Cross-layer target relay tracking method and system for multi-layer heterogeneous perception node network

Abstract

The application relates to a cross-layer target relay tracking method and system of a multi-layer heterogeneous perception node network. The method comprises the following steps: acquiring tracking data of a target by a current perception layer and judging whether a cross-layer relay triggering condition is met; based on target motion state parameters and scene constraint information, a predicted appearance position and an arrival time window of the target in a field of view of a candidate relay node of an adjacent perception layer are calculated, and a relay node list is generated; a relay activation instruction carrying target feature description information and a predicted arrival area is sent to the candidate relay node; candidate target information detected after active search of the candidate relay node is received, and an identity confirmation operation is performed; after successful confirmation, a continuous global target trajectory is generated by connecting trajectory segments in a unified three-dimensional space-time coordinate system. By using the method, active predictive cross-layer relay tracking of a target in a multi-layer heterogeneous perception node network can be realized.

Description

Technical Field

[0001] This application relates to the fields of multi-sensor collaborative sensing, target tracking and spatial computing technology, and in particular to a cross-layer target relay tracking method and system for a multi-layer heterogeneous sensing node network. Background Technology

[0002] With the development of multi-sensor collaborative sensing and spatial computing technologies, a technological trend has emerged to construct urban or regional sensing networks using heterogeneous sensing nodes deployed at different heights and locations. These technologies aim to achieve wide-area perception and continuous tracking of moving targets within a monitored area through the network deployment of various types of sensors (cameras, LiDAR, millimeter-wave radar, etc.). For example, urban surveillance networks based on fixed ground cameras can provide side-view image acquisition at the ground level; V2X vehicle-to-everything (V2X) systems based on roadside units and vehicle-mounted sensors can achieve the exchange of dual-layer perception data between the roadside and vehicles; and patrol sensing platforms based on aerial drones or balloons can provide wide-area coverage from a top-down perspective. While each of these technologies possesses certain sensing advantages, they are all limited to a single or limited level of sensing perspective.

[0003] However, current urban sensing and target tracking methods, or traditional approaches, have the following structural limitations, which prevent continuous, full-area target tracking from being achieved: First, single-layer perception has numerous blind spots. Ground-based fixed camera surveillance networks, deployed on the ground and low-rise buildings, create many blind spots in densely built-up urban canyon environments due to building obstruction. While aerial platforms offer wide coverage, their overhead view cannot penetrate building roofs or dense vegetation, and they also obstruct views of building backstreets and underground passage entrances. A single-height perception layer cannot fundamentally eliminate these blind spots.

[0004] Secondly, cross-camera target re-identification is difficult and lacks an active relay mechanism. When a target moves from the field of view of one camera to the field of view of another, existing systems generally rely on appearance feature re-matching technology for target identification. This technology is affected by factors such as changes in lighting, viewing angle, and partial occlusion of the target, resulting in a high re-identification failure rate and easily leading to broken target trajectories or target identity confusion. More importantly, existing multi-camera tracking systems can only passively wait for the target to appear in the images of other cameras after it leaves the current camera's field of view. There is no active relay mechanism that proactively predicts which sensor's field of view the target will appear in based on its current motion state and activates that sensor in advance to prepare for receiving the data. This passive working mode is one of the root causes of cross-regional and cross-level tracking interruptions.

[0005] Furthermore, the structural blind spots of fixed sensing node networks cannot be adaptively filled. Fixed-deployment sensing node networks are limited by deployment locations and the building environment, resulting in inherent structural blind spots, such as areas behind buildings, entrances to underground passages, and behind moving obstructions. Existing systems cannot detect these blind spots in real time during operation, nor do they have the capability to proactively schedule available mobile sensing nodes (such as unmanned vehicles and drones) to temporarily deploy to blind spot locations to establish coverage, inevitably leading to tracking loss when a target enters the blind spot.

[0006] Furthermore, sensor data from different types of sensors cannot be effectively fused within a unified real-time spatiotemporal coordinate system. Point cloud data from aerial sensors (overhead view), point cloud data from fixed ground sensors (side view), and point cloud data from mobile vehicle sensors (driving direction) all have different data formats, sampling frequencies, and underlying local coordinate systems. Existing systems lack a unified mechanism for real-time mapping of these heterogeneous sensor data to a shared three-dimensional spatiotemporal coordinate system and for timestamp alignment and fusion. This makes it difficult to collaboratively improve the accuracy and reliability of target state estimation even when multiple sensors simultaneously observe the same target.

[0007] Existing monitoring systems are usually divided into two independent subsystems, outdoor monitoring and indoor monitoring, with the building's exterior wall as the boundary. The coordinate systems of the two are not unified, which causes the target's trajectory to be interrupted when entering or leaving the building entrance, making it impossible to achieve continuous indoor and outdoor tracking.

[0008] Compared to existing technologies, Skynet video surveillance only supports single-layer ground perception and lacks relay and blind spot filling capabilities; V2X vehicle-to-everything (V2X) involves two-layer perception but lacks a cross-layer active relay mechanism and has limited coordinate system fusion; drone patrol perception is a single-layer, non-continuous coverage and lacks the ability to reconstruct occlusion trajectories and achieve seamless indoor-outdoor connectivity. Therefore, there is currently no system globally that can organize aerial floating platforms, high-point fixed nodes, building surface nodes, and ground mobile nodes at different deployment heights into a collaborative, multi-layered heterogeneous perception network, while simultaneously achieving cross-layer active predictive relay tracking of targets, dynamic adaptive filling of coverage blind spots, continuous trajectory reconstruction during occlusion, and seamless indoor-outdoor connectivity. Summary of the Invention

[0009] Based on this, it is necessary to provide a method and system for cross-layer target relay tracking using a multi-layer heterogeneous sensing node network that can achieve continuous target relay tracking across layers, blind zones, and occlusions.

[0010] A method for cross-layer target relay tracking in a multi-layer heterogeneous sensing node network, wherein the multi-layer heterogeneous sensing node network comprises at least two sensing layers with different deployment heights, each sensing layer contains at least one sensing node, and the sensing data collected by each sensing node is mapped to a unified three-dimensional spatiotemporal coordinate system, characterized in that the method includes: Acquire tracking data generated by the sensing nodes in the current sensing layer tracking the target, determine whether the preset cross-layer relay triggering conditions are met based on the tracking data, and generate a relay triggering signal if they are met. Based on the target's motion state parameters extracted from the tracking data by the relay trigger signal and the pre-built scene constraint information, the predicted appearance position and predicted arrival time window of the target in the perception field of at least one candidate relay node in the adjacent perception layer are calculated, and a relay node list containing at least one candidate relay node is generated. Based on the relay node list, a relay activation instruction is sent to at least one candidate relay node, wherein the relay activation instruction carries at least the target's feature description information, the predicted arrival time window, and the predicted arrival area determined based on the predicted location. The system receives target detection information generated when a candidate relay node actively searches the predicted arrival area within the predicted arrival time window and detects a candidate target. Based on the target detection information and the information carried in the relay activation command, the system performs an identity verification operation on the candidate target. After successful identity verification, in a unified three-dimensional spatiotemporal coordinate system, trajectory continuation processing is performed on the first trajectory segment recorded by the current perception node and the second trajectory segment recorded by the candidate relay node to generate a continuous global target trajectory that spans the perception layer.

[0011] A multi-layer heterogeneous sensing node collaborative sensing system, the system comprising: A multi-layer sensing node network includes at least two sensing layers with different deployment heights. Each sensing layer contains at least one sensing node, and each sensing node is configured to map the collected sensing data to a unified three-dimensional spatiotemporal coordinate system. The relay triggering module is configured to acquire tracking data generated by a sensing node in the current sensing layer tracking a target, and determine whether the preset cross-layer relay triggering conditions are met based on the tracking data. The relay prediction module is configured to, in response to the fulfillment of the relay triggering conditions, calculate the predicted occurrence position and predicted arrival time window of the target within the perception field of at least one candidate relay node in the adjacent perception layer, based on the target's motion state parameters extracted from the tracking data and the pre-built scene constraint information. The active activation module is configured to send a relay activation instruction to at least one candidate relay node based on the predicted occurrence location and the predicted arrival time window. The relay activation instruction carries at least target feature description information. The identity verification module is configured to generate target detection information when a candidate relay node detects a candidate target after receiving a relay activation command, and to perform identity verification on the candidate target based on the target detection information and the target feature description information carried in the relay activation command. The trajectory continuation module is configured to perform continuation processing on the first trajectory segment recorded by the perception node in the current perception layer and the second trajectory segment recorded by the candidate relay node in a unified three-dimensional spatiotemporal coordinate system in response to successful identity verification, thereby generating a continuous global target trajectory that spans the perception layer.

[0012] The aforementioned method and system for cross-layer target relay tracking in a multi-layer heterogeneous sensing node network addresses the problem of real-time fusion of data from different types of sensors in existing technologies by constructing a multi-layer heterogeneous sensing node network containing at least two sensing layers with different deployment heights and mapping the sensing data from each layer to a shared three-dimensional spatiotemporal coordinate system. This provides a spatiotemporal reference foundation for cross-layer collaborative tracking. A cross-layer target relay tracking protocol is established. When the tracking confidence of the current sensing layer decreases, the target's position in an adjacent layer is actively predicted based on its motion state, and relay nodes are activated in advance to perform target search and identity verification. This achieves a shift from passive re-identification to active prediction of relay, eliminating cross-layer tracking gaps. Furthermore, by establishing a coverage blind spot detection and active blind spot filling scheduling mechanism, when it is predicted that the target trajectory will enter the blind spot of the fixed node network, available mobile sensing nodes are actively scheduled to be deployed to the blind spot location to establish temporary sensing coverage. This enables the sensing network to dynamically and adaptively fill structural blind spots, ensuring continuous observation of the target within the blind spot. A continuous target trajectory reconstruction mechanism is established during occlusion. When all perception layer nodes lose target observation, trajectory extrapolation is performed based on the pre-occlusion motion state and scene constraints, with uncertainty annotations added. Trajectory continuation is completed when the target reappears, maintaining the continuity of trajectory recording and solving the trajectory loss problem in completely occluded scenarios. The above mechanisms work together to achieve continuous target relay tracking across layers, blind spots, and occlusions. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a cross-layer target relay tracking method for a multi-layer heterogeneous sensing node network in one embodiment. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0015] In one embodiment, such as Figure 1As shown, a cross-layer target relay tracking method for a multi-layer heterogeneous sensing node network is provided. The multi-layer heterogeneous sensing node network includes at least two sensing layers with different deployment heights. Each sensing layer contains at least one sensing node. The sensing data collected by each sensing node is mapped to a unified three-dimensional spatiotemporal coordinate system. Specifically, the establishment of the unified three-dimensional spatiotemporal coordinate system (WCS) and the mapping of sensing data from each sensing node include the following processes: During system initialization, each sensing node determines its own coordinate transformation matrix through a three-level calibration process. This coordinate transformation matrix transforms the node's local sensor coordinate system to the unified three-dimensional spatiotemporal coordinate system. The three-level calibration process can refer to the coordinate calibration methods in related technologies of dimensional infrastructure. Each sensing node synchronizes its clock via NTP or PTP protocols, with a time synchronization accuracy better than 1 millisecond, to ensure that the timestamps of the cross-layer sensing data can be aligned and fused.

[0016] The cross-layer target relay tracking method in a multi-layer heterogeneous sensing node network includes the following steps: Step 102: Obtain the tracking data generated by the sensing nodes in the current sensing layer tracking the target. Determine whether the preset cross-layer relay triggering conditions are met based on the tracking data. If they are met, generate a relay triggering signal.

[0017] The current perception layer refers to the perception layer to which the perception node currently performing the target tracking task in the multi-layer heterogeneous perception node network belongs. For example, it might be the first perception layer (air layer) in the initial tracking phase, and switch to the second perception layer (high-point layer) after the target enters the street. Tracking data includes the target's position, velocity, acceleration, and tracking confidence. Cross-layer relay triggering conditions are used to determine whether the target is about to leave the effective perception range of the current perception layer. Specific triggering conditions may include: the current perception layer's tracking confidence value for the target is less than a first preset threshold (which can be set according to actual needs), and the confidence value shows a decreasing trend; or the distance between the target's current position and the effective perception boundary of the current perception layer is less than a second preset threshold, and the direction of movement is pointing outward from the boundary; or based on the scene map, it is predicted that the target will enter a static occlusion area within a preset time. When any condition is met, a relay trigger signal is generated to start the cross-layer relay process. This step actively monitors and predicts that the target is about to leave the coverage of the current layer, providing a triggering opportunity for subsequent relay prediction and activation, avoiding tracking interruption caused by passive re-identification after the target is lost.

[0018] Step 104: Based on the target's motion state parameters extracted from the tracking data by the relay trigger signal and the pre-constructed scene constraint information, calculate the predicted appearance position and predicted arrival time window of the target within the perception field of at least one candidate relay node in the adjacent perception layer, and generate a relay node list containing at least one candidate relay node.

[0019] Motion state parameters include the target's current position, velocity vector, and acceleration vector. Scene constraint information refers to physical accessibility constraints such as road networks, building outlines, and restricted areas. The calculation process for the Predicted Appearance Zone (PAZ) is as follows: First, extrapolate the set of possible locations of the target within a future period based on the motion model; then, use scene constraint information to trim physically inaccessible locations, forming a set of feasible locations; finally, for each sensing node in the adjacent sensing layer, calculate the spatial intersection between its sensing range and the set of feasible locations, and prioritize each sensing node according to the probability that the intersection is a non-empty set, generating a list of candidate relay nodes. The predicted arrival time window is the time interval during which the target is expected to enter the sensing range of a candidate node. This step utilizes the target's motion trend and spatial constraints to accurately pinpoint the next layer of sensing nodes and their visual field, providing precise guidance for active activation and improving the efficiency and success rate of relay search.

[0020] Step 106: Based on the relay node list, send a relay activation instruction to at least one candidate relay node, wherein the relay activation instruction carries at least the target's feature description information, the predicted arrival time window, and the predicted arrival area determined based on the predicted location.

[0021] The Handoff Activation command is control information sent to candidate relay nodes. The Target Feature Packet includes the target's appearance feature vector, 3D size estimation, and movement speed. If the target carries a wireless positioning tag, such as a UWB tag, it may also include a device identifier. The predicted arrival area is a spatial description of the predicted location within the node's coordinate system. Upon receiving this command, the candidate relay node enters an active search mode: within the predicted arrival time window, it prioritizes processing sensor data for the predicted arrival area at a high frame rate and high resolution to improve the detection response speed when the target appears. This step transforms the traditional passive waiting into active preparation, ensuring the relay node is ready before the target arrives, which is a crucial step in achieving seamless relay.

[0022] Step 108: Receive the target detection information generated when the candidate relay node detects a candidate target after performing an active search of the predicted arrival area within the predicted arrival time window, and perform an identity verification operation on the candidate target based on the target detection information and the information carried in the relay activation instruction.

[0023] The identity verification process ensures that the candidate target detected by the relay node is the same target being tracked in the current layer. Verification methods include: feature matching verification, which calculates the similarity between the candidate target's feature vector and the target feature description information in the relay activation command; if the similarity exceeds a third preset threshold, it is confirmed; motion consistency verification, which verifies whether the position and velocity deviations between the candidate target's detected position and the predicted trajectory are within acceptable limits; and when the target carries a wireless positioning tag, it is preferentially confirmed directly through the tag's device identifier. Once identity verification is successful, it indicates that the relay node has successfully captured the target and can proceed with trajectory continuation. This step, through a multi-verification mechanism, ensures the correctness of the target's identity during cross-layer handover, preventing trajectory confusion caused by mismatches.

[0024] Step 110: After successful identity verification, in a unified three-dimensional spatiotemporal coordinate system, perform trajectory continuation processing on the first trajectory segment recorded by the current perception node and the second trajectory segment recorded by the candidate relay node to generate a continuous global target trajectory that crosses the perception layer.

[0025] The trajectory continuation processing includes aligning the endpoint of the first trajectory segment with the starting point of the second trajectory segment in a unified three-dimensional spatiotemporal coordinate system, and filling the relay transition segment using trajectory interpolation methods to form a continuous, unbroken global target trajectory. When the relay occurs between sensing nodes deployed outdoors and indoors, it is also necessary to use pre-built Building Information Modeling (BIM) data to determine the coordinate transformation relationship, accurately aligning the outdoor coordinate system with the indoor coordinate system to ensure that the trajectory does not change position when crossing indoor and outdoor boundaries. This step ultimately outputs the complete motion trajectory of the target within the entire coverage area of ​​the multi-layer sensing network, providing a continuous data foundation for subsequent trajectory analysis, evidence storage, and other applications.

[0026] In the above method, tracking data of the target tracking nodes in the current perception layer is acquired. Based on this tracking data, it is determined whether the preset cross-layer relay triggering conditions are met. If they are met, a relay triggering signal is generated. By actively monitoring the preset triggering conditions, the relay initiation time is advanced from "passive re-identification after the target has been lost" to "active prediction before the target is about to be lost," fundamentally changing the timing of the relay behavior and avoiding the generation of tracking gaps. At the same time, the tracking data acquisition and triggering judgment involved in this step rely on the premise that the perception data of each perception node has been mapped to a unified three-dimensional spatiotemporal coordinate system. This premise itself solves the problem of the inability to fuse data from different types of sensors in real time—the aerial top-down point cloud, the ground side view image, and the dynamic data of the moving node all have comparable and fusionable spatial position benchmarks under the same coordinate system. Based on the relay triggering signal, the motion state parameters of the target are extracted from the tracking data. Combined with the pre-constructed scene constraint information, the predicted appearance position and predicted arrival time window of the target within the perception field of vision of the candidate relay nodes in the adjacent perception layers are calculated to generate a relay node list. Existing technologies require extensive re-identification and matching across multiple camera feeds after a target is lost, which is not only inefficient but also prone to target confusion due to feature mismatches. This invention, by calculating the predicted location and arrival time window, compresses the search range of relay nodes into a highly defined spatiotemporal window, significantly improving the efficiency and success rate of relay search. Simultaneously, the introduction of scene constraint information solves the blind spot problem of single-layer perception—through a pre-constructed scene map, the system can accurately identify which areas are physically inaccessible, thus constraining the predicted location to the target's actual possible path, avoiding relay failures caused by invalid predictions such as passing through walls or buildings; scene constraint information eliminates invalid predictions of physically inaccessible locations, preventing relay failures due to inaccurate predictions. A relay activation command is sent to candidate relay nodes based on the relay node list. This command carries target feature description information, the predicted arrival time window, and the predicted arrival area. The system sends a target feature description package, including target appearance features, 3D dimensions, movement speed, and even the wireless tag device identifier, along with the predicted spatiotemporal window to the candidate relay nodes. This step is the key turning point that distinguishes this invention from the "passive waiting" mode of existing technologies. In existing systems, each camera operates independently, performing feature comparison only after a target is detected; however, this invention, through a relay activation command, enables candidate relay nodes to enter an active search mode before the target arrives—prioritizing the processing of sensor data at high frame rate and high resolution within a specified time window and a specified spatial area. This mechanism transforms the role of the relay node from a "passive discoverer" to an "active preparer," solving the problem of target misses caused by response lag in existing technologies.Furthermore, the advance transmission of feature description information allows relay nodes to immediately verify the identity of candidate targets upon detection, eliminating the need to request feature data from the central node and reducing the impact of communication delays on real-time performance. The system receives candidate target information actively searched and detected by candidate relay nodes within the prediction spatiotemporal window. Based on this information and the target feature description in the relay activation command, it performs feature similarity matching, motion consistency verification, and wireless tag identifier confirmation to complete the identity verification operation. A multi-dimensional joint verification strategy is employed: feature matching provides appearance-level similarity judgment, motion consistency verification provides spatiotemporal rationality verification, and the wireless tag device identifier provides absolute identity anchoring. These three verification methods form a complementary redundancy. Even when drastic changes in lighting or partial target occlusion reduce the confidence of appearance feature matching, motion consistency verification can still eliminate false matches based on the spatiotemporal constraint that "the target cannot appear in a physically inaccessible location in a short time." When the target carries a wireless tag, device ID confirmation directly bypasses the uncertainties of appearance and motion verification. This multi-layered verification mechanism significantly improves the accuracy of target identity verification during cross-layer handover, fundamentally solving the trajectory breakage problem caused by identity confusion. After successful identity verification, trajectory continuation processing is performed on the first trajectory segment recorded by the current sensing node and the second trajectory segment recorded by the candidate relay node in a unified three-dimensional spatiotemporal coordinate system to generate a continuous global target trajectory across the sensing layer. Precise alignment and interpolation continuation of trajectory segments are achieved using unified physical space coordinates, eliminating splicing errors caused by inconsistencies in sensor coordinate systems. When the relay involves indoor and outdoor boundaries, seamless indoor and outdoor connections are achieved through building information model coordinate alignment, resolving the trajectory interruption problem caused by the separation of indoor and outdoor sensing.

[0027] In one embodiment, at least two sensing layers include at least two of a first sensing layer, a second sensing layer, a third sensing layer, and a fourth sensing layer; the first sensing layer consists of aerial floating sensing nodes deployed at a height of 50 meters or more, providing wide-area sensing coverage in a top-down direction; the second sensing layer consists of fixed sensing nodes deployed at high points between 8 and 30 meters, providing street-level sensing coverage in a side-view direction; the third sensing layer consists of building surface sensing nodes deployed on or near the ground, providing near-field high-precision sensing coverage; and the fourth sensing layer consists of mobile sensing nodes with variable deployment heights, providing on-demand sensing coverage with dynamic perspectives.

[0028] Specifically, the multi-layered heterogeneous sensing node network consists of at least two of the four sensing layers mentioned above, with each layer having different deployment heights, sensing perspectives, and coverage characteristics. The first sensing layer (L1) consists of aerial sensing nodes, which can be tethered balloons or solar-powered airships. Their top-down perspective provides global target location anchoring, covering wide areas inaccessible to fixed nodes. The second sensing layer (L2) consists of high-point fixed sensing nodes deployed at high locations such as lampposts and towers. Their side-view perspective compensates for the vertical occlusion blind spots of L1 in urban canyons, forming an orthogonal perspective with L1 to jointly constrain the target's three-dimensional position. The third sensing layer (L3) consists of building surface sensing nodes deployed on building exteriors, rooftops, or entrances, providing high-precision near-field sensing of buildings. This covers the building shadow blind spots of L1 / L2 and extends into interior spaces. The fourth sensing layer (L4) consists of mobile sensing nodes, which can be vehicles, robots, or temporarily deployed drones, used to dynamically fill coverage blind spots in the fixed node network or provide continuous near-range sensing of important targets. An overlapping area of ​​sensing range is established between adjacent sensing nodes. This overlapping area is the physical basis for cross-layer relay tracking. Its width is required to ensure that the joint observation time of adjacent layer nodes on the same spatial area is not less than the minimum time required for target identification. By constructing a multi-layer heterogeneous sensing network, this method comprehensively utilizes the sensing advantages of different heights and perspectives, eliminates the inherent blind spots of a single sensing layer, and provides a network foundation for achieving continuous tracking across the entire domain.

[0029] In one embodiment, determining whether a preset cross-layer relay triggering condition is met based on tracking data includes performing at least one of the following judgment operations: determining whether the tracking confidence value output by the current sensing node tracking the target is less than a first preset threshold; determining whether the distance between the target's current position and the effective sensing boundary of the current sensing layer is less than a second preset threshold; and predicting whether the target will enter a static occlusion area within a preset time based on the scene map.

[0030] Specifically, the first preset threshold can range from 0.5 to 0.7, and the second preset threshold can range from 5 meters to 15 meters. The tracking confidence score reflects the reliability of the current sensing node's observation of the target. When the confidence score continuously decreases and falls below the threshold, it indicates that the target is gradually leaving the effective sensing range of the node. The distance between the current position and the sensing boundary, combined with the direction of movement, can determine whether the target is about to leave the current layer coverage. Scene map prediction uses prior environmental information to identify occluded areas in advance. By combining the above-mentioned multiple triggering conditions, this embodiment can initiate the relay process in time before the target is about to be lost from observation, avoiding tracking gaps caused by trigger lag.

[0031] In one embodiment, calculating the predicted location of the target within the perceptual field of view of at least one candidate relay node in an adjacent perception layer includes: Based on the target's motion model, the motion state of the target over a future time period is extrapolated to generate a set of possible locations. By using pre-built scene constraint information, the set of possible locations is spatially clipped, and the location points corresponding to physically inaccessible areas are removed to form a set of feasible locations; For each sensing node in an adjacent sensing layer, calculate the spatial intersection between the sensing coverage area of ​​each sensing node and the set of feasible locations; Based on the probability that the spatial intersection is a non-empty set, the priority of each sensing node is sorted to determine the relay node list.

[0032] Specifically, the motion model can employ a uniform velocity model, a uniform acceleration model, or a motion pattern model based on historical trajectory learning. A future time period serves as the prediction time window, the length of which can be dynamically set according to the target's motion speed and scene complexity. Scene constraint information includes spatial constraint data such as road boundaries, building outlines, and restricted areas. Spatial clipping removes locations where the target is unlikely to appear, such as inside buildings or in restricted areas. When calculating spatial intersection, the perception coverage is determined by the node's field of view, effective perception distance, and current pose. Priority ranking ensures that the node most likely to observe the target is selected as the primary relay node. This calculation process, by combining motion prediction and spatial constraints, improves the accuracy of relay node selection and reduces invalid activations.

[0033] In one embodiment, performing an identity verification operation on a candidate target includes at least one of the following methods: Calculate the first similarity between the candidate target feature vector extracted from the target detection information and the target feature description information carried in the relay activation instruction. If the first similarity exceeds the third preset threshold, the identity is determined to be successfully confirmed. The value of the third preset threshold ranges from 0.80 to 0.90. The position and velocity deviations between the detected position of the candidate target and the predicted trajectory extrapolated from the target's motion state parameters are verified. If the position deviation is within the first allowable range and the velocity deviation is within the second allowable range, the identity is determined to be successfully confirmed. When a target carrying a wireless positioning tag is detected, the device identifier of the wireless positioning tag is parsed first to confirm that the candidate target and the tracked target are the same target.

[0034] Specifically, feature vectors can be based on appearance features, motion features, etc., extracted by deep learning models. The first and second allowable ranges can be pre-calibrated based on sensor accuracy and motion model uncertainty. Wireless positioning tags have unique device identifiers, therefore, device ID verification has the highest priority. When a wireless positioning tag is unavailable, the system reverts to a joint verification method combining feature matching and motion consistency verification. This multi-factor authentication mechanism ensures the reliability of target identity during cross-layer handover, effectively preventing target confusion caused by feature matching errors.

[0035] In one embodiment, the method further includes a coverage blind spot filling step, which includes: Maintain a three-dimensional sensing coverage map that is updated in real time and represents the coverage status of the monitoring area by a multi-layered heterogeneous sensing node network. Based on the target's current trajectory, predict the time when the target will enter a blind zone in the 3D perception coverage map where the coverage is below the fourth preset threshold. From the available set of mobile sensing nodes, select at least one mobile sensing node that can reach the boundary of the blind zone before the target enters the blind zone. Send scheduling instructions to the selected mobile sensing nodes. The scheduling instructions are used to instruct the mobile sensing nodes to move to the designated deployment location to perform sensing data collection tasks in the blind spots. The sensing data collected by the mobile sensing nodes are incorporated into a unified three-dimensional spatiotemporal coordinate system to participate in cross-layer data fusion processing; Before sending scheduling instructions to the selected mobile sensing nodes, the process also includes: The calculation process involves finding the optimal deployment location for the mobile sensing node with the objective of maximizing the joint sensing coverage area of ​​the mobile sensing node and the existing fixed sensing nodes after deployment.

[0036] Specifically, the 3D perception coverage map is constructed by fusing the real-time pose, perception range, and environmental occlusion information of perception nodes at each layer. Coverage represents the probability that a spatial location is effectively perceived by at least one perception node. When the predicted target trajectory will enter a blind zone with coverage below a fourth preset threshold, the system triggers a blind zone filling scheduling process. When selecting candidate mobile perception nodes, the node's current position, movement speed, and time required to reach the blind zone boundary must be considered. The calculation of the optimal deployment location aims to maximize the joint coverage area. When multiple mobile perception nodes are available, the optimal combination of locations for joint deployment of multiple nodes can be calculated. After a mobile perception node reaches a designated location, its perception data is incorporated into a unified coordinate system for fusion, thereby eliminating blind zones. This step enables the structural blind zones of the fixed node network to be dynamically filled by mobile nodes, ensuring continuous perception of the target within the blind zones.

[0037] In one embodiment, the method further includes an occlusion trajectory reconstruction step, which includes: When it is detected that all sensing layers in the multi-layer heterogeneous sensing node network have lost tracking of the target within a continuous time period, the known motion state information of the target at the moment of loss is recorded. Based on known motion state information and scene constraint information, the motion trajectory of the target during occlusion is extrapolated and estimated using a kinematic model to generate estimated trajectory data. Parameters representing estimation uncertainty are added to each location point in the estimated trajectory data, and the parameters are dynamically adjusted as the occlusion duration increases. When any sensing node in the sensing layer detects a candidate target again, it performs a consistency match between the detection location of the candidate target and the location indicated by the estimated trajectory data. If the match is successful, it determines that the candidate target is the reappearance of the tracked target, and after correcting the endpoint of the estimated trajectory data with the detection location, it performs trajectory continuation processing.

[0038] Specifically, complete occlusion refers to a state where the tracking confidence of all nodes in the perception layer for the same target drops to zero within a continuous time window. Known motion state information includes the target's last observed position, velocity, and acceleration. The kinematic model can be a uniform linear model or a path constraint model incorporating road orientation. The parameter characterizing the estimation uncertainty can be the radius of the uncertain ellipsoid, which expands with increasing occlusion duration to reflect the accumulation of estimation errors. Upon re-detection, the system compares the detection position of the candidate target with the end position of the estimated trajectory. If the deviation is within the range determined based on the uncertainty parameter, the target is confirmed. After confirmation, the endpoint of the estimated trajectory is corrected using the actual detection position, and trajectory continuation is completed. The generated estimated trajectory segment is labeled "Inferred Trajectory" in the global trajectory record, distinguishing it from the directly observed trajectory. This step maintains the continuity of the trajectory record even in the extreme case where the target is completely invisible, and provides a reliability reference for the inferred segment for trajectory users.

[0039] In one embodiment, the method further includes a cross-layer data fusion step, which includes the following steps when the target is simultaneously observed by multiple sensing nodes from different sensing layers: The tracking confidence score and geometric precision factor corresponding to the position estimation of the target by each sensing node are obtained; the geometric precision factor is determined by the observation distance between the sensing node and the target and the observation angle. Based on the tracking confidence score and geometric precision factor, a fusion weight is calculated for the position estimation results of each sensing node, where the fusion weight is expressed as:

[0040] in, For the first The fusion weights of each sensing node For the first Tracking confidence of each sensing node For the first Geometric precision factor of each sensing node; The position estimation results of each sensing node are weighted and summed according to the fusion weight to calculate the fusion position estimate of the target.

[0041] Specifically, when the target is located in an area covered by overlapping sensing layers, multiple nodes may simultaneously provide effective observations. The geometric accuracy factor reflects the geometric quality of the sensing node's observations; for example, the closer the distance and the more upright the viewing angle, the higher the geometric accuracy factor. The fusion weight is the product of the tracking confidence and the geometric accuracy factor, and the target position estimate obtained through weighted fusion (…). This method offers higher accuracy and reliability than any single-node estimation. It fully utilizes the redundant information from simultaneous observations by multiple heterogeneous nodes, improving the accuracy of target location estimation.

[0042] The geometric precision factor can be determined in the following ways: ,in For the first The observation distance between each sensing node and the target. For the first The angle between the observation line of each sensing node and the normal vector of the target motion plane.

[0043] An exemplary calculation method is as follows: ,in This represents the distance attenuation factor. The closer the observation distance and the more upright the observation angle, the higher the geometric precision factor. In one embodiment, when trajectory continuity processing occurs between an outdoor-deployed sensing node and an indoor-deployed sensing node, the trajectory continuity processing further includes: Obtain pre-built building information model data corresponding to the current building; Based on building information model data, coordinate transformation relationships are determined to precisely align the coordinate systems of outdoor sensing nodes with those of indoor sensing nodes, ensuring that the continuous global target trajectory maintains positional continuity when crossing indoor and outdoor boundaries.

[0044] Specifically, Building Information Modeling (BIM) data contains precise information such as the structural dimensions and coordinate system of the building's interior spaces. By calibrating and aligning the unified outdoor spatiotemporal coordinate system with the indoor coordinate system in the BIM, the rigid body transformation matrix between the two can be determined. When a target enters the building from the outside (e.g., from a shop entrance), the trajectory continuity between the outdoor and indoor nodes of the third perception layer (L3) can be seamlessly connected using this coordinate transformation relationship, avoiding trajectory jumps caused by inconsistencies in coordinate systems. This feature overcomes the limitation of the existing system's separation of indoor and outdoor perception.

[0045] In one embodiment, for multiple trajectory segments that constitute a continuous global target trajectory and are recorded by sensing nodes of different sensing layers, the cryptographic hash values ​​corresponding to the multiple trajectory segments are calculated respectively. The hash values ​​of subsequent trajectory segments generated by the same target during the cross-layer relay process are associated and stored with the hash values ​​of the corresponding trajectory segments of the previous sensing layer in a chain data structure, forming an immutable trajectory evidence chain to prove the continuity of the target's cross-layer tracking process.

[0046] Specifically, each trajectory segment has its cryptographic hash value calculated immediately after generation. The hash value of subsequent segments is calculated using the hash value of the previous segment as input, thus constructing a hash chain. This hash chain proves the temporal relationship and integrity between trajectory segments; any tampering with historical trajectory segments will result in a mismatch in subsequent hash values. This trajectory evidence chain is specifically designed to prove the continuity of the target in a cross-layer relay tracking process. Unlike other trajectory hash chains used to prove spatial location continuity, the former emphasizes proving the behavioral continuity of the cross-layer relay, while the latter emphasizes proving the positional continuity of the spatial trajectory.

[0047] The following is an example illustrating the practical use of the proposed solution: In one embodiment, the method of this application is used to perform cross-layer continuous tracking of a target in an urban scene. First, the scene is set up, including a commercial street in a city with high building density. The first perception layer (L1) balloons cover the entire area, the second perception layer (L2) lamppost nodes are deployed along the main road with a spacing of about 80 meters, and the third perception layer (L3) building exterior wall nodes cover the main shop entrances. The target enters the street from the square, passes through multiple building obstruction areas, and then enters a shop.

[0048] In the initial tracking phase (L1 dominant), when the target is walking in the square, the first perception layer balloon can directly observe the target from a top-down perspective. The L1 node continuously tracks the target with a tracking confidence of 0.92, and the target's coordinate position in the unified three-dimensional spatiotemporal coordinate system (WCS) is updated in real time.

[0049] During the L1-to-L2 relay triggering and execution phase, after the target enters the street, building occlusion causes the tracking confidence of the L1 node to gradually decrease from 0.92 to 0.58, falling below the first preset threshold, thus triggering the cross-layer relay process. Based on the target's current motion state (position, velocity, acceleration) and scene constraint information (road direction, building outline), the relay prediction module predicts that the target will appear within the perception range of node L2-Node-07 in the second perception layer approximately 13 seconds later. The system generates a relay node list, designating L2-Node-07 as a candidate relay node, and sends a relay activation command to it. The command includes a description of the target's appearance features, the predicted arrival time window, and the predicted arrival area. Upon receiving the command, L2-Node-07 enters active search mode, increasing the sensor frame rate to 30fps within the predicted arrival area. Approximately 11 seconds later, L2-Node-07 detected a candidate target within the prediction area. The first similarity between the candidate target's feature vector and the target feature description information in the relay activation command was calculated to be 0.89, exceeding the third preset threshold of 0.85. Simultaneously, the position and velocity deviations in the motion consistency verification were within acceptable ranges, indicating successful identity confirmation. The trajectory continuation module aligned the endpoint of the first trajectory segment recorded by L1 with the starting point of the second trajectory segment recorded by L2-Node-07 in a unified coordinate system. Through trajectory interpolation, a relay transition segment of approximately 0.3 meters was filled to generate a continuous global target trajectory.

[0050] During the occlusion processing phase, the target continues walking along the street and is completely occluded by a building at a building corner. At this point, the tracking confidence of both L1 and L2 nodes simultaneously drops to zero within a continuous time period, and the system determines that the target has entered a state of complete occlusion. The occlusion trajectory reconstruction module records the target's last known state (position, velocity approximately 1.3 m / s, acceleration) at the moment of occlusion. Based on this known state and road direction constraints, the system uses a kinematic model to extrapolate the trajectory, generating estimated trajectory data for approximately 5.2 seconds during the occlusion period. An uncertainty ellipsoid radius is added to each location point on the estimated trajectory, and this radius gradually increases from an initial 0.5 meters to 1.8 meters as the occlusion duration increases.

[0051] During the L3 relay and indoor tracking phase, as the target approaches the shop entrance and exits the obstructed area, the nodes deployed on the building's exterior wall in the third sensing layer regain observation of the target and detect the candidate target. The system performs consistency matching between the detected position of the candidate target and the estimated trajectory data's end position. The deviation between the detected position and the estimated position is 0.4 meters, which is within the allowable range determined based on the radius of the uncertainty ellipsoid, indicating a successful match. The system determines that the candidate target has reappeared as the tracked target, corrects the estimated trajectory's endpoint with the actual detected position, and completes the trajectory continuation. The estimated trajectory segment generated during the obstruction period is labeled "Inferred Trajectory" in the global trajectory record to distinguish it from the measured trajectory segment generated by direct observation. After the target enters the shop, the indoor sensing nodes in the third sensing layer achieve precise alignment between the outdoor and indoor coordinate systems through pre-built Building Information Model (BIM) data, continuously tracking the target's movement trajectory in the indoor space, achieving seamless indoor-outdoor transition.

[0052] In another embodiment, the method of this application is used to actively schedule mobile sensing nodes to fill the coverage blind spot when a mobile target enters the coverage blind spot of the fixed sensing node network, specifically including: First, the scenario is set up as an industrial park. The first sensing layer (L1) consists of balloons covering the main area of ​​the park, and the second sensing layer (L2) consists of streetlight nodes deployed along the main roads. Behind Warehouse B, there is a space of approximately 35 meters x 20 meters. This area has a blind spot because L1 is obscured by the warehouse roof and no L2 nodes are deployed. A forklift is heading towards the back of Warehouse B.

[0053] During the blind spot prediction and triggering phase, the system maintains a real-time updated 3D perception coverage map, which is generated by integrating the real-time pose, perception range, and environmental occlusion information of nodes at each layer. Based on the forklift's current trajectory, facing the back of warehouse area B at a speed of approximately 3 m / s, the system predicts that the forklift will enter a blind spot in the 3D perception coverage map with a coverage level lower than the fourth preset threshold in approximately 12 seconds, triggering the blind spot filling scheduling process.

[0054] During the candidate node selection and deployment location calculation phase, the system filters from the available set of fourth-layer (L4) mobile sensing nodes. Currently, there are four available L4 mobile nodes. The system selects two nodes that can reach the blind zone boundary before the target enters the blind zone: node L4-A is currently about 20 meters from the eastern boundary of the blind zone, and node L4-B is about 30 meters from the southern boundary. The system aims to maximize the joint sensing coverage area of ​​the mobile sensing nodes and existing fixed sensing nodes, and calculates the optimal deployment location for each candidate node. Calculations show that when L4-A is deployed at the midpoint of the eastern boundary of the blind zone, it can cover approximately 68% of the blind zone area on its own; when L4-B is deployed at the midpoint of the southern boundary of the blind zone, it can additionally cover the remaining approximately 32% of the area. The joint deployment of these two nodes can achieve complete coverage of the blind zone.

[0055] During the scheduling execution and blind spot perception phase, the system issues scheduling instructions to L4-A and L4-B respectively. These instructions include the target deployment location, arrival time limit, and perception task parameters. L4-A arrives at the designated deployment location within 9 seconds, establishing perception coverage 3 seconds before the forklift enters the blind spot. After the forklift enters the blind spot, its complete movement trajectory within the blind spot is continuously recorded jointly by L4-A and L4-B without any trajectory interruption. Once the forklift leaves the blind spot, L1 resumes direct observation and perception of the target. The system sends a release scheduling instruction to L4-A and L4-B nodes. Upon receiving the instruction, both nodes end the blind spot task and return to their standby positions.

[0056] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0057] In one embodiment, a multi-layer heterogeneous sensing node collaborative sensing system is provided, comprising: a multi-layer sensing node network, including at least two sensing layers with different deployment heights, each sensing layer including at least one sensing node, and each sensing node configured to map the collected sensing data to a unified three-dimensional spatiotemporal coordinate system; The relay triggering module is configured to acquire tracking data generated by a sensing node in the current sensing layer tracking a target, and determine whether the preset cross-layer relay triggering conditions are met based on the tracking data. The relay prediction module is configured to, in response to the fulfillment of the relay triggering condition, calculate the predicted occurrence position and predicted arrival time window of the target within the perception field of at least one candidate relay node in the adjacent perception layer, based on the motion state parameters of the target extracted from the tracking data and the pre-constructed scene constraint information. The active activation module is configured to send a relay activation instruction to at least one of the candidate relay nodes based on the predicted occurrence location and the predicted arrival time window, wherein the relay activation instruction carries at least target feature description information. The identity verification module is configured to generate target detection information when a candidate relay node detects a candidate target after receiving the relay activation instruction, and to perform identity verification on the candidate target based on the target detection information and the target feature description information carried in the relay activation instruction. The trajectory continuation module is configured to perform continuation processing on the first trajectory segment recorded by the perception node in the current perception layer and the second trajectory segment recorded by the candidate relay node in the unified three-dimensional spatiotemporal coordinate system in response to successful identity confirmation, thereby generating a continuous global target trajectory of the target across the perception layer.

[0058] Specific limitations regarding a multi-layer heterogeneous sensing node collaborative sensing system can be found in the limitations of the cross-layer target relay tracking method in multi-layer heterogeneous sensing node networks described above, and will not be repeated here. Each module in the aforementioned multi-layer heterogeneous sensing node collaborative sensing system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0059] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0060] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0061] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A cross-layer target relay tracking method for a multi-layer heterogeneous sensing node network, characterized in that, The multi-layer heterogeneous sensing node network comprises at least two sensing layers with different deployment heights, each sensing layer containing at least one sensing node, and the sensing data collected by each sensing node is mapped to a unified three-dimensional spatiotemporal coordinate system. The method is characterized by comprising: Acquire tracking data generated by the sensing nodes in the current sensing layer tracking the target, and determine whether the preset cross-layer relay triggering conditions are met based on the tracking data. If they are met, generate a relay triggering signal. Based on the motion state parameters of the target extracted from the tracking data by the relay trigger signal and the pre-constructed scene constraint information, the predicted appearance position and predicted arrival time window of the target in the perception field of at least one candidate relay node in the adjacent perception layer are calculated, and a relay node list containing the at least one candidate relay node is generated. Based on the relay node list, a relay activation instruction is sent to at least one of the candidate relay nodes, wherein the relay activation instruction carries at least the feature description information of the target, the predicted arrival time window, and the predicted arrival area determined based on the predicted occurrence location. The system receives target detection information generated when the candidate relay node detects a candidate target after actively searching the predicted arrival area within the predicted arrival time window, and performs identity verification operation on the candidate target based on the target detection information and the information carried in the relay activation instruction. After the identity verification operation is successful, in the unified three-dimensional spatiotemporal coordinate system, trajectory continuation processing is performed on the first trajectory segment recorded by the perception node in the current perception layer and the second trajectory segment recorded by the candidate relay node to generate the continuous global target trajectory of the target across the perception layer.

2. The method according to claim 1, characterized in that, The at least two sensing layers include at least two of the following: a first sensing layer, a second sensing layer, a third sensing layer, and a fourth sensing layer. The first sensing layer consists of aerial floating sensing nodes deployed at a height of 50 meters or more, providing wide-area sensing coverage in a top-down direction. The second sensing layer consists of fixed sensing nodes deployed at high points between 8 and 30 meters, providing street-level sensing coverage in a side-view direction. The third sensing layer consists of sensing nodes deployed on building surfaces or near the ground, providing near-field high-precision sensing coverage. The fourth sensing layer consists of mobile sensing nodes with variable deployment heights, providing on-demand sensing coverage with dynamic perspectives.

3. The method according to claim 1, characterized in that, The step of determining whether the preset cross-layer relay triggering conditions are met based on the tracking data includes performing at least one of the following judgment operations: determining whether the tracking confidence value output by the current sensing node tracking the target is less than a first preset threshold; determining whether the distance between the current position of the target and the effective sensing boundary of the current sensing layer is less than a second preset threshold; and predicting whether the target will enter the static occlusion area within a preset time based on the scene map.

4. The method according to claim 1, characterized in that, Calculating the predicted location of the target within the perceptual field of view of at least one candidate relay node in an adjacent perceptual layer includes: Based on the motion model of the target, the motion state of the target in a future time period is extrapolated to generate a set of possible locations. Using the pre-built scene constraint information, the set of possible locations is spatially clipped, and location points corresponding to physically inaccessible areas are removed to form a set of feasible locations; For each sensing node in the adjacent sensing layer, calculate the spatial intersection between the sensing coverage area of ​​each sensing node and the set of feasible locations; The relay node list is determined by prioritizing each sensing node based on the probability that the spatial intersection is a non-empty set.

5. The method according to claim 1, characterized in that, Performing an identity verification operation on the candidate target includes at least one of the following methods: Calculate the first similarity between the candidate target feature vector extracted from the target detection information and the target feature description information carried in the relay activation instruction. If the first similarity exceeds a third preset threshold, the identity is determined to be successfully confirmed. The positional and velocity deviations between the detected position of the candidate target and the predicted trajectory extrapolated from the motion state parameters of the target are verified. If the positional deviation is within a first allowable range and the velocity deviation is within a second allowable range, the identity is determined to be successfully confirmed. When the target is detected to be carrying a wireless positioning tag, the device identifier of the wireless positioning tag is parsed first to confirm that the candidate target and the tracked target are the same target.

6. The method according to claim 1, characterized in that, The method further includes a coverage blind spot filling step, which includes: Maintain a three-dimensional sensing coverage map that is updated in real time and represents the coverage status of the monitoring area by the multi-layer heterogeneous sensing node network. Based on the target's current trajectory, predict the time when the target will enter a blind zone in the 3D perception coverage map where the coverage is below a fourth preset threshold; From the available set of mobile sensing nodes, at least one mobile sensing node is selected that can reach the boundary of the blind zone before the target enters the blind zone. A scheduling instruction is sent to the selected mobile sensing nodes, the scheduling instruction being used to instruct the mobile sensing nodes to move to a designated deployment location to perform sensing data collection tasks in the blind area; The sensing data collected by the mobile sensing nodes are incorporated into the unified three-dimensional spatiotemporal coordinate system to participate in cross-layer data fusion processing. Before sending scheduling instructions to the selected mobile sensing nodes, the method further includes: The calculation process for the specified deployment location includes finding the optimal deployment location of the mobile sensing node with the objective of maximizing the joint sensing coverage area of ​​the mobile sensing node and the existing fixed sensing nodes after deployment.

7. The method according to claim 1, characterized in that, The method further includes an occlusion trajectory reconstruction step, which includes: When it is detected that all sensing layers in the multi-layer heterogeneous sensing node network have lost tracking of the target within a continuous time period, the known motion state information of the target at the moment of loss is recorded. Based on the known motion state information and the scene constraint information, the motion trajectory of the target during the occlusion period is extrapolated and estimated using a kinematic model to generate estimated trajectory data. Parameters representing the estimation uncertainty are added to each position point in the estimated trajectory data, and these parameters are dynamically adjusted as the occlusion duration increases. When any sensing node in the sensing layer re-detects the candidate target, the detection position of the candidate target is matched with the position indicated by the estimated trajectory data. If the match is successful, the candidate target is determined to be the reappearance of the tracked target. After correcting the endpoint of the estimated trajectory data with the detection position, trajectory continuation processing is performed.

8. The method according to claim 1, characterized in that, The method further includes a cross-layer data fusion step, which is performed when the target is simultaneously observed by multiple sensing nodes of different sensing layers. The tracking confidence score and geometric precision factor corresponding to the position estimation of the target by each sensing node are obtained; the geometric precision factor is determined by the observation distance and observation angle between the sensing node and the target. Based on the tracking confidence score and the geometric precision factor, a fusion weight is calculated for the position estimation results of each sensing node, wherein the fusion weight is expressed as: in, For the first The fusion weights of each sensing node For the first Tracking confidence of each sensing node For the first Geometric precision factor of each sensing node; The position estimation results of each sensing node are weighted and summed according to the fusion weight to calculate the fusion position estimate of the target.

9. The method according to claim 1, characterized in that, When the trajectory continuity processing occurs between an outdoor-deployed sensing node and an indoor-deployed sensing node, the trajectory continuity processing further includes: Obtain pre-built building information model data corresponding to the current building; Based on the building information model data, the coordinate transformation relationship is determined to accurately align the coordinate system on which the outdoor sensing nodes are based with the coordinate system on which the indoor sensing nodes are based, ensuring that the continuous global target trajectory maintains positional continuity when crossing indoor and outdoor boundaries.

10. The method according to claim 1, characterized in that, The method further includes: For the multiple trajectory segments that constitute the continuous global target trajectory and are recorded by sensing nodes of different sensing layers, the cryptographic hash values ​​corresponding to the multiple trajectory segments are calculated respectively. The hash values ​​of subsequent trajectory segments generated by the same target during the cross-layer relay process are associated and stored with the hash values ​​of the corresponding trajectory segments of the previous perception layer in a chain data structure to form an immutable trajectory evidence chain used to prove the continuity of the target's cross-layer tracking process.

11. A multi-layer heterogeneous sensing node collaborative sensing system, characterized in that, The system includes: A multi-layer sensing node network includes at least two sensing layers with different deployment heights. Each sensing layer contains at least one sensing node, and each sensing node is configured to map the collected sensing data to a unified three-dimensional spatiotemporal coordinate system. The relay triggering module is configured to acquire tracking data generated by a sensing node in the current sensing layer tracking a target, and determine whether the preset cross-layer relay triggering conditions are met based on the tracking data. The relay prediction module is configured to, in response to the fulfillment of the relay triggering condition, calculate the predicted occurrence position and predicted arrival time window of the target within the perception field of at least one candidate relay node in the adjacent perception layer, based on the motion state parameters of the target extracted from the tracking data and the pre-constructed scene constraint information. The active activation module is configured to send a relay activation instruction to at least one of the candidate relay nodes based on the predicted occurrence location and the predicted arrival time window, wherein the relay activation instruction carries at least target feature description information. The identity verification module is configured to generate target detection information when a candidate relay node detects a candidate target after receiving the relay activation instruction, and to perform identity verification on the candidate target based on the target detection information and the target feature description information carried in the relay activation instruction. The trajectory continuation module is configured to perform continuation processing on the first trajectory segment recorded by the perception node in the current perception layer and the second trajectory segment recorded by the candidate relay node in the unified three-dimensional spatiotemporal coordinate system in response to successful identity confirmation, thereby generating a continuous global target trajectory of the target across the perception layer.

Images