A park closed management method based on AI intelligent identification

Abstract

This invention discloses an AI-based intelligent recognition method for closed-loop management of a park, relating to the field of park management technology. The method includes: acquiring video frame sequences, area boundaries, access equipment status data, and access authorization data for a closed access area; identifying personnel and vehicle target boxes through the video frame sequences and generating target trajectory segments; obtaining access target detection data based on the target trajectory segments; generating authorized access windows based on the area boundaries, access equipment status data, and access authorization data, according to the authorized object's entry crossing time, access equipment release time, and authorized object's exit crossing time; calculating the target box overlap value of the access target detection data to obtain occlusion segment data. This invention maintains the stability of the cross-frame correspondence of targets when personnel and vehicles pass simultaneously and occlusion occurs, and establishes consistent constraints on time and spatial location, thereby improving the accurate identification of abnormal following behavior.

Description

Technical Field

[0001] This invention relates to the field of park management technology, specifically to a closed-loop management method for parks based on AI intelligent recognition. Background Technology

[0002] With the increasing demand for closed management in industrial parks, video surveillance can be used to identify and record the passage of people and vehicles. By combining area boundaries, the status of access equipment, and access authorization information to constrain passage behavior, dynamic monitoring and process management of the passage process can be achieved. At the same time, the passage behavior can be verified by the correspondence between time information and spatial location, providing basic support for the identification of abnormal behavior.

[0003] Currently, when people and vehicles pass through a closed passage area simultaneously, the differences in the position, scale, and appearance of different targets in consecutive video frames can easily lead to confusion in the correspondence between targets when multiple targets pass through in a short period of time. This results in the same target not being able to be stably corresponded between consecutive frames, which in turn makes the correlation between passage time and spatial location unstable, thus affecting the accuracy of matching passage behavior with authorization information.

[0004] Secondly, when people and vehicles overlap and cause occlusion, the target may disappear and reappear briefly in consecutive video frames. It is difficult to accurately maintain the correspondence between different targets before and after occlusion. Furthermore, within the same time range, there may be targets that do not correspond to the authorization information that enter and pass through, making it difficult to form a consistent constraint between the positional and temporal relationships during the passage process. This increases the chances of misjudgment and missed judgment in the identification of abnormal following passage behavior. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for closed-loop management of a park based on AI intelligent recognition.

[0006] A method for closed-loop management of a park based on AI intelligent recognition includes:

[0007] Acquire video frame sequences, area boundaries, access equipment status data, and access authorization data for closed access areas. Identify personnel and vehicle target boxes through the video frame sequences and generate target trajectory segments. Obtain access target detection data based on the target trajectory segments.

[0008] Based on the area boundary, access equipment status data, and access authorization data, an authorized access window is generated according to the time when the authorized object crosses the line at the entrance, the time period for the access equipment to release the passage, and the time when the authorized object crosses the line at the exit.

[0009] The target bounding box overlap value is calculated on the target detection data to obtain occlusion fragment data. Based on the occlusion fragment data and the authorized passage window, the identity continuation matching of the target trajectory fragment is performed to generate occlusion recovery passage data.

[0010] Based on the data of occlusion recovery and the authorized passage window, the system verifies the entrance position, time period, and exit position of personnel or vehicle target frames that are not bound to the authorized passage window within the authorized passage window, and generates an abnormal following passage judgment result.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0012] This invention constrains the continuous correspondence between personnel target boxes and vehicle target boxes by using traffic target detection data formed based on video frame sequences. This ensures that target trajectory segments maintain temporal continuity and spatial consistency even under conditions of mutual occlusion and staggered passage. As a result, a stable correspondence is formed between the authorized object's entry crossing time, the authorized object's exit crossing time, and the verification and release period, reducing target identity confusion and improving the matching accuracy between the authorized passage window and the actual passage behavior.

[0013] Furthermore, this invention also associates occluded fragment data with authorized access windows and combines occlusion recovery access data to continuously verify the target boxes of people or vehicles that are not bound to authorized access windows. This ensures that the data occupied by unbound target windows remains consistent in terms of time and location. Based on this, the entrance location, travel time, and exit location are jointly determined to form a consistent abnormal following access determination result, reducing the impact of mismatches in the occlusion recovery process and thus improving the accuracy of abnormal following access behavior identification.

[0014] In summary, this invention improves the accurate identification of abnormal following behavior by maintaining the stability of the target cross-frame correspondence when people and vehicles pass simultaneously and obstruction occurs, and by forming consistent constraints on time and spatial location. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 The flowchart illustrates a closed-loop management method for a park based on AI intelligent recognition, as provided by this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, not all embodiments. 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.

[0018] Please see Figure 1 As shown in the figure, this embodiment discloses a method for closed-loop management of a park based on AI intelligent recognition. The method includes:

[0019] S11: Obtain video frame sequences, area boundaries, access equipment status data, and access authorization data for the closed access area; identify personnel and vehicle target boxes through the video frame sequences and generate target trajectory segments; obtain access target detection data based on the target trajectory segments.

[0020] In one specific embodiment, the video frame sequence of the closed passage area, the area boundary, the status data of the passage equipment, and the passage authorization data are used as input data. The video frame sequence includes video frames arranged according to the acquisition time, frame number, and frame time. The area boundary includes the entrance side boundary, the exit side boundary, and the channel side boundary. The passage equipment status data includes the equipment number, equipment status, status start time, and status end time. The passage authorization data includes the authorized object identifier, authorized object category, authorization start time, authorization end time, and authorized passage direction.

[0021] The video frame sequence is acquired by existing image acquisition equipment set in the closed passage area. The area boundary is manually calibrated by the monitoring screen or converted by combining the park passage plan with the camera calibration parameters. The status data of the passage equipment is obtained from the existing control records of the access control, barrier gate, turnstile or retractable gate. The passage authorization data is obtained from the park access control system, visitor reservation system or vehicle reservation system.

[0022] Specifically, the steps for generating transit target detection data are as follows:

[0023] S111: Obtain the image data corresponding to each video frame in the video frame sequence, identify the person target box and vehicle target box in the image data, and record the frame time, target box center point and appearance feature vector corresponding to the person target box and vehicle target box.

[0024] In one specific embodiment, each video frame in the video frame sequence is read in the order of frame time, each video frame is converted into image data, and an existing object detection network is used to identify the person target bounding box and vehicle target bounding box in the image data. The person target bounding box is used to represent the rectangular detection area where the person target is located in the video frame, and the vehicle target bounding box is used to represent the rectangular detection area where the vehicle target is located in the video frame.

[0025] Both personnel and vehicle bounding boxes are represented using the video frame coordinate system. Each bounding box includes the coordinates of its top-left and bottom-right corners. The bounding box is represented as follows:

[0026]

[0027] in, For frame time The Middle One target box, and The coordinates of the top left corner of the target box. and The coordinates of the bottom right corner of the target box. The horizontal pixel coordinates of the image. These are the vertical pixel coordinates of the image.

[0028] The center point of the target box is calculated based on the coordinates of the top left and bottom right corners of the personnel and vehicle target boxes. The center point of the target box is represented as:

[0029]

[0030] in, For frame time The Middle The center point of the target box. The x-coordinate of the center point of the target box. The ordinate of the center point of the target box. , , , For frame time The Middle The coordinates of the four boundaries of the target bounding box.

[0031] The image regions corresponding to the bounding boxes for personnel and vehicles are cropped, and appearance feature vectors are extracted using an existing target re-identification feature extraction network. The appearance feature vectors are represented as follows:

[0032]

[0033] in, For frame time The Middle The appearance feature vector corresponding to each target bounding box The dimension of the appearance feature vector. , , , These are the feature values ​​of each dimension in the appearance feature vector.

[0034] It should be noted that the appearance feature vector is output by the existing target re-identification feature extraction network. The target re-identification feature extraction network can be trained based on historical monitoring image samples of the park, or it can adopt the existing pedestrian re-identification network, vehicle re-identification network or shared feature extraction network.

[0035] The personnel target bounding box, vehicle target bounding box, frame time, target bounding box center point, and appearance feature vector are written into the corresponding single-frame target record. The single-frame target record serves as the data source for calculating the target bounding box continuation parameters in step S112.

[0036] S112: Calculate the intersection and union value, center point displacement value, center point movement direction value, and feature vector similarity of target boxes of the same category in adjacent video frames to obtain the target box continuation parameters;

[0037] In a specific embodiment, target boxes within adjacent video frames are read in the order of frame time. When there are dropped frames in the video frame sequence, adjacent valid video frames are read in the order of frame time, and only the intersection and union value, center point displacement value, center point movement direction value, and feature vector similarity are calculated for target boxes of the same category.

[0038] Let the frame time corresponding to the current video frame be . The frame time corresponding to the next adjacent valid video frame is , frame time target box in With frame time Same category target boxes The intersection and union values ​​are expressed as:

[0039]

[0040] in, For the target box With target box Intersection and union values ​​between them This represents the area of ​​the overlapping region between the two bounding boxes. This represents the area of ​​the merged region of the two target boxes.

[0041] The displacement value of the center point is calculated based on the center point of the target box. The displacement value of the center point is expressed as:

[0042]

[0043] in, The displacement value of the center point. For the target box The center point of the target bounding box. For the target box The center point of the target bounding box.

[0044] The movement direction value of the center point is calculated based on the lateral and longitudinal displacement values ​​of the center point of the target box. The movement direction value of the center point is expressed as follows:

[0045]

[0046] in, The value represents the direction of movement of the center point. It is the arctangent function in the fourth quadrant. This represents the lateral displacement value of the center point of the target bounding box. This represents the longitudinal displacement value of the center point of the target bounding box.

[0047] The feature vector similarity is calculated based on the appearance feature vectors corresponding to two bounding boxes of the same category. The feature vector similarity is expressed as:

[0048]

[0049] in, For feature vector similarity, For the target box The corresponding appearance feature vector, For the target box The corresponding appearance feature vector, Represents the vector norm.

[0050] The intersection and union values, center point displacement values, center point movement direction values, and feature vector similarity are written into the same continuation record to obtain the target box continuation parameters, which are expressed as follows:

[0051]

[0052] in, For the target box With target box The parameters connecting the target boxes between them For intersection and union values, The displacement value of the center point. The value represents the direction of movement of the center point. This represents the similarity of feature vectors.

[0053] It should be noted that: the target box continuation parameters for personnel target boxes are only calculated with personnel target boxes, and the target box continuation parameters for vehicle target boxes are only calculated with vehicle target boxes. The target box continuation parameters for personnel target boxes and vehicle target boxes used for continuation of the same target are not calculated.

[0054] S113: Based on the target box continuation parameters, continuate the target boxes of the same person or the same vehicle to generate target trajectory segments;

[0055] In one specific embodiment, the target box continuation parameter is used to determine whether target boxes of the same category in adjacent video frames belong to the same target, and the target trajectory segment is data formed by continuating the same person target box or the same vehicle target box in the order of frame time.

[0056] Specifically, the steps for generating the target trajectory segment are as follows:

[0057] S113.1: Read the target bounding box continuation parameters in the order of frame time, and filter the target bounding boxes of people or vehicles with the same target category in adjacent video frames;

[0058] In one specific embodiment, the target bounding box continuation parameters are read in frame time sequence. And read the target box of the previous frame corresponding to the target box continuation parameters. and the target box of the next adjacent valid frame When the target box With target box When the target categories are the same, the target box will be... As the target box Candidate continuation boxes are not considered as candidate continuation boxes for the same target if the target categories are different.

[0059] Candidate continuation pairs are represented as:

[0060]

[0061] in, For candidate successor pairs, The target bounding box of the previous frame. For the next adjacent valid frame of the same category, Parameters for the target bounding box.

[0062] S113.2: Based on the intersection and union values, center point displacement values, center point movement direction values, and feature vector similarity of the same type of target boxes, determine the contiguous boxes belonging to the same target in adjacent video frames;

[0063] In one specific embodiment, read candidate continuation pairs intersection and union values ​​in Displacement value of center point , center point movement direction value Similarity with feature vectors The continuation frame is determined according to the continuation judgment rules, which include the intersection and merger values ​​satisfying the intersection and merger conditions, the center point displacement values ​​satisfying the displacement conditions, the center point movement direction values ​​satisfying the direction conditions, and the feature vector similarity satisfying the similarity conditions. The parameters corresponding to the intersection and merger conditions, displacement conditions, direction conditions, and similarity conditions are determined based on the installation height of the park cameras, video frame rate, channel width, target passage speed, and historical passage sample experimental data.

[0064] The continuation cost is calculated based on the target bounding box continuation parameters. The continuation cost is expressed as:

[0065]

[0066] in, For the candidate to continue The value of successor generations, The reference displacement value is in pixels. The data was determined based on video frame rate, camera calibration parameters, and experimental data on target passage speed within the closed passage area. This represents the direction of movement of the current target trajectory segment within the previous continuation segment. It is calculated from the center points of the two adjacent connected target boxes in the current target trajectory segment. , , , These are dimensionless weighting coefficients. , , , Determined experimentally based on correct and incorrect continuation samples from historical common samples. Pi is the mathematical constant of a circle.

[0067] When a candidate continuation pair meets the continuation determination rules, the next adjacent valid frame of the same category in the candidate continuation pair is determined as the continuation frame. If the same previous frame target frame corresponds to multiple candidate continuation frames, they are sorted in ascending order of continuation generation value, and the candidate continuation frame with the highest continuation generation value is determined as the continuation frame. If the same next adjacent valid frame target frame is selected by multiple previous frame target frames, the continuation result with the highest continuation generation value is retained.

[0068] S113.3: Connect the continuation frames corresponding to the same target in the order of frame time to generate target trajectory segments;

[0069] In one specific embodiment, the continuation frame is written into the trajectory record of the corresponding target in the order of frame time. For the first appearance of a personnel target frame or vehicle target frame, a new target trajectory segment is established. For personnel target frames or vehicle target frames that already have target trajectory segments, the continuation frame is appended to the end of the existing target trajectory segment.

[0070] The target trajectory segment is represented as:

[0071]

[0072] in, For the first A target trajectory segment, These are the frame times arranged in chronological order. For the first The target at frame time The corresponding target box, For the first The target at frame time The corresponding center point of the target box, For the first The target at frame time The corresponding appearance feature vector, For the first The target category of each objective.

[0073] When the target trajectory segment does not obtain a continuation frame in consecutive video frames, the target trajectory segment is recorded as a temporary interruption state, and the end target frame, end frame time, and end appearance feature vector of the target trajectory segment are retained.

[0074] S114: Write the personnel target bounding box, vehicle target bounding box, frame time, target bounding box center point, appearance feature vector, and target trajectory fragment into the same target record to generate traffic target detection data.

[0075] In one specific embodiment, the person target bounding box, vehicle target bounding box, frame time, target bounding box center point, appearance feature vector, and target trajectory fragment are written into the same target record to form traffic target detection data. The traffic target detection data is stored in a table structure, key-value structure, or database record structure.

[0076] The target record in the passable target detection data is represented as:

[0077]

[0078] in, For the first One target record, Number the target record. For the target category, For the target trajectory segment, For the first Each frame, For the first The target at frame time The corresponding target box, For the first The target at frame time The corresponding center point of the target box, For the first The target at frame time The corresponding appearance feature vector, For the first The target at frame time The corresponding trajectory state, The target record contains the number of frame times.

[0079] S12: Based on the area boundary, access equipment status data, and access authorization data, generate an authorized access window according to the entry crossing time of the authorized object, the access equipment release time period, and the exit crossing time of the authorized object;

[0080] In a specific embodiment, the area boundary, access device status data, and access authorization data are used as input data for this step. The area boundary is represented using the video frame coordinate system. The access device status data includes device number, device status, status start time, and status end time. The access authorization data includes authorized object identifier, authorized object category, authorization start time, authorization end time, and authorized passage direction.

[0081] The area boundary is obtained in step S11, the access equipment status data is obtained from the existing control records of the access control, barrier gate, turnstile or retractable gate, and the access authorization data is obtained from the park access control system, visitor reservation system or vehicle reservation system.

[0082] Specifically, the steps to generate the authorization pass window are as follows:

[0083] S121: Determine the authorized object based on the passage authorization data, and extract the personnel target box or vehicle target box corresponding to the authorized object from the passage target detection data;

[0084] In one specific embodiment, the authorized object identifier, authorized object category, authorization start time, authorization end time, and authorized passage direction in the passage authorization data are read, and the personnel or vehicle targets that have the authorized object identifier and are between the authorization start time and the authorization end time are identified as authorized objects.

[0085] When the authorized object category is personnel, the target record corresponding to the personnel target box is extracted from the traffic target detection data. When the authorized object category is vehicle, the target record corresponding to the vehicle target box is extracted from the traffic target detection data.

[0086] The personnel or vehicle target boxes corresponding to the authorized objects are bound to the authorized object identifier and the target record in the access target detection data. The personnel target box can be determined based on the frame time and entrance area corresponding to the access control card swipe record, face recognition record, or visitor reservation record. The vehicle target box can be determined based on the frame time and entrance area corresponding to the license plate recognition record, vehicle reservation record, or vehicle pass record.

[0087] It should be noted that the traffic target detection data is generated in step S11. The traffic target detection data includes personnel target boxes, vehicle target boxes, frame time, target box center point, appearance feature vector, and target trajectory fragment. The personnel target boxes or vehicle target boxes corresponding to the authorized objects are extracted from the traffic target detection data.

[0088] S122: Based on the center point of the target box corresponding to the region boundary and the authorized object, obtain the crossing time data, which includes the crossing time of the authorized object entrance and the crossing time of the authorized object exit;

[0089] In a specific embodiment, the entrance boundary line and the exit boundary line are determined based on the region boundary, and the center point of the target box corresponding to the authorized object is read. The intersection frame time of the center point of the target box corresponding to the authorized object with the entrance boundary line and the exit boundary line is calculated according to the frame time sequence.

[0090] Specifically, the steps to obtain the data on the moment of crossing the line are as follows:

[0091] S122.1: Determine the entrance and exit boundary lines based on the region boundary, and extract the target trajectory segments corresponding to the authorized objects;

[0092] In a specific embodiment, the entrance side boundary and the exit side boundary in the area boundary are read, and the entrance boundary line and the exit boundary line are determined according to the authorized passage direction. The side boundary on which the authorized object enters the closed passage area is determined as the entrance boundary line, and the side boundary on which the authorized object leaves the closed passage area is determined as the exit boundary line. The target trajectory segment corresponding to the authorized object is extracted from the passage target detection data.

[0093] The entrance boundary line is represented as:

[0094]

[0095] in, This is the entrance boundary line. The first endpoint of the entrance boundary line. It is the second endpoint of the entrance boundary line. and The horizontal pixel coordinates of the image. and These are the vertical pixel coordinates of the image.

[0096] The export boundary line is represented as:

[0097]

[0098] in, For the export boundary line, The first endpoint of the export boundary line. It is the second endpoint of the export boundary line. and The horizontal pixel coordinates of the image. and These are the vertical pixel coordinates of the image.

[0099] The target trajectory segment corresponding to the authorized object is represented as follows:

[0100]

[0101] in, For the first The target trajectory segment corresponding to each authorized object These are the frame times arranged in chronological order. For the first An authorized object at frame time The corresponding target box, For the first An authorized object at frame time The corresponding center point of the target box, For the first An authorized object at frame time The corresponding appearance feature vector, For the first The authorized object category of an authorized object.

[0102] It should be noted that both the entrance and exit boundary lines are obtained from the area boundary, and the target trajectory segment corresponding to the authorized object is extracted from the passing target detection data generated in step S11.

[0103] S122.2: Read the center point of the target box corresponding to the authorized object in the order of frame time, and calculate the intersection frame time of the center point of the target box with the entrance boundary line and the exit boundary line respectively;

[0104] In a specific embodiment, the center point of the target box in the target trajectory segment corresponding to the authorized object is read in the order of frame time, the center points of the target boxes corresponding to two adjacent frame times are connected to form a center point movement line segment, and it is determined whether the center point movement line segment intersects with the entrance boundary line and the exit boundary line.

[0105] The line segment representing the movement of the center point is represented as:

[0106]

[0107] in, For the first An authorized object at frame time to frame time The center point of the line segment between them is moved. For the first An authorized object at frame time The corresponding center point of the target box, For the first An authorized object at frame time The center point of the corresponding target bounding box.

[0108] For the line segment that moves from the center point and boundary segments The intersection of two line segments can be determined by the cross product of two-dimensional vectors. The condition for line segment intersection is expressed as:

[0109]

[0110]

[0111] in, and Move the two endpoints of the line segment to the center point. and These are the two endpoints of the inlet or outlet boundary line. It is the cross product function of two-dimensional vectors.

[0112] The cross product function of two-dimensional vectors is expressed as:

[0113]

[0114] in, and It is a two-dimensional vector. and For the horizontal coordinate component of the image, and This represents the vertical coordinate component of the image.

[0115] When the moving line segment of the center point intersects with the entrance boundary line, the next frame time of the moving line segment of the center point is recorded as the intersection frame time of the target box center point and the entrance boundary line. When the moving line segment of the center point intersects with the exit boundary line, the next frame time of the moving line segment of the center point is recorded as the intersection frame time of the target box center point and the exit boundary line.

[0116] S122.3: Record the intersection frame time of the target box center point and the entrance boundary line as the entrance crossing time of the authorized object, and record the intersection frame time of the target box center point and the exit boundary line as the exit crossing time of the authorized object, thus obtaining the crossing time data;

[0117] The data for the moment of crossing the line is represented as follows:

[0118]

[0119] in, For the first Data on the time of crossing the line for each authorized object. For the first When an authorized object's entry point crosses the line, For the first The moment when an authorized object crosses the exit line.

[0120] It should be noted that the boundary crossing time data includes the boundary crossing time of the authorized object's entry and the boundary crossing time of the authorized object's exit. The boundary crossing time of the authorized object's entry and the boundary crossing time of the authorized object are obtained from the intersection frame time of the center point of the target box corresponding to the authorized object with the entry boundary line and the exit boundary line.

[0121] S123: Extract the passage release period of the passage equipment based on the status data of the passage equipment, and perform cross-validation between the passage release period and the crossing time data to obtain the validated passage release period;

[0122] In one specific embodiment, the device number, device status, status start time, and status end time in the access device status data are read, and the time period when the device status is in the open state or the open state is extracted as the access device release period.

[0123] The passage period for the passage equipment is indicated as follows:

[0124]

[0125] in, The designated time period for allowing passage equipment to operate. This is the starting time when the passage equipment enters the passage release state. This is the end time when the passage equipment ceases its release state.

[0126] The data on the time period for allowing passage through the equipment and the time of crossing the line are cross-validated. The validated time period is represented as follows:

[0127]

[0128] in, To verify the release period, This is the starting time when the passage equipment enters the passage release state. This is the end time when the passage equipment ceases its release state. When the entry point of an authorized object crosses the line, The moment when the authorized object crosses the exit line.

[0129] when When there is a time overlap between the passage release period and the crossing time data of the passage equipment, the verification release period is output. If there is no time overlap between the passage release time and the crossing time data of the passage device, the corresponding authorized object will be recorded as a window verification exception object.

[0130] It should be noted that: the passage period is obtained by extracting the passage status data, the crossing time data is obtained by step S122, and the verification passage period is jointly limited by the passage period and the crossing time data.

[0131] S124: Generate an authorized passage window based on the authorized object's entry crossing time, verification and release time period, and authorized object's exit crossing time;

[0132] In a specific embodiment, the entry crossing time of the authorized object, the verification and release period, and the exit crossing time of the authorized object are read. The start time of the verification and release period is used as the start time of the authorized passage window, and the end time of the verification and release period is used as the end time of the authorized passage window. The authorized object identifier, authorized object category, and authorized passage direction are written into the same window record to generate the authorized passage window.

[0133] The authorized access window is displayed as follows:

[0134]

[0135] in, For the first The authorization access window corresponding to each authorized object The start time of the authorization access window. This is the end time of the authorization window. For the authorized object identifier, For the authorized object category, For authorized traffic direction.

[0136] The start and end times of the authorized access window are represented as follows:

[0137]

[0138]

[0139] in, The start time of the authorization access window. This is the end time of the authorization window. This is the starting time when the passage equipment enters the passage release state. This is the end time when the passage equipment ceases its release state. When the entry point of an authorized object crosses the line, The moment when the authorized object crosses the exit line.

[0140] It should be noted that the authorized passage window is generated jointly by the time the authorized object crosses the line at the entrance, the verification and release period, and the time the authorized object crosses the line at the exit. The authorized passage window serves as the data source for subsequent occlusion recovery passage data generation and abnormal follow-through passage determination.

[0141] S13: Calculate the target bounding box overlap value of the passing target detection data to obtain occluded segment data; based on the occluded segment data and the authorized passage window, perform identity continuation matching on the target trajectory segment to generate occlusion recovery passage data;

[0142] Specifically, the steps for generating data to restore traffic flow after occlusion are as follows:

[0143] S131: Extract the bounding boxes of people and vehicles at the same time from the traffic target detection data, and calculate the overlap value between the bounding boxes of people and vehicles.

[0144] In one specific embodiment, the bounding boxes of people and vehicles are read from the traffic target detection data according to the frame time, the bounding boxes of people and vehicles that exist at the same frame time are filtered, and the overlap value between the bounding boxes of people and vehicles is calculated.

[0145] The personnel target box is represented as:

[0146]

[0147] in, For frame time Next Individual target boxes, and The coordinates of the top left corner of the personnel target box. and The coordinates of the bottom right corner of the personnel target box. The horizontal pixel coordinates of the image. These are the vertical pixel coordinates of the image.

[0148] The vehicle target box is represented as:

[0149]

[0150] in, For frame time Next One vehicle target box, and The coordinates of the top left corner of the vehicle target box. and The coordinates are the bottom right corner of the vehicle's target bounding box.

[0151] The target bounding box overlap value is expressed as:

[0152]

[0153] in, For frame time Next Individual target box and the first The overlap value of the bounding boxes between vehicle bounding boxes. This represents the area of ​​the overlapping region between the personnel target bounding box and the vehicle target bounding box. The target area for personnel. Let this be the area of ​​the vehicle's target frame. This is a function that takes the minimum value.

[0154] It should be noted that the target box overlap value is used to characterize the degree of spatial overlap between the person target box and the vehicle target box at the same frame time. The parameters corresponding to the determination of the target box overlap value are determined based on the camera installation height in the closed passage area, the video resolution, and the experimental data of historical occlusion samples.

[0155] S132: Obtain occlusion fragment data based on the target bounding box overlap value and the missing frame time of the target trajectory fragment;

[0156] In one specific embodiment, the overlap value of the target bounding box and the missing frame time of the target trajectory segment are read, and the frame time corresponding to the overlap value of the target bounding box is matched with the missing trajectory time period. When the frame time corresponding to the overlap value of the target bounding box falls into the missing trajectory time period, occlusion segment data is generated.

[0157] The occluded segment data is represented as follows:

[0158]

[0159] in, To obscure fragment data, For the first The record is obscured. The occlusion record number.

[0160] Specifically, the steps to obtain occluded segment data are as follows:

[0161] S132.1: Arrange the target box overlap values ​​in the order of frame time, and extract the personnel target boxes, vehicle target boxes and corresponding frame times corresponding to the target box overlap values;

[0162] In a specific embodiment, the target box overlap values ​​obtained in step S131 are arranged in the order of frame time, and the personnel target box, vehicle target box and corresponding frame time corresponding to each target box overlap value are read to obtain the target box overlap value sorting record.

[0163] The target bounding box overlap value sorting record is represented as follows:

[0164]

[0165] in, Sort the records by target box overlap value. These are the frame times arranged in chronological order. For frame time Next Individual target boxes, For frame time Next One vehicle target box, For frame time The corresponding target bounding box overlap value.

[0166] It should be noted that the target box overlap values ​​in the target box overlap value sorting record are derived from step S131, and the personnel target boxes, vehicle target boxes and corresponding frame times corresponding to the target box overlap values ​​are all derived from the traffic target detection data.

[0167] S132.2: Based on the missing frame times and recurring frame times of the person or vehicle target bounding boxes in consecutive video frames, obtain the trajectory missing time periods;

[0168] In one specific embodiment, the target trajectory segment corresponding to the person target box or vehicle target box is read, the continuity of the same target in consecutive video frames is determined according to the frame time sequence, and the trajectory missing time period is obtained based on the missing frame time and the recurring frame time.

[0169] The missing time periods in the trajectory are represented as follows:

[0170]

[0171] in, For the first The missing time period corresponding to each target trajectory segment For the frame time corresponding to the missing start frame, To reproduce the frame time corresponding to the starting frame.

[0172] Specifically, the steps to obtain the missing time periods of the trajectory are as follows:

[0173] S132.2.1: Read the target trajectory segments of the same target in the order of frame time, and extract the frame number interval value between adjacent continuation frames in the target trajectory segments;

[0174] In one specific embodiment, target trajectory segments of the same target are read in the order of frame time, and the frame numbers corresponding to adjacent continuation frames in the target trajectory segments are read. The frame number interval value is calculated based on the frame numbers corresponding to adjacent continuation frames.

[0175] The frame number interval value is represented as:

[0176]

[0177] in, For the first In the target trajectory segment, the first The frame number interval value between adjacent continuation frames. For frame time The corresponding frame number, For frame time The corresponding frame number.

[0178] It should be noted that the frame number comes from the video frame sequence, which is obtained in step S11. The frame number interval is used to determine whether there are missing frames between adjacent connecting frames in the target trajectory segment.

[0179] S132.2.2: Determine the missing start frame and the reappearance start frame of the person or vehicle target frame in consecutive video frames based on the frame number interval value;

[0180] In one specific embodiment, the frame number interval value is compared with the continuous continuation condition. When the frame number interval value does not meet the continuous continuation condition, the first frame after the frame number corresponding to the previous continuation frame is determined as the missing start frame, and the frame number corresponding to the next continuation frame is determined as the reproduction start frame.

[0181] The missing start frame number is represented as:

[0182]

[0183] The starting frame number for reproduction is represented as:

[0184]

[0185] in, For the first The missing start frame number corresponding to each target trajectory segment. For the first The starting frame number corresponding to the re-enactment of each target trajectory segment. This is the frame number corresponding to the previous continuation frame. This is the frame number corresponding to the next continuation frame.

[0186] It should be noted that the parameters corresponding to the continuous continuation condition are determined based on the video frame rate, target detection period and historical pass sample experimental data. When there are lost frame records in the video frame sequence, the frame number corresponding to the lost frame record is excluded before the missing start frame and the re-enactment start frame are determined.

[0187] S132.2.3: Pair the frame time corresponding to the missing starting frame with the frame time corresponding to the recurring starting frame to obtain the trajectory missing time period.

[0188] In one specific embodiment, the missing start frame number is mapped to the frame time corresponding to the missing start frame, the recurrence start frame number is mapped to the frame time corresponding to the recurrence start frame, and the frame time corresponding to the missing start frame and the frame time corresponding to the recurrence start frame are paired to obtain the trajectory missing time period.

[0189] The missing time periods in the trajectory are represented as follows:

[0190]

[0191] in, For the first The missing time period corresponding to each target trajectory segment For the frame time corresponding to the missing start frame, To reproduce the frame time corresponding to the starting frame.

[0192] S132.3: Write the missing target identifier, missing target category, personnel target box, vehicle target box, target box overlap value, corresponding frame time and trajectory missing time period of the frame time corresponding to the target box overlap value into the same occlusion record to obtain occlusion segment data;

[0193] In a specific embodiment, the target box overlap value sorting record and trajectory missing time period are read, it is determined whether the frame time corresponding to the target box overlap value falls within the trajectory missing time period, and the missing target identifier and missing target category corresponding to the trajectory missing time period are read. The missing target identifier, missing target category, personnel target box, vehicle target box, target box overlap value, corresponding frame time and trajectory missing time period are written into the same occlusion record to obtain occlusion segment data.

[0194] The occlusion record is represented as:

[0195]

[0196] in, For the first The record is obscured. To obscure the record number, This refers to the missing target identifiers corresponding to the time periods when the trajectory is missing. For the missing target categories corresponding to the time periods of the missing trajectory, The target bounding boxes are the personnel target bounding boxes corresponding to the target bounding box overlap value. The target bounding box is the vehicle target bounding box corresponding to the target bounding box overlap value. This is the overlap value of the target bounding box. This refers to the frame time corresponding to when the overlap value of the target bounding box falls within the time period of trajectory loss. This refers to the period when the trajectory is missing.

[0197] It should be noted that the missing target identifier and missing target category are determined by the target trajectory segment corresponding to the missing trajectory period. When the missing trajectory period is generated by the target trajectory segment corresponding to the personnel target box, the missing target category is the personnel category. When the missing trajectory period is generated by the target trajectory segment corresponding to the vehicle target box, the missing target category is the vehicle category.

[0198] S133: Based on the occlusion fragment data and the authorized passage window, filter the target trajectory fragment before the interruption and the target trajectory fragment after the recurrence;

[0199] In one specific embodiment, the missing target identifier, missing target category, and trajectory missing time period in the occlusion fragment data are read, and the start time and end time of the authorized passage window in the authorized passage window are read. Within the frame time range corresponding to the authorized passage window, the target trajectory fragment before the interruption before the trajectory missing time period is filtered based on the missing target identifier, and the target trajectory fragment after the trajectory missing time period and with the same target category is filtered based on the missing target category.

[0200] The target trajectory segment before the interruption is represented as follows:

[0201]

[0202] in, Missing target identifier The corresponding target trajectory segment before the interruption, This refers to the last frame moment before the period of missing trajectory. This is the target bounding box corresponding to the last frame. The center point of the target bounding box at the last frame time. This is the appearance feature vector corresponding to the last frame time. For the target category.

[0203] The reproduced target trajectory segment is represented as follows:

[0204]

[0205] in, For target category The corresponding candidate re-constructed target trajectory segment, This refers to the start frame time after the period of missing trajectory. This is the target bounding box corresponding to the starting frame. The center point of the target bounding box at the start frame time. This is the appearance feature vector corresponding to the starting frame time. For the target category.

[0206] S134: Based on the target trajectory segment before the interruption and the target trajectory segment after the recurrence, complete the identity continuation matching and generate occlusion recovery data.

[0207] In one specific embodiment, the center point of the end target box, the end frame time, and the end appearance feature vector of the target trajectory segment before the interruption are read, and the center point of the start target box, the start frame time, and the start appearance feature vector of the target trajectory segment after the recurrence are read. Identity continuation matching is completed based on the restored displacement value, the restored time interval, and the restored similarity.

[0208] The recovered displacement value is expressed as:

[0209]

[0210] in, For the first The target trajectory segment before the interruption and the first The recovered displacement values ​​between the target trajectory segments after reproduction The center point of the target bounding box at the end of the target trajectory segment before the interruption. The center point of the starting target bounding box for the reproduced target trajectory segment.

[0211] The recovery time interval is expressed as:

[0212]

[0213] in, For the first The target trajectory segment before the interruption and the first The recovery time interval between each reproduced target trajectory segment This refers to the last frame time of the target trajectory segment before the interruption. This is the starting frame time of the reconstructed target trajectory segment.

[0214] The restored similarity is represented as:

[0215]

[0216] in, For the first The target trajectory segment before the interruption and the first The recovery similarity between the reproduced target trajectory segments This is the final appearance feature vector of the target trajectory segment before the interruption. This represents the initial appearance feature vector of the reconstructed target trajectory segment. Represents the vector norm.

[0217] The value of identity continuation is represented as:

[0218]

[0219] in, To perpetuate the value of identity. To restore the reference displacement value in pixels, The data was determined based on video frame rate, camera calibration parameters, and experimental data on target passage speed within the closed passage area. To restore the reference time, the unit is seconds. Determined based on video frame rate and historical occlusion sample experimental data. , , These are dimensionless weighting coefficients. , , Determined experimentally based on correct and incorrect recovery samples from historical occlusion recovery samples.

[0220] When the target category of the target trajectory segment before interruption and the target category of the target trajectory segment after reappearance are both the same as the missing target category, and the identity continuation value meets the identity continuation matching condition, the target trajectory segment before interruption and the target trajectory segment after reappearance are determined to be the target trajectory segments before and after the same target occlusion.

[0221] If multiple re-reproduced target trajectory segments corresponding to the same pre-interruption target trajectory segment meet the identity continuation matching condition, they are sorted in ascending order of identity continuation value, and the re-reproduced target trajectory segment with the highest identity continuation value is taken as the identity continuation matching result.

[0222] Data on restoring traffic flow after obstruction is represented as follows:

[0223]

[0224] in, For the first Data on data restoration after obstruction This is the target trajectory segment before the interruption. For periods with missing data, This is a segment of the target trajectory after reconstruction. To perpetuate the value of identity. This is the target identifier after recovery.

[0225] It should be noted that the data for restoring passage after occlusion is obtained by connecting the target trajectory fragment before the interruption, the time period of missing trajectory, and the target trajectory fragment after recurrence, and retains the identity continuation value and the target identifier after restoration;

[0226] When the target trajectory segment before the interruption and the target trajectory segment after the recurrence are determined to be the same target trajectory segment before and after the target occlusion, the missing target identifier is used as the target identifier after recovery.

[0227] S14: Based on the occlusion recovery data and authorized passage window, verify the entrance position, travel time and exit position of the personnel target box or vehicle target box that is not bound to the authorized passage window within the authorized passage window, and generate abnormal following passage judgment results;

[0228] Specifically, the steps for generating the abnormal following passage determination result are as follows:

[0229] S141: Extract the personnel or vehicle target boxes from the occlusion recovery traffic data that are not bound to authorized traffic windows, and obtain the corresponding target trajectory segments;

[0230] In one specific embodiment, the restored target identifier, target category, and target trajectory fragment are read from the occlusion recovery passage data, and the authorized object identifier, authorized passage window start time, and authorized passage window end time are read from the authorized passage window. Personnel target frames or vehicle target frames that do not correspond to the authorized object identifier in the current authorized passage window and whose target trajectory fragments overlap with the current authorized passage window in frame time are determined as personnel target frames or vehicle target frames that are not bound to an authorized passage window.

[0231] The target trajectory fragment corresponding to the personnel or vehicle target boxes that are not bound to an authorized access window is represented as follows:

[0232]

[0233] in, For the target trajectory segments corresponding to the target boxes of personnel or vehicles that are not bound to authorized access windows. These are the frame times arranged in chronological order. For the target bounding box of personnel or vehicle that is not bound to an authorized access window at frame time The corresponding target box, For the target bounding box of a person or vehicle that is not bound to an authorized access window at frame time The corresponding center point of the target box, For the target bounding box of personnel or vehicle that is not bound to an authorized access window at frame time The corresponding appearance feature vector, The target category corresponding to the personnel or vehicle target boxes that are not bound to the authorized access window.

[0234] It should be noted that: the occlusion recovery data is generated in step S13, the authorized passage window is generated in step S12, and the personnel target box or vehicle target box that is not bound to the authorized passage window refers to the personnel target box or vehicle target box that exists in the occlusion recovery data and does not correspond to the authorized object identifier of the current authorized passage window.

[0235] S142: Based on the target trajectory fragments corresponding to the personnel or vehicle target frames that are not bound to authorized access windows and the authorized access windows, obtain the unbound target window occupancy data;

[0236] In a specific embodiment, the target trajectory segment corresponding to the personnel target frame or vehicle target frame that is not bound to the authorized access window is read, and the start time and end time of the authorized access window in the authorized access window are read. The content of the target trajectory segment that falls within the frame time range of the authorized access window is written into the occupancy record to obtain the unbound target window occupancy data.

[0237] The data occupied by the unbound target window is represented as follows:

[0238]

[0239] in, Data is occupied by an unbound target window. For the target identifier corresponding to the personnel or vehicle target boxes that are not bound to the authorized access window. For authorized access windows, For a trajectory segment within the window, The center point of the target bounding box corresponding to the trajectory segment within the window. This represents the frame time corresponding to the trajectory segment within the window.

[0240] Specifically, the steps to obtain the data occupied by the unbound target window are as follows:

[0241] S142.1: Read the target trajectory segment corresponding to the personnel target box or vehicle target box of the unbound authorized access window, and extract the center point of the target box in the target trajectory segment;

[0242] In one specific embodiment, the target trajectory segment corresponding to the personnel target frame or vehicle target frame of the unbound authorized access window is read. And extract the center point of the target box from each target record in the target trajectory segment.

[0243] The center point of the target bounding box is represented as:

[0244]

[0245] in, For the target bounding box of personnel or vehicle that is not bound to an authorized access window at frame time The corresponding center point of the target box, The x-coordinate of the center point of the target box. The ordinate is the y-coordinate of the center point of the target bounding box.

[0246] The center point of the target box corresponding to the personnel or vehicle target box that is not bound to the authorized access window is represented as follows:

[0247]

[0248] in, The center point of the target frame corresponding to the personnel or vehicle target frame that is not bound to an authorized access window. The center points of the target bounding boxes are arranged in the order of frame time.

[0249] S142.2: Extract the target trajectory segment whose center point falls within the frame time range corresponding to the authorized passage window as the trajectory segment within the window;

[0250] In one specific embodiment, the authorization pass window is read. The start time of the authorization access window in the middle and the end time of the authorization window It reads the target trajectory segments corresponding to the personnel or vehicle target frames of the unbound authorized access windows in the order of frame time.

[0251] When a frame moment in a target trajectory segment satisfies the following formula, the corresponding target record is extracted as the target record within the window:

[0252]

[0253] in, The start time of the authorization access window. For the first segment of the target trajectory Each frame, This is the end time of the authorized passage window.

[0254] The trajectory segment within the window is represented as follows:

[0255]

[0256] in, For a trajectory segment within the window, The frame time that falls within the frame time range corresponding to the authorized passage window. For the trajectory segment within the window at frame time The corresponding target box, For the trajectory segment within the window at frame time The corresponding center point of the target box, For the trajectory segment within the window at frame time The corresponding appearance feature vector, The target category corresponding to the personnel or vehicle target boxes that are not bound to the authorized access window.

[0257] It should be noted that the trajectory fragments within the window are captured from the target trajectory fragments corresponding to the personnel or vehicle target frames that are not bound to the authorized passage window, and the capture is based on the frame time range corresponding to the authorized passage window.

[0258] S142.3: Write the trajectory fragment within the window, the center point of the target box corresponding to the trajectory fragment within the window, and the frame time corresponding to the trajectory fragment within the window into the same occupancy record to obtain the occupancy data of the unbound target window;

[0259] In one specific embodiment, a trajectory fragment within the window is read. Extract the center point of the target box corresponding to the trajectory segment within the window and the frame time corresponding to the trajectory segment within the window. Write the trajectory segment within the window, the center point of the target box corresponding to the trajectory segment within the window, and the frame time corresponding to the trajectory segment within the window into the same occupancy record to obtain the occupancy data of the unbound target window.

[0260] The center point of the target bounding box corresponding to the trajectory segment within the window is represented as follows:

[0261]

[0262] in, The center point of the target bounding box corresponding to the trajectory segment within the window. This represents the center point of the target bounding box corresponding to the trajectory segment within the window at each frame time.

[0263] The frame time corresponding to the trajectory segment within the window is represented as follows:

[0264]

[0265] in, The frame time corresponding to the trajectory segment within the window. These represent the frame times corresponding to the trajectory segment within the window.

[0266] The data occupied by the unbound target window is represented as follows:

[0267]

[0268] in, Data is occupied by an unbound target window. For the target identifier corresponding to the personnel or vehicle target boxes that are not bound to the authorized access window. For authorized access windows, For a trajectory segment within the window, The center point of the target bounding box corresponding to the trajectory segment within the window. This represents the frame time corresponding to the trajectory segment within the window.

[0269] S143: Extract the entry location, same time period, and exit location based on the data occupied by the unbound target window;

[0270] In a specific embodiment, the system reads the trajectory fragments within the window, the center point of the target frame corresponding to the trajectory fragments within the window, and the frame time corresponding to the trajectory fragments within the window from the data occupied by the unbound target window. It also reads the entrance boundary line and the exit boundary line corresponding to the authorized passage window. Based on the intersection of the center point of the target frame corresponding to the trajectory fragments within the window with the entrance boundary line and the exit boundary line, the system extracts the entrance and exit positions. Based on the overlap of the frame time of the trajectory fragments within the authorized object window with the data occupied by the unbound target window, the system extracts the time period of the same object.

[0271] The entrance location is indicated as follows:

[0272]

[0273] in, This is the entrance location. The x-coordinate of the entrance location. The vertical coordinate of the entrance location is . This is the frame time corresponding to the entry position.

[0274] The export location is indicated as:

[0275]

[0276] in, For export location, The x-coordinate of the exit location. The vertical coordinate of the exit location is . This is the frame time corresponding to the exit position.

[0277] The time period for traveling together is represented as:

[0278]

[0279] in, For the same time period, The start time of the same period. This is the end time of the same period.

[0280] It should be noted that: the entrance position is obtained by the intersection frame time of the center point of the target box corresponding to the trajectory segment within the window and the entrance boundary line; the exit position is obtained by the intersection frame time of the center point of the target box corresponding to the trajectory segment within the window and the exit boundary line; and the same-pass time period is obtained by intersecting the frame time corresponding to the data occupied by the unbound target window and the frame time corresponding to the authorized object within the authorized passage window.

[0281] S144: Generate abnormal following judgment results based on entrance location, time period of travel, and exit location;

[0282] In one specific embodiment, the entry position, the time period of the same line, and the exit position are read. The entry boundary distance value is calculated based on the entry position, the exit boundary distance value is calculated based on the exit position, the number of consecutive frames in the same line is calculated based on the time period of the same line, and an abnormal following passage judgment result is generated based on the entry boundary distance condition, the consecutive frame condition, and the exit boundary distance condition.

[0283] The entrance boundary distance value is expressed as:

[0284]

[0285] in, This is the distance value from the entrance boundary. and The x and y coordinates of the entrance location are given. , , , These are the coordinates of the two endpoints of the entrance boundary line.

[0286] The export boundary distance value is expressed as:

[0287]

[0288] in, This is the distance value from the export boundary. and The x and y coordinates of the exit location are given. , , , These are the coordinates of the two endpoints of the export boundary line.

[0289] The number of consecutive frames in the same row is expressed as:

[0290]

[0291] in, The number of consecutive frames in the same row. This is the frame number corresponding to the end time of the same time period. This is the frame number corresponding to the start time of the same time period.

[0292] When the entrance location exists and satisfies The same time period exists and meets the requirements. The export location exists and satisfies When this happens, the judgment category will be recorded as "abnormal follow-through".

[0293] When a travel period exists but the entrance or exit location is missing, the classification will be recorded as "passage pending review".

[0294] When an entrance location exists, a time period exists, and an exit location exists, and two of the conditions for entrance boundary distance, continuous frame, and exit boundary distance are met while the other condition is not met, the judgment category will be recorded as passage pending review.

[0295] When the data occupied by the unbound target window is not available, or the same time period is not available, or less than two of the following conditions are met: entrance boundary distance condition, continuous frame condition, and exit boundary distance condition, the judgment category will be recorded as non-following passage.

[0296] The result of the abnormal following passage determination is expressed as follows:

[0297]

[0298] in, This is the result of the abnormal following passage judgment. For the target identifier corresponding to the personnel or vehicle target boxes that are not bound to the authorized access window. For authorized access windows, This is the entrance location. For the same time period, For export location, To determine the category.

[0299] It should be noted that the entrance reference distance value Export reference distance value Number of reference frames and peers All parameters were determined based on the width of the closed passage area, the camera installation height, the video frame rate, and historical passage sample experimental data.

[0300] The above embodiments are only used to illustrate the technical methods 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 methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for closed-loop management of a park based on AI intelligent recognition, characterized in that: The method includes: Acquire video frame sequences, area boundaries, access equipment status data, and access authorization data for closed access areas. Identify personnel and vehicle target boxes through the video frame sequences and generate target trajectory segments. Obtain access target detection data based on the target trajectory segments. Based on the area boundary, access equipment status data, and access authorization data, an authorized access window is generated according to the time when the authorized object crosses the line at the entrance, the time period for the access equipment to release the passage, and the time when the authorized object crosses the line at the exit. The target bounding box overlap value is calculated on the target detection data to obtain occlusion fragment data. Based on the occlusion fragment data and the authorized passage window, the identity continuation matching of the target trajectory fragment is performed to generate occlusion recovery passage data. Based on the data of occlusion recovery and the authorized passage window, the system verifies the entrance position, time period, and exit position of personnel or vehicle target frames that are not bound to the authorized passage window within the authorized passage window, and generates an abnormal following passage judgment result.

2. The method for closed-loop management of a park based on AI intelligent recognition according to claim 1, characterized in that, The steps to obtain the target detection data are as follows: Acquire the image data corresponding to each video frame in the video frame sequence, identify the person target box and vehicle target box in the image data, and record the frame time, target box center point and appearance feature vector corresponding to the person target box and vehicle target box. Calculate the intersection and union value, center point displacement value, center point movement direction value, and feature vector similarity of target boxes of the same category in adjacent video frames to obtain the target box continuation parameters; Based on the target box continuation parameters, the same person target box or the same vehicle target box is continuated to generate target trajectory segments; Personnel target bounding boxes, vehicle target bounding boxes, frame time, target bounding box center point, appearance feature vector, and target trajectory fragment are written into the same target record to generate traffic target detection data.

3. The method for closed-loop management of a park based on AI intelligent recognition according to claim 2, characterized in that, The steps to generate the target trajectory segment are as follows: Read the target bounding box continuation parameters in the order of frame time, and filter the target bounding boxes of people or vehicles with the same target category in adjacent video frames; Based on the intersection and union values, center point displacement values, center point movement direction values, and feature vector similarity of the same type of target boxes, the contiguous boxes belonging to the same target in adjacent video frames are determined. Connect the continuation frames corresponding to the same target in the order of frame time to generate target trajectory segments.

4. The method for closed-loop management of a park based on AI intelligent recognition according to claim 1, characterized in that, The steps to generate an authorization pass window are as follows: The authorized objects are determined based on the passage authorization data, and the personnel or vehicle target boxes corresponding to the authorized objects are extracted from the passage target detection data. Based on the center point of the target box corresponding to the region boundary and the authorized object, the crossing time data is obtained, which includes the crossing time of the authorized object's entrance and the crossing time of the authorized object's exit. Extract the passage release period of the passage equipment based on the status data of the passage equipment, and perform cross-validation between the passage equipment release period and the crossing time data to obtain the validated release period; An authorized passage window is generated based on the time when the authorized object crosses the line at the entrance, the verification and release period, and the time when the authorized object crosses the line at the exit.

5. The method for closed-loop management of a park based on AI intelligent recognition according to claim 4, characterized in that, The steps to obtain the data on the moment of crossing the line are as follows: The entry and exit boundary lines are determined based on the region boundary, and the target trajectory segments corresponding to the authorized objects are extracted. Read the center point of the target box corresponding to the authorized object in the order of frame time, and calculate the intersection frame time of the center point of the target box with the entrance boundary line and the exit boundary line respectively; The intersection frame time of the target box center point and the entrance boundary line is recorded as the entrance crossing time of the authorized object, and the intersection frame time of the target box center point and the exit boundary line is recorded as the exit crossing time of the authorized object, thus obtaining the crossing time data.

6. The method for closed-loop management of a park based on AI intelligent recognition according to claim 1, characterized in that, The steps for generating data to restore passage after occlusion are as follows: Extract the bounding boxes of people and vehicles at the same time point from the traffic target detection data, and calculate the overlap value between the bounding boxes of people and vehicles; Based on the overlap value of the target bounding box and the missing frame time of the target trajectory segment, the occlusion segment data is obtained; Based on the occlusion fragment data and the authorized passage window, filter the target trajectory fragment before the interruption and the target trajectory fragment after the recurrence; Identity continuation matching is completed based on the target trajectory fragment before the interruption and the target trajectory fragment after the recurrence, and occlusion recovery data is generated.

7. A method for closed-loop management of a park based on AI intelligent recognition as described in claim 6, characterized in that, The steps to obtain occluded fragment data are as follows: Arrange the target bounding box overlap values ​​in frame time order, and extract the personnel target bounding boxes, vehicle target bounding boxes and corresponding frame times corresponding to the target bounding box overlap values; Based on the missing frame times and recurring frame times of the person or vehicle target bounding boxes in consecutive video frames, the trajectory missing time period is obtained; The missing target identifier, missing target category, personnel target box, vehicle target box, target box overlap value, corresponding frame time, and trajectory missing time period are written into the same occlusion record to obtain occlusion segment data.

8. The method for closed-loop management of a park based on AI intelligent recognition according to claim 7, characterized in that, The steps to obtain the missing time period of the trajectory are as follows: Read the target trajectory segments of the same target in the order of frame time, and extract the frame number interval value between adjacent continuation frames in the target trajectory segments; Determine the missing start frame and the recurrence start frame of the person or vehicle target box in consecutive video frames based on the frame number interval value; By pairing the frame time corresponding to the missing starting frame with the frame time corresponding to the recurring starting frame, the time period of the missing trajectory can be obtained.

9. A method for closed-loop management of a park based on AI intelligent recognition according to claim 1, characterized in that, The steps to generate the abnormal following passage determination result are as follows: Extract the bounding boxes of people or vehicles from the occlusion recovery traffic data and obtain the corresponding target trajectory fragments; Based on the target trajectory fragments corresponding to the personnel or vehicle target frames that are not bound to authorized access windows and the authorized access windows, the occupancy data of unbound target windows is obtained. Extract the entry location, time period, and exit location based on the data occupied by the unbound target window; Based on the entrance location, time period of travel, and exit location, an abnormal following passage judgment result is generated.

10. A method for closed-loop management of a park based on AI intelligent recognition according to claim 9, characterized in that, The steps to obtain the data occupied by the unbound target window are as follows: Read the target trajectory segment corresponding to the personnel target box or vehicle target box of the unbound authorized access window, and extract the center point of the target box in the target trajectory segment; The target trajectory segment whose center point of the target box falls within the frame time range corresponding to the authorized passage window is extracted as the trajectory segment within the window; Write the trajectory fragment within the window, the center point of the target box corresponding to the trajectory fragment within the window, and the frame time corresponding to the trajectory fragment within the window into the same occupancy record to obtain the occupancy data of the unbound target window.

Images