A smart community property management method based on image recognition technology

Abstract

The application discloses a kind of intelligent community property management methods based on image recognition technology, it is related to image recognition technology field, including, calibration each camera picture in property space anchor point, and the object handover line between adjacent camera is associated, configuration space connection relationship, form anchor point connection set;Using anchor point connection set to identify the property object in real-time video stream, determine the contact relationship and action state of property object and property space anchor point, and arrange anchor point contact sequence according to time sequence, form event fragment set;Verify the reset state of the corresponding terminal anchor point of disposal page, and continue video identification and event continuation when terminal anchor point is not reset, form property disposal record.The application realizes the closed-loop management of event disposal by the continuous verification of terminal anchor point reset state and the dynamic continuation under the condition of not resetting.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a smart community property management method based on image recognition technology. Background Technology

[0002] With the advancement of smart community construction, intelligent property management technology based on video images has been widely applied. Existing methods typically involve deploying surveillance cameras in public areas of the community to detect and track objects in real-time video streams, identifying property objects such as people, vehicles, and items. Based on single frames or short clips, they determine whether abnormal behavior exists, such as blocked fire lanes, electric vehicles entering elevators, or haphazardly piled garbage. This type of technology mainly relies on image classification, object detection, and behavior recognition models to perform generalized analysis of the surveillance footage and trigger alarms when suspected anomalies are detected.

[0003] However, existing technologies have the following shortcomings: First, the recognition in the surveillance footage lacks deep coupling with the physical spatial structure, making it impossible to accurately determine the precise spatial relationship between the object and specific property facilities (such as unit doors, elevator doors, and fire escape boundaries), which easily leads to false alarms and missed alarms. Second, cross-camera event tracking usually relies solely on matching the appearance features of the object, making it difficult to maintain the continuity of events when there are camera blind spots or changing perspectives. Furthermore, the lack of a complete record and semantic description of the event's development process results in isolated and fragmented alarm information, failing to provide property management personnel with complete traceability evidence from the occurrence to the end of the event, and also making it difficult to effectively verify the event handling results. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a smart community property management method based on image recognition technology to solve the problem that the lack of deep coupling with physical space and insufficient semantic association of events makes it impossible to achieve accurate tracing and handling from spatial perception to event closure.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a smart community property management method based on image recognition technology, which includes: calibrating property space anchor points in the images of each camera, associating object handover lines between adjacent cameras, configuring spatial connection relationships, and forming an anchor point connection set;

[0008] Anchor point connection sets are used to identify property objects in real-time video streams, determine the contact relationship and action status between property objects and anchor points in the property space, and arrange the anchor point contact sequence in chronological order to form an event fragment set;

[0009] The event clips from adjacent cameras are spliced ​​together according to object continuation, action state continuation, and anchor point contact sequence to form a property event chain.

[0010] Extract the starting point, turning point, and ending point of the event chain in the industry, and based on the extracted starting point, turning point, and ending point, complete the event characterization, locate the responsible party, and arrange the capture path to form a handling page;

[0011] The system verifies the reset status of the corresponding endpoint anchor point on the processing page, and continues video recognition and event continuation if the endpoint anchor point is not reset, thus forming a property management processing record.

[0012] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the property space anchor points include unit door frame anchor points, door leaf anchor points, equipment room door handle anchor points, doorway boundary anchor points, elevator threshold anchor points, car boundary anchor points, fire lane edge anchor points, garbage disposal outlet boundary anchor points, corridor passable zone boundary anchor points, and public facility reset reference anchor points.

[0013] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the object handover line between adjacent cameras refers to marking a virtual handover line at the boundary of the overlapping field of view and blind zone of adjacent cameras, connecting the last visible anchor point of the previous camera with the first visible anchor point of the next camera, marking the crossing direction and assigning camera number pairs.

[0014] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the coordinates of the property space anchor points in each camera and the connection direction of the intersection line between adjacent cameras are extracted.

[0015] Connect the property space anchor points within the same camera according to their spatial adjacency to form an anchor point topology map, and use the intersection line as the cross-camera anchor point connection edge, while configuring traversable type and direction constraints.

[0016] Summarize the anchor point topology and cross-camera connection edges to generate an anchor point connection set.

[0017] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the step of forming an event fragment set specifically involves:

[0018] Based on the anchor point connection set, the contours of objects in the real-time video stream are extracted, and the intersection-union ratio and positional changes of the contours and anchor point regions are detected.

[0019] When the outline touches the anchor point area and the action triggering condition is met, a relational unit containing the object type, anchor point identifier, and action type is generated.

[0020] Connect the continuously generated relational units of the same property object group within the same camera in chronological order, and extract the anchor point identifiers corresponding to each relational unit to form anchor point contact sequence identifiers;

[0021] Each continuous relational unit and its corresponding anchor point contact sequence are encapsulated as an event fragment and summarized to form an event fragment set.

[0022] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the formation of the property event chain specifically includes:

[0023] Load the anchor point connection set and the event fragment set of the front and rear cameras, and extract the intersection line and direction constraints between adjacent cameras;

[0024] Compare the object contour similarity and motion direction consistency between the previous event segment and the next event segment, filter candidate segment pairs for object continuation, and parse the termination anchor point and last action state of the previous event segment, and the starting anchor point and first action state of the next event segment.

[0025] Query whether there is a junction line connecting the terminating anchor point and the starting anchor point in the anchor point connection set, and whether the last action state and the first action state meet the traversable type constraint. At the same time, extract the continuous anchor point segments at the end of the anchor point contact sequence of the previous event segment and the continuous anchor point segments at the beginning of the anchor point contact sequence of the next event segment, and verify whether a spatial continuous path passing through the junction line is formed in the anchor point topology graph.

[0026] When the sequence of objects, actions, and anchor points is continuous, the event fragments before and after merging constitute the same property event chain.

[0027] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the qualitative analysis of the completion event specifically includes:

[0028] Traverse all relational units in the property event chain, extract the object type and anchor point identifier of the first relational unit as the event start point feature, and extract the anchor point identifier of the last relational unit as the event end point position;

[0029] Detect nodes where the action state of the relationship unit changes abruptly, and extract nodes where the action changes from moving to stationary, from approaching to contacting, or from complete to separating as event turning points;

[0030] Read the anchor point type corresponding to the event endpoint and the sequence of action states that appear consecutively in the event chain, combine and match them with a preset violation type library, and output the event qualitative result.

[0031] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the location of the responsible object specifically refers to...

[0032] Extract the object type and outline corresponding to the first relational unit from the event origin features, and mark it as the main responsible object;

[0033] Extract the identifiers of auxiliary objects associated with the main responsible party in the property event chain. When the main responsible party and auxiliary objects are continuously bound in the contact path corresponding to the property event chain, they are marked as joint responsible parties and their identifiers are extracted.

[0034] Extract the first clear snapshot frame of the main responsible party in the event chain as the initial image of the responsible party, and extract the snapshot frame of the main responsible party at the turning point as the image confirming the responsible behavior.

[0035] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the capture path arrangement is specifically as follows:

[0036] The starting image is the initial image of the responsible party, the turning image is the image confirming the responsible behavior, and the corresponding unit's image is extracted from the end position of the property event chain as the ending image.

[0037] Read the event type identifier from the event qualitative results, and extract the endpoint anchor identifier from the anchor contact sequence identifier of the property event chain;

[0038] Load the camera layout diagram and handover line connection relationship associated with the anchor point connection set, map the anchor point contact sequence identifier of the property event chain to the camera, and generate a cross-area path map;

[0039] The starting point image, turning point image, ending point image, cross-regional route map, event type identifier, ending point anchor point identifier, and related liability object identifier are combined to generate a processing page.

[0040] As a preferred embodiment of the smart community property management method based on image recognition technology described in this invention, the step of forming property disposal records specifically involves:

[0041] Call the endpoint anchor point identifier in the processing page and load the corresponding pre-stored baseline contour map;

[0042] Real-time detection of the current frame image of the endpoint anchor point area, extraction of the property component outline associated with the endpoint anchor point, and calculation of the overlap between the current outline and the baseline outline map;

[0043] When the overlap of the outline exceeds the preset reset threshold, it is determined that the endpoint anchor point has been reset, and the property event chain, the handling page and the reset timestamp are merged and archived as a property handling record.

[0044] When the contour overlap does not exceed the reset threshold, it is determined that the endpoint anchor point has not been reset, and real-time video recognition of the current anchor point area continues to generate new relational units and new anchor point contact sequences.

[0045] The newly generated relational unit is connected to the end of the original property event chain, and the new anchor contact sequence is connected to the anchor contact sequence of the original property event chain. The duration, end anchor identifier and termination action status of the property event chain are updated until the contour overlap exceeds the preset reset threshold, and then a property disposal record is generated.

[0046] The beneficial effects of this invention are as follows: By constructing a property space anchor point connection set, video recognition is elevated from generalized image analysis to a precise perception level coupled with physical space semantics, effectively solving the problem of false alarms and missed alarms caused by fuzzy spatial relationships in traditional methods; by arranging the contact sequence between objects and anchor points, discrete single-frame detection is transformed into event fragments with spatial logic, providing a structured data foundation for event understanding; through a triple continuation mechanism of object, action, and anchor point sequences, cross-camera event chain splicing is achieved, overcoming the problem of event breakage caused by blind spots and perspective switching, ensuring the integrity and continuity from the start to the end of the event; In the event characterization phase, a comprehensive judgment is made based on the starting point, turning point, and ending point of the event chain, replacing the traditional single-frame or short-term judgment, which significantly improves the accuracy and interpretability of violation identification. At the same time, by locating the responsible party and generating cross-regional path maps, clear responsibility identification and tracing basis are provided for property management. Through continuous verification of the endpoint anchor point reset status and dynamic continuation in the case of non-reset, closed-loop management of event handling is realized, ensuring that violations are effectively corrected and retaining the complete event chain as a reference for subsequent processing, thereby improving the overall intelligence level and execution efficiency of smart community property management. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a smart community property management method based on image recognition technology.

[0049] Figure 2 Build a flowchart for the anchor point connection set.

[0050] Figure 3 Generate a flowchart for the event fragment set.

[0051] Figure 4 A flowchart for assembling the property event chain. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figures 1-4 As an embodiment of the present invention, this embodiment provides a smart community property management method based on image recognition technology, including the following steps:

[0056] S1. Mark the property space anchor points in the images of each camera, associate the object handover lines between adjacent cameras, configure the spatial connection relationship, and form an anchor point connection set.

[0057] S1.1. In the real-time video footage of the existing cameras in the community, fixedly mark image anchor points directly related to property management to generate a property space anchor point map. The specific marking method is as follows: In the video footage of each camera, use image annotation tools to mark the anchor points of the unit door frame, the unit door leaf, the equipment room door handle, the equipment room door boundary, the elevator threshold, the elevator car boundary, the fire escape side lines, the garbage disposal area, the corridor passable area, and the public facility reset reference anchor points in their normal fixed placement positions. Each anchor point is stored in the video footage as a coordinate point or coordinate area, forming the property space anchor point map of the camera.

[0058] S1.2. Between the property space anchor point maps of adjacent cameras, mark the object handover line used to transmit the object entry and exit relationship. The specific marking method is as follows: for each pair of adjacent cameras, mark a virtual handover line at the boundary of the overlapping area or blind zone of the adjacent camera's field of view.

[0059] The property space anchor point corresponding to the last visible position within the field of view of the preceding camera is used as the starting anchor point of the crossover line, and the property space anchor point corresponding to the first visible position within the field of view of the following camera is used as the ending anchor point of the crossover line. The starting and ending anchor points are connected by a virtual crossover line. The object crossing direction attribute is marked on the virtual crossover line, which includes forward crossing and reverse crossing. At the same time, a corresponding pair of adjacent camera numbers is assigned to each virtual crossover line. The pair of adjacent camera numbers consists of the number of the preceding camera and the number of the following camera.

[0060] S1.3. Based on the property space anchor point map of each camera and the object intersection line between each adjacent camera, configure spatial connection relationships to form an anchor point connection set. The specific configuration method is as follows: extract the coordinate position of each property space anchor point in the property space anchor point map of each camera, and extract the connection direction of each object intersection line between each adjacent camera. The connection direction includes the direction from the starting anchor point to the ending anchor point; connect all property space anchor points in the same camera according to the spatial adjacency relationship to construct the anchor point topology map of the camera. The spatial adjacency relationship means that the Euclidean distance between two property space anchor points in the image coordinate system is less than a preset distance threshold.

[0061] To further explain, the distance threshold is set based on the camera's image resolution and the typical spacing between adjacent property components in the actual scene, with a value of 100 pixels. This is because the distance between the unit door frame anchor point and the door leaf anchor point in the image is usually 30 to 80 pixels, the distance between the elevator threshold anchor point and the car boundary anchor point is usually 50 to 100 pixels, and the distance between the equipment room door handle anchor point and the doorway boundary anchor point is usually 20 to 60 pixels. Therefore, setting the distance threshold to 100 pixels can cover the maximum distance between adjacent anchor points in the same local area, while effectively avoiding incorrect connection of anchor points in different areas (such as the unit door anchor point and the fire escape edge anchor point, whose image distance is usually more than 200 pixels), thereby ensuring the accuracy of the anchor point topology map.

[0062] The calibrated object intersection line is used as the anchor point connection edge across the camera. The anchor point connection edge is added between the anchor point topology graphs of two adjacent cameras. The traversable type and direction constraint are configured for each anchor point connection edge. The traversable type includes people traversable, vehicles traversable, and objects traversable. The direction constraint includes allowing only forward traversal, allowing only reverse traversal, and allowing bidirectional traversal.

[0063] Summarize the anchor point topology of all cameras and all anchor point connection edges across cameras, and generate an anchor point connection set containing anchor point identifiers, anchor point coordinates, anchor point topology, cross-camera connection edges, traversable types, and directional constraints.

[0064] S2. Use anchor point connection sets to identify property objects in real-time video streams, determine the contact relationship and action status between property objects and anchor points in the property space, and arrange the anchor point contact sequence in chronological order to form an event fragment set.

[0065] S2.1. Based on the anchor point connection set, perform property object recognition and contour extraction on each frame of the real-time video stream. The specific recognition method is as follows: perform target detection on each frame of the real-time video stream, extract the contours of property objects belonging to people, non-motorized vehicles, electric vehicles, motorized vehicles, packages, decoration materials, garbage bags, handcarts, pets, as well as door leaf, garbage can lid, and equipment room door, and assign a unique object identifier to each extracted property object contour to form a property object group. At the same time, record the contour coordinate set and bounding rectangle of each property object contour in the current frame image.

[0066] Based on the extracted property object outlines, the contact relationship between each property object outline and the anchor point areas of each property space in the anchor point connection set is detected one by one, and the current action state is determined. Specifically, for each property object outline, the intersection-union ratio (IUR) between the property object outline and each property space anchor point area is calculated. The IUR calculation expression is:

[0067] ;

[0068] in, This represents the intersection-union ratio of the object outline and the anchor point region. This represents the pixel area covered by the object's outline. This indicates the anchor point area corresponding to the property's spatial anchor point. This represents the pixel area of ​​the overlapping portion between the object's outline and the anchor point region. This represents the total pixel area covered after the object outline and anchor point region are merged.

[0069] Simultaneously, the Euclidean distance between the center point of the outer rectangle of the property object's outline and the center point of the property's spatial anchor point area is calculated, expressed as:

[0070] ;

[0071] in, This represents the Euclidean distance between the center point of the bounding rectangle of the object's outline and the center point of the anchor region. This represents the pixel coordinates of the center point of the bounding rectangle of the object's outline. This represents the pixel coordinates of the center point of the anchor region.

[0072] In addition, the displacement change of the property object outline between different frames is calculated using the following expression:

[0073] ;

[0074] in, Indicates the first Frame relative to the first The amount of displacement change in the frame. Indicates the first The pixel coordinates of the center point of the bounding rectangle of the property object in the frame. Indicates the first The pixel coordinates of the center point of the bounding rectangle of the same property object in the frame.

[0075] S2.2. Based on the calculated intersection-over-union ratio, center point distance, and displacement change, determine whether the property object touches the property space anchor point area and what action triggering conditions are met. When the intersection-over-union ratio of the property object outline and the property space anchor point area is greater than the first threshold and the center point distance is less than the second threshold, it is determined that the property object touches the property space anchor point area. Further, based on the displacement change and contact duration at the time of touch, determine the action type. Action types include crossing, occupying, opening, leaving, and resetting. Among them, crossing refers to the property object outline moving from one side of the property space anchor point area to the other side; occupying refers to the property object outline staying in the property space anchor point area for more than the third threshold; opening refers to the angle between the door leaf outline and the door frame anchor point area changing; leaving refers to the property object outline appearing in the property space anchor point area and the property object no longer moving in subsequent frames; and resetting refers to the property object outline returning from the deviated state to the public facility reset reference anchor point area.

[0076] To further explain, the first threshold is the intersection-union ratio threshold, which is set to 0.1. This is because the property object and the anchor point area only need to make slight contact to be considered as touching. Therefore, using a smaller intersection-union ratio threshold of 0.1 can cover the situation where the edge of the object just enters the anchor point area, while avoiding misjudgment due to slight jitter in target detection.

[0077] The second threshold is the center point distance threshold, which is set to 50 pixels. This is because the anchor point area in the image is usually about 100×100 pixels in size, and the average distance from its center point to the boundary of the area is about 50 pixels. Therefore, setting the center point distance threshold to 50 pixels can ensure that the object outline is considered to have been touched when it is close enough to the anchor point area, while excluding targets that are far away.

[0078] The third threshold is the dwell time threshold, which is set at 5 seconds. According to property management experience, people or vehicles staying briefly in areas such as fire lanes and corridors (e.g., within 3 seconds) are usually not considered to occupy spaces, while staying for more than 5 seconds constitutes occupying space. Therefore, setting the dwell time threshold to 5 seconds can effectively distinguish between normal passage and illegal occupation.

[0079] S2.3. When it is determined that a property object touches the property space anchor point area and meets the action triggering conditions, a relation unit is generated. The relation unit contains a triplet of information: object type, anchor point identifier, and action type. The object type is obtained from the identified property object type, the anchor point identifier is obtained from the anchor point identifier corresponding to the touched property space anchor point in the anchor point connection set, and the action type is obtained from the determined action type. At the same time, the timestamp when the relation unit is generated is recorded.

[0080] For the same property object group with the same object identifier within the same camera, all relational units of the generated property object group are connected in chronological order according to timestamps to form a relational unit sequence of the property object group. Based on the relational unit sequence, the anchor point identifier corresponding to each relational unit is extracted, and the anchor point identifiers are arranged in chronological order according to the relational units to form the anchor point contact sequence identifier of the property object group in the current camera.

[0081] The continuous relational unit sequence and corresponding anchor contact sequence identifier of each property object group are encapsulated to form an event fragment of the property object group in the current camera. The event fragment includes the object identifier, relational unit sequence, anchor contact sequence identifier, start anchor identifier, end anchor identifier, and complete contact path in the camera of the property object group. The event fragments of all property object groups in all cameras are summarized to form an event fragment set.

[0082] S3. The event clips from adjacent cameras are spliced ​​together according to object continuation, action state continuation, and anchor point contact sequence to form a property event chain.

[0083] S3.1. Load the anchor point connection set and the event fragment set of all cameras, and extract the intersection lines and directional constraints between adjacent cameras. Specifically, read all cross-camera anchor point connection edges from the anchor point connection set. Each anchor point connection edge corresponds to an object intersection line. Obtain the starting anchor point identifier, ending anchor point identifier, connection direction, traversable type, and directional constraints of each object intersection line. At the same time, extract the event fragments generated by the previous camera and the event fragments generated by the next camera from the event fragment set as candidate objects to be spliced.

[0084] For the extracted previous and subsequent camera event segments, candidate segment pairs that meet the object continuation criteria are selected by comparing the contour similarity and motion direction consistency of the property object groups in the two event segments. Specifically, the comparison method is as follows: extract the contour feature vectors of the property object groups in the last consecutive frames of the previous event segment, extract the contour feature vectors of the property object groups in the first consecutive frames of the subsequent event segment, and calculate the cosine similarity between the two contour feature vectors. The expression is:

[0085] ;

[0086] in, This indicates the similarity in outline between a preceding event segment and a following event segment. The feature vector representing the outline of the property object group in the previous event segment. The feature vector representing the outline of the property object group in the subsequent event segment. This represents the dot product of two eigenvectors. and These represent the magnitudes of the two eigenvectors, respectively.

[0087] Meanwhile, by comparing the motion direction of the property object group in the last consecutive frame of the previous event segment with the motion direction of the property object group in the first consecutive frame of the subsequent event segment, if the angle between the motion directions is less than the preset motion direction consistency threshold, the motion directions are determined to be consistent; if the contour similarity is greater than the preset contour similarity threshold and the motion directions are consistent, the previous event segment and the subsequent event segment are marked as candidate segment pairs that meet the object continuation conditions.

[0088] To further explain, the outline similarity threshold is set at 0.85. While the appearance of the same property object may differ due to variations in lighting and angle under different camera views, the overall outline features still maintain a high degree of similarity. 0.85 effectively matches the same object while avoiding misidentification of different objects (such as different people or vehicles) as the same object. The motion direction consistency threshold is set at 45 degrees. The motion direction of a property object between adjacent cameras typically does not undergo drastic changes. Considering potential angular deviations during camera field-of-view switching, the 45-degree allowable range covers most normal movement and turning situations while excluding mismatches with significantly inconsistent directions.

[0089] S3.2. For each selected candidate fragment pair, parse the termination anchor point identifier and the last action state of the previous event fragment, and parse the start anchor point identifier and the first action state of the next event fragment. The specific parsing method is as follows: read the termination anchor point identifier field from the data structure of the previous event fragment to obtain the anchor point identifier corresponding to the last relation unit in the event fragment; at the same time, read the action type field of the last relation unit as the last action state; read the start anchor point identifier field from the data structure of the next event fragment to obtain the anchor point identifier corresponding to the first relation unit in the event fragment; at the same time, read the action type field of the first relation unit as the first action state.

[0090] The system checks if a crossover connection exists between the terminating and starting anchor points in the anchor point connection set, and verifies whether the last action state and the first action state meet the traversable type constraints configured for the crossover connection. Specifically, it searches the anchor point connection set for cross-camera anchor point connections that start from the terminating anchor point and end at the starting anchor point, or vice versa. If such a connection exists, it further reads the traversable type configured for the connection, determines whether the property object type corresponding to the last action state falls within the allowed traversable type range, and checks whether the transition between the last and first action states conforms to the directional constraints configured for the connection. When a connection exists and meets both the traversable type and directional constraints, the action state continuation is considered successful.

[0091] S3.3. Extract the continuous anchor point segments at the end of the anchor point contact sequence of the previous event segment and the continuous anchor point segments at the beginning of the anchor point contact sequence of the next event segment, and verify whether a continuous spatial path passing through the intersection line is formed in the anchor point topology map. The specific verification method is as follows: extract the last N consecutive anchor point identifiers from the anchor point contact sequence identifiers of the previous event segment to form the end segment of the sequence; extract the first M consecutive anchor point identifiers from the anchor point contact sequence identifiers of the next event segment to form the beginning segment of the sequence; where the values ​​of N and M are set according to the actual scenario, and are usually 2 to 3 anchor point identifiers.

[0092] The end segment of the sequence is concatenated with the beginning segment of the sequence to form a complete anchor point sequence segment. The anchor point topology maps of the previous and next cameras are loaded into the anchor point connection set. The concatenated anchor point sequence segment is mapped onto the anchor point topology map. It is checked whether there are spatially adjacent connecting edges between adjacent anchor points in the sequence, and it is verified whether the entire sequence segment transitions continuously from the last anchor point of the end segment to the first anchor point of the beginning segment through the found intersection connecting edge. When every pair of adjacent anchor points in the sequence has a connecting edge in the anchor point topology map, and the last anchor point of the end segment and the first anchor point of the beginning segment pass through the intersection connecting edge, the anchor point sequence is determined to be successfully connected.

[0093] When a selected candidate segment pair simultaneously meets both the action state continuation condition and the anchor point sequence beginning-end connection condition, the preceding event segment and the following event segment are merged into a single property event chain. The specific merging method is as follows: a new property event chain data structure is created; the relation unit sequences of the preceding and following event segments are concatenated in chronological order to form a merged relation unit sequence; the anchor point contact sequence identifiers of the preceding and following event segments are concatenated in chronological order to form a merged anchor point contact sequence identifier; the starting anchor point identifier of the preceding event segment is used as the starting anchor point identifier of the property event chain, and the ending anchor point identifier of the following event segment is used as the ending anchor point identifier of the property event chain; simultaneously, the camera identifier sequence and the intersection line identifier sequence traversed by the property event chain are retained; the merging operation is performed on all candidate segment pairs that meet the conditions to generate a complete property event chain composed of multiple event segments.

[0094] S4. Extract the starting point, turning point, and ending point of the event chain in the industry, and based on the extracted starting point, turning point, and ending point, complete the event characterization, locate the responsible party, and arrange the capture path to form a handling page.

[0095] S4.1. For each complete property event chain generated, traverse all relational units contained in the property event chain, and extract the event start-up features, event end-up positions, and event turning points. Specifically, the extraction method is as follows: read the relational unit with the earliest timestamp in the property event chain, extract the object type and anchor point identifier contained in the relational unit, and use them as the event start-up features; read the relational unit with the latest timestamp in the property event chain, and extract the anchor point identifier contained in the relational unit, and use it as the event end-up position.

[0096] Simultaneously, the system detects the node positions where action states change within the relational unit sequence of the property event chain. An action state change refers to a switch in the action type of adjacent relational units that conforms to a specific change pattern. The specific detection method is as follows: traverse each pair of adjacent relational units in the relational unit sequence. When the action type of the first relational unit is movement and the action type of the second relational unit is stationary, the position is determined as the turning point from movement to stationary; when the action type of the first relational unit is approaching and the action type of the second relational unit is contacting, the position is determined as the turning point from approaching to contacting; when the action type of the first relational unit is complete and the action type of the second relational unit is separation, the position is determined as the turning point from complete to separation.

[0097] S4.2. Based on the extracted event endpoint location and the sequence of action states that appear continuously in the property event chain, the event is characterized. The specific characterization method is as follows: read the anchor point identifier corresponding to the event endpoint location, and query the anchor point type corresponding to the anchor point identifier from the anchor point connection set. The anchor point types include unit door frame anchor point, unit door leaf anchor point, elevator threshold anchor point, car boundary anchor point, fire passage edge anchor point, garbage disposal outlet boundary anchor point, and corridor passable zone boundary anchor point, etc.

[0098] Simultaneously, all action types are extracted from the relational unit sequence of the property event chain and arranged chronologically to form an action state sequence. The endpoint anchor type is combined with the action state sequence to generate an event feature combination. This event feature combination is then matched with standard feature combinations in a pre-built violation type library. The violation type library pre-stores several standard feature combinations of violation events and their corresponding event type identifiers. For example, if the endpoint anchor type is a fire lane edge anchor point and the action state sequence includes "driving in, parking, personnel leaving the vehicle, and the passage being continuously occupied," it matches as a fire lane occupancy event type. If the endpoint anchor type is an elevator car boundary anchor point and the action state sequence includes "pushing closer, crossing the threshold, vehicle entering, and personnel following," it matches as an electric vehicle entering the elevator event type. If the endpoint anchor type is a unit door leaf anchor point or equipment room door handle anchor point and the action state sequence includes "opening, personnel leaving, and the door still in the open position," it matches as a door not in its correct position event type. When an event feature combination successfully matches a standard feature combination in the violation type library, the corresponding event type identifier is output as the event qualitative result.

[0099] S4.3. Based on the extracted event starting point features and the detected event turning points, complete the location of the responsible object. The specific location method is as follows: extract the object type and object outline corresponding to the first relational unit from the event starting point features, and mark the property object as the main responsible object; in the relational unit sequence of the property event chain, find the auxiliary object identifier that appears in association with the main responsible object. The auxiliary object refers to other property objects that appear at the same time as the main responsible object and have an interactive relationship in the relational unit at the same or adjacent timestamps.

[0100] When the primary responsible party and a certain auxiliary party are continuously bound in the contact path corresponding to the property event chain, that is, when the primary responsible party and the auxiliary party appear simultaneously at multiple consecutive anchor point positions in the anchor point contact sequence identifier and remain relatively stable, the auxiliary party is marked as a joint responsible party, and the object identifier corresponding to the joint responsible party in the anchor point connection set is extracted as the joint responsible party identifier.

[0101] Meanwhile, the first clear snapshot frame of the main responsible party in the event chain is extracted from the stored data of the property event chain as the initial image of the responsible party; the snapshot frames corresponding to each turning point of the event are extracted as the images confirming the responsible behavior. When there are multiple turning points, the snapshot frame corresponding to the latest turning point is selected as the final image confirming the responsible behavior.

[0102] S4.4. Based on the location of the responsible party, the qualitative event type, and the generated property event chain, complete the capture path arrangement and generate the handling page. The specific arrangement method is as follows: take the extracted initial image of the responsible party as the starting capture image, take the extracted image of the confirmed responsible behavior as the turning capture image, and extract the capture frame at the time of generation of the relationship unit from the relationship unit corresponding to the extracted event endpoint position as the endpoint capture image.

[0103] Read the event type identifier from the output event qualitative results; extract the last anchor point identifier as the endpoint anchor point identifier from the anchor point contact sequence identifiers of the generated property event chain. Load the camera layout diagram and crossover connection relationships associated with the generated anchor point connection set. The camera layout diagram includes the installation positions and field of view of all cameras, and the crossover connection relationships include the camera pairs and spatial positions connected to each object crossover line.

[0104] The anchor point sequence contained in the generated property event chain is mapped onto the camera layout map in chronological order, and the spatial locations corresponding to each anchor point are connected sequentially. The intersection lines of the objects traversed are marked to generate a complete cross-district path map. The starting point capture image, turning point capture image, ending point capture image, cross-district path map, event type identifier, ending point anchor point identifier, and extracted related liability object identifier are combined and encapsulated to generate a processing page containing complete event tracing information.

[0105] S5. Verify the reset status of the corresponding endpoint anchor point on the handling page, and continue video recognition and event continuation if the endpoint anchor point is not reset, to form a property handling record.

[0106] S5.1. Call the disposal page and read the endpoint anchor point identifier from the disposal page; based on the read endpoint anchor point identifier, load the pre-stored benchmark contour map corresponding to the endpoint anchor point identifier from the generated anchor point connection set. The pre-stored benchmark contour map is stored in the property space anchor point calibration stage and reflects the standard contour image when the property component associated with the anchor point is in the normal benchmark position.

[0107] Real-time video detection is performed on the anchor point area corresponding to the endpoint anchor point identifier on the processing page. The specific detection method is as follows: the current frame image is captured from the real-time video stream, the anchor point area corresponding to the endpoint anchor point identifier in the current frame image is located, the anchor point area is segmented and target recognition is performed, and the property component outline associated with the endpoint anchor point in the current frame is extracted. The property component outline includes the outline of door leaf, electric vehicle, garbage bag, stacked items, etc. At the same time, the outline coordinate set of the property component outline in the current frame is extracted.

[0108] S5.2. Calculate the contour overlap between the extracted current property component contour and the loaded pre-stored reference contour. The formula for calculating the contour overlap is:

[0109] ;

[0110] in, This indicates the degree of overlap between the current property component outline and the pre-stored baseline outline. This represents the pixel area covered by the extracted outline of the current property component. This represents the reference pixel area covered by the pre-stored reference contour map. This represents the pixel area of ​​the overlap between the current contour and the reference contour. Represents the total pixel area of ​​the baseline contour.

[0111] The calculated contour overlap is compared with a preset reset threshold. Different processing branches are executed based on the comparison results. The reset threshold is set to 0.9. When the property component is fully reset to the reference position, the contour should basically overlap with the reference contour. Considering the slight contour differences that may be caused by factors such as changes in lighting and shooting angle, the reset threshold is set to 0.9. That is, when the overlap area between the current contour and the reference contour reaches more than 90% of the area of ​​the reference contour, it can be determined that it has been reset. This ensures the accuracy of the reset judgment and avoids misjudgment caused by environmental factors.

[0112] When the overlap of the contours is greater than the reset threshold, it is determined that the endpoint anchor point has been reset. At this time, the current timestamp is recorded as the reset timestamp. The generated complete property event chain, the generated disposal page, and the reset timestamp are merged and archived to generate the final property disposal record.

[0113] When the contour overlap is less than or equal to the reset threshold, it is determined that the endpoint anchor point has not been reset. At this time, real-time video recognition continues to be performed on the anchor point area corresponding to the endpoint anchor point identifier. The specific recognition method is as follows: for each frame of the anchor point area in the real-time video stream, property object recognition, action state determination and relation unit generation are performed to generate new relation units and corresponding new anchor point contact sequences. The newly generated relation units are continued to the end of the relation unit sequence of the original property event chain in chronological order, and the new anchor point contact sequence is continued to the end of the anchor point contact sequence identifier of the original property event chain.

[0114] Simultaneously, the relevant attributes of the original property event chain are updated, specifically including: updating the duration of the original property event chain to the time length from the start timestamp to the current timestamp; updating the end anchor point identifier of the original property event chain to the anchor point identifier corresponding to the last newly generated relation unit; and updating the termination action status of the original property event chain to the action type contained in the last newly generated relation unit. The contour overlap of the current frame is repeatedly checked and compared with the reset threshold. New events are continuously added if the frame is not reset, until the contour overlap exceeds the reset threshold. Then, the operations of the reset branch are executed, generating the final property disposal record.

[0115] In summary, this invention elevates video recognition from generalized image analysis to a precise perception level coupled with physical space semantics by constructing a property space anchor point connection set, effectively solving the false alarm and missed alarm problems caused by ambiguous spatial relationships in traditional methods. By arranging the contact sequences between objects and anchor points, discrete single-frame detection is transformed into event fragments with spatial logic, providing a structured data foundation for event understanding. Through a triple continuation mechanism of object, action, and anchor point sequences, cross-camera event chain splicing is achieved, overcoming the event breakage problems caused by blind spots and perspective switching, ensuring the integrity and continuity from the start to the end of the event. In the event characterization phase, a comprehensive judgment is made based on the starting point, turning point, and ending point of the event chain, replacing the traditional single-frame or short-term judgment, which significantly improves the accuracy and interpretability of violation identification. At the same time, by locating the responsible party and generating cross-regional path maps, clear responsibility identification and tracing basis are provided for property management. Through continuous verification of the endpoint anchor point reset status and dynamic continuation in the case of non-reset, closed-loop management of event handling is realized, ensuring that violations are effectively corrected and retaining the complete event chain as a reference for subsequent processing, thereby improving the overall intelligence level and execution efficiency of smart community property management.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart community property management method based on image recognition technology, characterized in that: include, Mark the property space anchor points in the images of each camera, associate the object handover lines between adjacent cameras, configure the spatial connection relationship, and form an anchor point connection set; Anchor point connection sets are used to identify property objects in real-time video streams, determine the contact relationship and action status between property objects and anchor points in the property space, and arrange the anchor point contact sequence in chronological order to form an event fragment set; The event clips from adjacent cameras are spliced ​​together according to object continuation, action state continuation, and anchor point contact sequence to form a property event chain. Extract the starting point, turning point, and ending point of the event chain in the industry, and based on the extracted starting point, turning point, and ending point, complete the event characterization, locate the responsible party, and arrange the capture path to form a handling page; The system verifies the reset status of the corresponding endpoint anchor point on the processing page, and continues video recognition and event continuation if the endpoint anchor point is not reset, thus forming a property management processing record.

2. The smart community property management method based on image recognition technology as described in claim 1, characterized in that: The property space anchor points include unit door frame anchor points, door leaf anchor points, equipment room door handle anchor points, doorway boundary anchor points, elevator threshold anchor points, car boundary anchor points, fire escape edge anchor points, garbage disposal outlet boundary anchor points, corridor passable zone boundary anchor points, and public facility reset reference anchor points.

3. The smart community property management method based on image recognition technology as described in claim 1, characterized in that: The object handover line between adjacent cameras refers to marking a virtual handover line at the boundary of the overlapping field of view and blind zone of adjacent cameras, connecting the last visible anchor point of the previous camera with the first visible anchor point of the next camera, marking the crossing direction and assigning camera number pairs.

4. The smart community property management method based on image recognition technology as described in claim 1, characterized in that: The configuration space connection relationship is specifically as follows: Extract the coordinates of the property space anchor points within each camera and the connection direction of the intersection lines between adjacent cameras; Connect the property space anchor points within the same camera according to their spatial adjacency to form an anchor point topology map, and use the intersection line as the cross-camera anchor point connection edge, while configuring traversable type and direction constraints. Summarize the anchor point topology and cross-camera connection edges to generate an anchor point connection set.

5. The smart community property management method based on image recognition technology as described in claim 4, characterized in that: The formation of the event fragment set specifically refers to... Based on the anchor point connection set, the contours of objects in the real-time video stream are extracted, and the intersection-union ratio and positional changes of the contours and anchor point regions are detected. When the outline touches the anchor point area and the action triggering condition is met, a relational unit containing the object type, anchor point identifier, and action type is generated. Connect the continuously generated relational units of the same property object group within the same camera in chronological order, and extract the anchor point identifiers corresponding to each relational unit to form anchor point contact sequence identifiers; Each continuous relational unit and its corresponding anchor point contact sequence are encapsulated as an event fragment and summarized to form an event fragment set.

6. The smart community property management method based on image recognition technology as described in claim 5, characterized in that: The formation of the property event chain specifically refers to... Load the anchor point connection set and the event fragment set of the front and rear cameras, and extract the intersection line and direction constraints between adjacent cameras; Compare the object contour similarity and motion direction consistency between the previous event segment and the next event segment, filter candidate segment pairs for object continuation, and parse the termination anchor point and last action state of the previous event segment, and the starting anchor point and first action state of the next event segment. Query whether there is a junction line connecting the terminating anchor point and the starting anchor point in the anchor point connection set, and whether the last action state and the first action state meet the traversable type constraint. At the same time, extract the continuous anchor point segments at the end of the anchor point contact sequence of the previous event segment and the continuous anchor point segments at the beginning of the anchor point contact sequence of the next event segment, and verify whether a spatial continuous path passing through the junction line is formed in the anchor point topology graph. When the sequence of objects, actions, and anchor points is continuous, the event fragments before and after merging constitute the same property event chain.

7. The smart community property management method based on image recognition technology as described in claim 6, characterized in that: The completion event is characterized as follows: Traverse all relational units in the property event chain, extract the object type and anchor point identifier of the first relational unit as the event start point feature, and extract the anchor point identifier of the last relational unit as the event end point position; Detect nodes where the action state of the relationship unit changes abruptly, and extract nodes where the action changes from moving to stationary, from approaching to contacting, or from complete to separating as event turning points; Read the anchor point type corresponding to the event endpoint and the sequence of action states that appear consecutively in the event chain, combine and match them with a preset violation type library, and output the event qualitative result.

8. The smart community property management method based on image recognition technology as described in claim 7, characterized in that: The specific definition of the responsible party is as follows: Extract the object type and outline corresponding to the first relational unit from the event origin features, and mark it as the main responsible object; Extract the identifiers of auxiliary objects associated with the main responsible party in the property event chain. When the main responsible party and auxiliary objects are continuously bound in the contact path corresponding to the property event chain, they are marked as joint responsible parties and their identifiers are extracted. Extract the first clear snapshot frame of the main responsible party in the event chain as the initial image of the responsible party, and extract the snapshot frame of the main responsible party at the turning point as the image confirming the responsible behavior.

9. The smart community property management method based on image recognition technology as described in claim 8, characterized in that: The capture path arrangement is as follows: The starting image is the initial image of the responsible party, the turning image is the image confirming the responsible behavior, and the corresponding unit's image is extracted from the end position of the property event chain as the ending image. Read the event type identifier from the event qualitative results, and extract the endpoint anchor identifier from the anchor contact sequence identifier of the property event chain; Load the camera layout diagram and handover line connection relationship associated with the anchor point connection set, map the anchor point contact sequence identifier of the property event chain to the camera, and generate a cross-area path map; The starting point image, turning point image, ending point image, cross-regional route map, event type identifier, ending point anchor point identifier, and related liability object identifier are combined to generate a processing page.

10. The smart community property management method based on image recognition technology as described in claim 9, characterized in that: The formation of property disposal records specifically refers to... Call the endpoint anchor point identifier in the processing page and load the corresponding pre-stored baseline contour map; Real-time detection of the current frame image of the endpoint anchor point area, extraction of the property component outline associated with the endpoint anchor point, and calculation of the overlap between the current outline and the baseline outline map; When the overlap of the outline exceeds the preset reset threshold, it is determined that the endpoint anchor point has been reset, and the property event chain, the handling page and the reset timestamp are merged and archived as a property handling record. When the contour overlap does not exceed the reset threshold, it is determined that the endpoint anchor point has not been reset, and real-time video recognition of the current anchor point area continues to generate new relational units and new anchor point contact sequences. The newly generated relational unit is connected to the end of the original property event chain, and the new anchor contact sequence is connected to the anchor contact sequence of the original property event chain. The duration, end anchor identifier and termination action status of the property event chain are updated until the contour overlap exceeds the preset reset threshold, and then a property disposal record is generated.

Images