Service area targeted broadcast method and device based on association of vehicle and person identities
By introducing head detection-first and progressive association logic in highway service areas, combined with high-dimensional algorithm addressing and graph database, strong association between people and vehicles and continuous global trajectory tracking are achieved. This solves the problems of fragmented people and vehicle identities and low broadcast efficiency in existing systems, and improves the accuracy of emergency response and broadcast intervention.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN METROPOLITAN RING ROAD NORTHERN SECTION INVESTMENT MANAGEMENT CO LTD
- Filing Date
- 2026-03-11
- Publication Date
- 2026-06-09
AI Technical Summary
The existing monitoring system in highway service areas cannot effectively link the identities of people and vehicles, resulting in low emergency response efficiency, blind spots with overlapping fields of view and lens distortion, and the inability to make accurate broadcasts to specific individuals or groups. Furthermore, the existing broadcasting system cannot make "roll call" announcements to specific individuals or groups, resulting in low intervention efficiency.
By introducing the logic of "head detection first and progressive association", the system uses a small target head detector to lock the pedestrian movement trajectory, combines multiple wide-angle stitched cameras to achieve full-field field of view coverage, forces hard binding of human and vehicle identities through graph database, and uses a combination of high-dimensional algorithm addressing to achieve precise directional broadcasting system to realize human and vehicle identity association and global trajectory continuous tracking.
It enables the sensitive extraction of the dynamic geometric deformation dependency of personnel in complex scenarios, reduces the false alarm rate of behavior analysis, realizes all-time early warning polling and accurate targeted voice broadcasting, and improves the efficiency of emergency response and the pertinence of broadcast intervention.
Smart Images
Figure CN122179043A_ABST
Abstract
Claims
1. A service area directional broadcasting method based on human and vehicle identity association, characterized in that, include: Data is collected on vehicles entering the service area through the video checkpoint at the highway service area entrance to generate a vehicle attribute information set and the vehicle door area. The vehicle attribute information set includes: associated vehicle brand information, vehicle color information and entry timestamp information. Pedestrian bounding boxes are captured in the vehicle door area of the entering vehicle, and the captured pedestrian bounding boxes are mapped and associated with the vehicle attribute information set to generate a human-vehicle topology relationship. Based on the vehicles entering the service area and a preset number of coplanar feature points, a continuous trajectory within the field of view is generated, wherein the preset number of coplanar feature points are located on the ground of the service area where the vehicles entering the service area are located; Based on the continuous trajectory within the field of view, an abnormal behavior alarm event is generated; Based on the pre-built broadcast terminal spatial index tree, the abnormal behavior alarm event is retrieved by node to generate the nearest speaker node and license plate number information; The license plate number information is concatenated with a preset warning message, and the concatenated text information is converted into speech to generate a directional speech audio stream. The directional voice audio stream is transmitted to the speaker node to generate a directional voice broadcast.
2. The method according to claim 1, characterized in that, The process of collecting data on vehicles entering the service area via video checkpoints at highway service areas to generate vehicle attribute information sets and vehicle door areas includes: Vehicle data is collected from vehicles entering the service area to generate a vehicle attribute information set, which includes: associated vehicle brand information, vehicle color information, and entry timestamp information. The driving status of vehicles entering the service area is detected to generate driving status detection results; In response to determining that the driving status detection result represents the parking status result, a preset area is delineated around the door of the vehicle entering the service area to generate the vehicle door area.
3. The method according to claim 1, characterized in that, The step of capturing pedestrian bounding boxes in the vehicle door area of the entering vehicle, and mapping and associating the captured pedestrian bounding boxes with the vehicle attribute information set to generate a human-vehicle topological relationship, includes: Head detection is performed on the pedestrian bounding box to generate a head bounding box; The head bounding box is time-tracked to generate a temporary head tracking identifier; Pedestrian visibility status detection is performed on the vehicle door area of the entering vehicle to generate a pedestrian bounding box pixel integrity index; In response to the pedestrian bounding box pixel integrity index exceeding a preset threshold, the camera's yaw angle, pitch angle, and focal length are calculated based on the wide-angle camera coordinates to drive the camera to capture images and obtain a high-definition full-body image of the pedestrian. Feature extraction is performed on the full-body high-resolution image of the pedestrian to generate a high-dimensional feature vector; The high-dimensional feature vector and the vehicle attribute information set are linked together in a graph database to create a topological relationship between people and vehicles.
4. The method according to claim 1, characterized in that, The step of generating a continuous trajectory within the field of view based on the vehicles entering the service area and a preset number of coplanar feature points includes: A homography matrix is determined for a predetermined number of coplanar feature points on the ground of the service area where the vehicles entering the service area are located, in order to generate a coordinate mapping transformation matrix; Multiply the target foot pixel coordinates with the coordinate mapping transformation matrix to generate the global bird's-eye view physical coordinates; The state of each target instance under the physical coordinates of the global bird's-eye view is predicted and updated to generate a continuous trajectory within the field of view; In response to the determination of a target instance that is temporarily occluded, virtual trajectory points are predicted and output to maintain the continuity of the target identifier and obtain a continuous trajectory within the field of view.
5. The method according to claim 4, characterized in that, The response to determining that a target instance is temporarily occluded, and obtaining a continuous trajectory within the field of view by predicting and outputting virtual trajectory points to maintain the continuity of the target identifier, includes: In response to the determination that a target instance has been briefly occluded, for a target that reappears after crossing a visual blind spot, the camera is triggered to capture the image and the captured image of the reappearing target is input into a multi-granularity network for global and local feature extraction to generate a feature probe of the reappearing target. The cosine similarity between the reproduced target feature probe and the vectors in the lost target feature library is determined to generate a re-identification matching result; In response to the re-identification matching result representing the successful matching result, the interrupted trajectory that was briefly occluded is stitched together in the graph database to generate a continuous trajectory within the field of view.
6. The method according to claim 1, characterized in that, The step of generating abnormal behavior alarm events based on the continuous trajectory within the field of view includes: The coordinates of key skeleton points are extracted from the continuous trajectory within the field of view to generate a sequence of key skeleton points; The sequence of key skeleton points is constructed into a three-dimensional spatiotemporal graph containing spatial and temporal edges to generate graph structure data; The graph structure data is processed by a spatiotemporal graph convolutional network to generate confidence scores for abnormal behavior classification. In response to determining that the confidence level of the abnormal behavior classification exceeds a preset safety threshold, an abnormal behavior alarm event is generated, wherein the abnormal behavior alarm event includes: target number, bird's-eye view coordinates and timestamp.
7. The method according to claim 1, characterized in that, The step of performing node retrieval on the abnormal behavior alarm event based on a pre-built broadcast terminal spatial index tree to generate the nearest speaker node and license plate number information includes: Obtain the three-dimensional installation coordinate set of multiple speakers; A balanced binary tree is constructed from the three-dimensional installation coordinate set to generate a broadcast terminal spatial index tree; Using the bird's-eye view coordinates in the abnormal behavior alarm event as the query point, the broadcast terminal spatial index tree is searched to generate a search node set; Hyperspherical pruning is performed on the retrieved node set to generate the nearest speaker node; Based on the target number of the abnormal behavior alarm event, the license plate number of the associated vehicle is searched in the graph database along the human-vehicle topology relationship to generate the license plate number information.
8. A service area directional broadcasting device based on human and vehicle identity association, characterized in that, include: The data acquisition unit is configured to collect data on vehicles entering the service area through the video checkpoint at the highway service area entrance and exit, in order to generate a set of vehicle attribute information and the vehicle door area of the entering vehicle. The set of vehicle attribute information includes: associated vehicle brand information, vehicle color information and entry timestamp information. The association unit is configured to capture pedestrian bounding boxes in the vehicle door area of the entering vehicle, and to map and associate the captured pedestrian bounding boxes with the vehicle attribute information set to generate a human-vehicle topology association relationship. The first generation unit is configured to generate a continuous trajectory within the field of view based on the service area entry vehicle and a preset number of coplanar feature points, wherein the preset number of coplanar feature points are located on the service area ground where the service area entry vehicle is located. The second generation unit is configured to generate abnormal behavior alarm events based on the continuous trajectory within the field of view; The retrieval unit is configured to perform node retrieval on the abnormal behavior alarm event based on a pre-built broadcast terminal space index tree, so as to generate the nearest speaker node and license plate number information; The conversion unit is configured to concatenate the license plate information with a preset warning message, and to convert the concatenated text information into speech to generate a directional speech audio stream; The transmission unit is configured to transmit the directional voice audio stream to the speaker node to generate directional voice broadcast.
9. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 7.
10. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 7.