frame are generated, wherein the acquisition conditions specifically include spatial layout information, target distribution information and scene composition information; Determine whether the scene description features can match pre-stored keyframe anchor points; If possible, an image region mapping relationship is established based on the feature correspondence between the monitoring video frame and the reference anchor point, the monitoring video frame is divided into several associated regions, and the anchor point source information corresponding to each associated region is recorded respectively.

8. An inter-frame differential compression storage system for a low-power camera, characterized in that, include: The calculation module is used to associate the monitoring video frames collected by the camera within a preset period with pre-stored key frame anchor points, establish a corresponding reference frame cache, and calculate the inter-frame difference value between the monitoring video frames and the reference frame cache. The judgment module is used to determine whether the inter-frame difference value is greater than a preset change threshold; The execution module is used to perform sparse sampling of still images according to a preset time interval if no, extract frame anchor representative points from the monitoring video frames to form corresponding monitoring index data, dynamically adjust the acquisition cycle of the next key frame based on the changing trend of inter-frame difference values ​​within several consecutive sampling periods, and generate a corresponding set of change event features through several consecutive changing frames. The monitoring index data specifically includes timestamps, device status information, and environmental parameter information, and the set of change event features specifically includes change duration, change area diffusion speed, movement direction, and target dwell time. The second judgment module is used to determine whether the set of change event features meets the preset abnormal event triggering conditions; The second execution module is used to, if satisfied, collect cached video data within a preset time range before the triggering of the change event, splice the cached video data with the real-time video stream after the triggering of the change event to obtain a complete event video, compress and manage the complete event video using a keyframe indexing method based on the event data of the complete event video, and dynamically update a preset inter-frame difference judgment model according to the target motion trajectory formed by the complete event video during the abnormal event. The event data specifically includes importance level, storage capacity and network status.

9. The inter-frame differential compression storage system for a low-power camera according to claim 8, characterized in that, Also includes: The generation module is used to calculate the stable parameters within the corresponding time window based on the image difference results acquired in several consecutive acquisition cycles, and to generate the corresponding scene stable sequence by combining the continuous stable parameters between adjacent time windows. The third judgment module is used to determine whether the scene stable sequence meets the preset stability maintenance conditions; The third execution module is used to establish a transitional cache set if not, temporarily store the current collected data in the transitional cache set, perform continuous consistency analysis on the transitional cache set, calculate the local consistency degree between each collection time, obtain the corresponding stability correction coefficient based on the local consistency degree, correct the scene stability sequence based on the stability correction coefficient, obtain the corrected stability sequence, and generate the corresponding stable evolution trajectory through the corrected stability sequence.

10. The inter-frame differential compression storage system for a low-power camera according to claim 8, characterized in that, The execution module further includes: A submodule is established to establish a corresponding change association chain based on the preset association data of the continuously changing frames. The association data specifically includes the order of occurrence, duration, spatial distribution, and location migration relationship. The judgment submodule is used to determine whether the changed association chain meets the preset continuous association condition; The execution submodule is used to, if not, split the set of change event features, divide the abnormal change information into several independent change units, calculate the persistence influence coefficient corresponding to the independent change unit, rearrange the processing priority of the independent change units according to the persistence influence coefficient, construct the corresponding candidate event queue according to the processing priority, store the candidate event queue in the set to be verified, dynamically receive new image information during subsequent acquisition, perform incremental matching on the candidate event queue, expand and update the candidate event queue through the continuous association relationship formed by the new image information and the candidate event queue, and regenerate the corresponding set of change event features. The abnormal change information specifically includes excessively long time intervals, excessively large spatial spans, and inconsistent evolution directions.

Images