Unmanned aerial vehicle highway traffic state monitoring system and method based on dynamic charging network

Abstract

The invention discloses an unmanned aerial vehicle highway traffic state monitoring system and method based on a dynamic charging network, the problem of detection failure in complex weather is solved through visible light, thermal imaging and millimeter-wave radar of a multi-mode sensing fusion technology, the endurance mileage of an unmanned aerial vehicle is increased by 270% and reaches 82km through magnetic resonance coupling of the dynamic charging network and a super capacitor, and the unmanned aerial vehicle traffic state monitoring system and method based on the dynamic charging network are provided. A hierarchical real-time distribution mechanism is subjected to P0-level response delay lt; and high-efficiency transmission of accident information is ensured within 15 seconds. The system covers the whole process of data acquisition, path planning, charging scheduling and information distribution, and significantly improves the highway monitoring efficiency and traffic safety.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation systems technology, specifically relating to an intelligent traffic monitoring system that integrates unmanned aerial vehicle (UAV) dynamic inspection, wireless charging network and real-time information distribution, and is particularly suitable for real-time traffic status monitoring and early warning of the entire highway. Background Technology

[0002] Fixed cameras have blind spots, accounting for 32% of accidents. They are limited by fixed installation angles and interference from obstructions such as bridge piers and large vehicles. They also have insufficient coverage of curves and merging areas of ramps. In low-light nighttime scenarios, the probability of recognition failure increases to 45%.

[0003] Geomagnetic sensors require road surface damage for installation, resulting in high maintenance costs, with annual maintenance costs per point exceeding 8,000 yuan. The installation of coils requires cutting the asphalt layer to a depth of ≥8cm, which leads to a 12%-15% decrease in road surface structural strength. During the rainy season, the base layer is prone to water damage, requiring rework and repair on average every 3 years.

[0004] Satellite remote sensing data has a long update cycle, >15 minutes, making it impossible to respond to sudden congestion in real time. The revisit cycle of low-orbit satellites is constrained by orbital parameters, such as Sentinel-1 which is 12 days. High-orbit satellites have insufficient resolution, >10 meters, and cloud cover results in an effective data acquisition rate of <70%.

[0005] Existing drones, such as the DJI Mavic 3, have a flight time of less than 30 minutes, which is insufficient to meet the continuous monitoring needs of highways. This is due to limitations in the energy density of lithium polymer batteries, currently at a maximum of 260Wh / kg, and aerodynamic power consumption, with hovering power consumption ≥200W. The maximum operating radius of a single drone is less than 10km, requiring frequent returns to change batteries, and an average of more than 20 takeoffs and landings per day.

[0006] Image analysis algorithms suffer from a sharp drop in accuracy under complex weather conditions. In rainy and foggy weather, the F1-score drops to 0.61. Raindrops / fog cause the signal-to-noise ratio of visible light sensors to drop by more than 15dB. Traditional algorithms rely on a single RGB channel and cannot effectively distinguish vehicle outlines from background noise, such as reflective water.

[0007] Traditional VMS, variable message signs, update delay > 5 minutes. Based on the multi-level relay architecture of fiber optic / 4G network, the average number of hops is ≥ 3, resulting in an end-to-end latency of 320±45 seconds. The centralized traffic management platform needs to poll and resolve 120,000+ roadside devices, and the instruction issuance cycle is ≥ 300 seconds.

[0008] Navigation apps rely on historical data and cannot reflect changes in road conditions in real time. Floating car data coverage is less than 15% and less than 8% during off-peak hours. Road condition prediction algorithms need 15-20 minutes to complete state corrections with a confidence level of >90% after a sudden accident, which lags behind actual road changes. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to overcome the above-mentioned technical defects and provide a UAV highway traffic condition monitoring system and method based on a dynamic charging network.

[0010] To address the aforementioned issues, the technical solution of this invention is: a drone-based highway traffic condition monitoring system based on a dynamic charging network, characterized in that: it comprises a closed-loop system consisting of a drone swarm, a dynamic charging network, edge computing nodes, and an information distribution platform; the dynamic charging network is composed of a magnetic resonance coupling device and a supercapacitor composite energy storage system, supporting 30-second fast charging; the modal data fusion adopts deformable convolution and CBAM attention mechanisms to improve recognition accuracy under complex weather conditions; and the information distribution platform triggers 5G broadcasting, BeiDou short messages, and navigation APP multi-channel push based on event priority.

[0011] Furthermore, the dynamic charging network includes a magnetic resonance coupling device, a supercapacitor energy storage system, and a fuzzy PID controller.

[0012] Furthermore, the multimodal data fusion algorithm includes deformable convolution to align visible light, thermal imaging, and millimeter-wave radar feature maps and an adaptive weighting mechanism. The adaptive weighting mechanism dynamically allocates feature weights based on the signal-to-noise ratio of each modality.

[0013] Furthermore, the information distribution platform's distribution strategy is as follows: P0 level events are synchronously triggered by 5G broadcast and BeiDou short message with 100% coverage; P1 level events are linked with the guidance screen through RSU with a coverage radius of 200m; and P2 level events are asynchronously pushed to the navigation APP through Protobuf compressed data.

[0014] The method for monitoring highway traffic conditions using unmanned aerial vehicles (UAVs) based on dynamic charging networks is characterized by the following steps: Step 1: Collaborative acquisition of multimodal traffic data, deployment of UAV formations equipped with multispectral sensors, division of responsibility areas according to highway grids, extraction of lane line texture features through ResNet-50, identification of vehicles crossing lane lines and littering, modeling of vehicle 3D geometric features using PointNet++, detection of abnormal parking and low-temperature objects such as ice, and scanning of road surface smoothness using 16-line LiDAR to generate millimeter-level elevation difference heatmaps;

[0015] Step 2: Dynamic Priority Time Discrimination and Data Compression: Run a multi-dimensional discrimination model, input traffic flow disruption index, speed anomaly coefficient, and accident risk probability, and output P0-P2 event classification. For P0 level data, improve Huffman coding is used to retain full-resolution images. For P1 / P2 level data, AV1 video coding is enabled, and the keyframe interval is reduced to 0.5 seconds.

[0016] Step 3: Multi-channel adaptive distribution and feedback control: P0 event: synchronously trigger 5G broadcast and Beidou short message, coverage rate 100%; P1 event: link with RSU roadside unit and guidance screen, coverage radius 200m; P2 event: WeChat mini program / navigation APP asynchronous push; feedback closed loop: vehicle terminal returns ACK signal, exponential backoff retransmission is enabled when packet loss occurs.

[0017] Step 4: Magnetic resonance supercapacitor charging scheduling: Magnetic resonance charging piles are deployed every 5km along the highway, covering a radius of 0.8-1.2m. The drone reports the SOC in real time, and the charging pile matches the power according to the remaining power. The receiving end is equipped with 6 sets of 2.7V / 3000F capacitors, which can store 243J of energy in 30 seconds of fast charging, support 15km of emergency range, and enable pulse heating in low temperature environment to maintain charging efficiency of more than 85%.

[0018] Step 5: Drone swarm collaborative path planning: Cruise along a preset grid path at an altitude of 50m and a speed of 10m / s. The A* algorithm plans the shortest path and avoids other drones in real time. When SOC < 15%, a charging request is triggered, and the scheduling system allocates the nearest available charging pile with higher priority than P2 level monitoring tasks.

[0019] Step 6: System self-test and fault-tolerant recovery: Self-test is performed every 30 minutes: sensor calibration, checkerboard calibration and temperature compensation, communication link stress test, fault switching strategy: when a single drone fails, the adjacent drone automatically expands its responsibility area, increasing the coverage radius by 50%; when the charging pile is offline, the supercapacitor is used to store power and return to the backup base station.

[0020] Step 7: Data Fusion and Decision Output: Visible light / thermal imaging / millimeter wave feature map fusion is achieved through deformable convolution, event visualization is generated to create a 3D traffic situation sand table, and Bayesian network parameters are updated based on historical data to improve the accuracy of accident prediction.

[0021] Furthermore, step 1, multimodal traffic data collaborative acquisition, includes a multimodal traffic state recognition algorithm. The multimodal traffic state recognition algorithm consists of: a data fusion mechanism that simultaneously processes visible light images, thermal imaging data, and millimeter-wave radar point clouds; an improved YOLOv5 algorithm that introduces the attention mechanism module CBAM to improve the recognition rate of vehicle occlusion scenes; and the design of a weather robustness training strategy and a congestion index calculation model.

[0022] Furthermore, step 2, dynamic priority time discrimination and data compression, includes dynamic charging network design, which includes an optimal deployment algorithm: calculating the distance between charging piles based on the drone's cruising speed v and battery capacity Q; a magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters; and a weather adaptability design: integrating a micro weather station on the top of the charging pile to trigger a dehumidification and heating device.

[0023] The advantages of this invention compared to existing technologies are: Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the system architecture of the present invention.

[0025] Figure 2 This is a schematic diagram of the system structure topology of the present invention. Detailed Implementation

[0026] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. Identical components are indicated by the same reference numerals.

[0027] It should be noted that the terms “front,” “back,” “left,” “right,” “up,” and “down” used in the following description refer to the directions shown in the attached diagram, while the terms “inside” and “outside” refer to the directions toward or away from the geometric center of a specific component, respectively.

[0028] To make the content of this invention easier to understand, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings.

[0029] like Figures 1 to 2 As shown,

[0030] Example 1

[0031] The UAV-based highway traffic condition monitoring system based on a dynamic charging network is characterized by comprising a closed-loop system consisting of a UAV swarm, a dynamic charging network, edge computing nodes, and an information distribution platform. The UAV swarm is equipped with multispectral sensors, visible light sensors, thermal imaging sensors, and LiDAR, and is divided into responsibility areas according to a grid. The dynamic charging network deploys magnetic resonance charging piles every 5km along the highway, supporting 30-second fast charging and low-temperature high-efficiency charging. The edge computing nodes run multimodal data fusion algorithms and event hierarchical models. The information distribution platform triggers 5G broadcasts, BeiDou short messages, RSUs, and navigation APP multi-channel push notifications based on event priorities.

[0032] The dynamic charging network includes a magnetic resonance coupling device, a supercapacitor energy storage system, and a fuzzy PID controller.

[0033] Optimal placement algorithm: Calculate the charging pile spacing based on the drone's cruising speed v (12m / s) and battery capacity Q (6000mAh).

[0034]

[0035] Where η is the charging efficiency and ≥85%, and θ is the highway gradient angle.

[0036] Magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters.

[0037] Weather-adaptive design: The top of the charging pile integrates a mini weather station for wind speed and precipitation detection, which triggers the dehumidification and heating device;

[0038] The multimodal perception fusion technology includes deformable convolution to align visible light, thermal imaging, and millimeter-wave radar feature maps and an adaptive weighting mechanism. The adaptive weighting mechanism dynamically allocates feature weights based on the signal-to-noise ratio of each modality.

[0039] Develop a dynamic feature weight calculation model:

[0040]

[0041] Where S(I) i ) represents the signal-to-noise ratio evaluation function for each modality of data.

[0042] The information distribution platform's distribution strategy is as follows: P0 level events are synchronously triggered by 5G broadcast and Beidou short message with 100% coverage; P1 level events are linked with the guidance screen through RSU with a coverage radius of 200m; and P2 level events are asynchronously pushed to the navigation APP through Protobuf compressed data.

[0043] Real-time information distribution system

[0044]

[0045] Path planning algorithm:

[0046] def dynamic_astar(nav_map,open_l ist=None,goal=None,traffic_weight=None)whi le open_l ist.not_empty():

[0047] current=open_list.pop_lowest_f()if current==goal:returnreconstruct_path()for neighbor in get neighbors(current):

[0048] tentative_g=current.g+movement_cost(current,neighbor)if tentative_g <neighbor.g:

[0049] neighbor.parent current

[0050] neighbor.g = tentative_g

[0051] neighbor.h = heuristic(neighbor,goal) * traffic_weight # Dynamic traffic weight neighbor.f = neighbor.g + neighbor.h.

[0052] Step 1: Multimodal traffic data collaborative collection, deploying drone formations equipped with multispectral sensors, dividing responsibility areas according to highway grids, extracting lane line texture features through ResNet-50, identifying vehicles crossing lane lines and littering, using PointNet++ to model vehicle 3D geometric features, detecting abnormal parking and low-temperature objects such as ice, and using 16-line LiDAR to scan road surface smoothness to generate millimeter-level elevation difference heatmaps;

[0053] Step 1, multimodal traffic data collaborative acquisition, includes a multimodal traffic state recognition algorithm. The multimodal traffic state recognition algorithm is as follows: a data fusion mechanism that simultaneously processes visible light images, thermal imaging data, and millimeter-wave radar point clouds; an improved YOLOv5 algorithm that introduces the attention mechanism module CBAM to improve the recognition rate of vehicle occlusion scenes; and the design of a weather robustness training strategy and a congestion index calculation model.

[0054] Robust training strategies:

[0055] def data_augmentation(img):

[0056] img = add rain filter(img, orm(0.1.0.5)) intens1 t # Add rain and fog noise img = adjust_contrast(img, orm(0.8.1.2) # Dynamic contrast adjustment Camma = ran return img.

[0057] Congestion Index Calculation Model:

[0058]

[0059] Where: N is the number of vehicles, L is the road segment length, v_i is the speed of a single vehicle, and σ_v is the speed variance.

[0060] Innovation Point Two: Dynamic Charging Network Design

[0061] Optimal placement algorithm: Calculate the charging pile spacing based on the drone's cruising speed v (12m / s) and battery capacity Q (6000mAh).

[0062]

[0063] Where η is the charging efficiency (≥85%), and θ is the highway gradient angle.

[0064] Magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters.

[0065] Weather-adaptive design: The top of the charging pile integrates a mini weather station (wind speed and precipitation detection) to trigger the dehumidification and heating device;

[0066] Step 2: Dynamic Priority Time Discrimination and Data Compression: Run a multi-dimensional discrimination model, input traffic flow disruption index, speed anomaly coefficient, and accident risk probability, and output P0-P2 event classification. For P0 level data, improve Huffman coding is used to retain full-resolution images. For P1 / P2 level data, AV1 video coding is enabled, and the keyframe interval is reduced to 0.5 seconds.

[0067] Step 2, dynamic priority time discrimination and data compression, includes dynamic charging network design, which includes an optimal deployment algorithm: calculating the distance between charging piles based on the drone's cruising speed v and battery capacity Q; a magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters; and a weather adaptability design: integrating a micro weather station on the top of the charging pile to trigger a dehumidification and heating device.

[0068] Dynamic charging network construction method: An optimal point placement algorithm based on an energy consumption model is proposed to achieve seamless switching between "charging" and "flight" states for UAVs.

[0069] 3D point layout optimization model:

[0070] Establish an energy consumption equation that takes into account topographic relief:

[0071]

[0072] Where h(t) is the altitude function, k1 = 0.012, k2 = 9.8 m / s 2 Fast charging technology for wind tunnel test calibration coefficients:

[0073] Developing a magnetic resonance-supercapacitor composite energy storage system:

[0074] The charging station's output power can be dynamically adjusted from 50W to 500W (matching the drone's remaining battery power). It employs a dual-coil LCCL resonant topology (transmitter: L1 = 15μH, C1 = 120nF; receiver: L2 = 12μH, C2 = 150nF), with a stable operating frequency of 85kHz. Power output is controlled by real-time Q-value adjustment (range 0.5-8) via DSP. It features six 2.7V / 3000F lithium-ion supercapacitors connected in series, providing a total energy storage of 243J, supporting continuous discharge at 500W peak power for 0.5 seconds, thus resolving instantaneous power fluctuations during drone charging. Based on SOC (remaining battery power) data from the drone's BMS, a fuzzy PID controller (proportional coefficient Kp = 0.8, integral time Ti = 0.2s) is used to achieve linear power adjustment.

[0075] SOC < 20%: 500W constant current fast charging (efficiency ≥ 90%)

[0076] 20% ≤ SOC < 80%: 200-350W constant voltage charging (92% efficiency)

[0077] SOC≥80%: 50W trickle charge (efficiency 95%).

[0078] Step 3: Multi-channel adaptive distribution and feedback control: P0 event: synchronously trigger 5G broadcast and Beidou short message, coverage rate 100%; P1 event: link with RSU roadside unit and guidance screen, coverage radius 200m; P2 event: WeChat mini program / navigation APP asynchronous push; feedback closed loop: vehicle terminal returns ACK signal, exponential backoff retransmission is enabled when packet loss occurs.

[0079] Step 4: Magnetic resonance supercapacitor charging scheduling: Magnetic resonance charging piles are deployed every 5km along the highway, covering a radius of 0.8-1.2m. The drone reports the SOC in real time, and the charging pile matches the power according to the remaining power. The receiving end is equipped with 6 sets of 2.7V / 3000F capacitors, which can store 243J of energy in 30 seconds of fast charging, support 15km of emergency range, and enable pulse heating in low temperature environment to maintain charging efficiency of more than 85%.

[0080] Step 5: Drone swarm collaborative path planning: Cruise along a preset grid path at an altitude of 50m and a speed of 10m / s. The A* algorithm plans the shortest path and avoids other drones in real time. When SOC < 15%, a charging request is triggered, and the scheduling system allocates the nearest available charging pile with higher priority than P2 level monitoring tasks.

[0081] Step 6: System self-test and fault-tolerant recovery: Self-test is performed every 30 minutes: sensor calibration, checkerboard calibration and temperature compensation, communication link stress test, fault switching strategy: when a single drone fails, the adjacent drone automatically expands its responsibility area, increasing the coverage radius by 50%; when the charging pile is offline, the supercapacitor is used to store power and return to the backup base station.

[0082] Step 7: Data Fusion and Decision Output: Visible light / thermal imaging / millimeter wave feature map fusion is achieved through deformable convolution, event visualization is generated to create a 3D traffic situation sand table, and Bayesian network parameters are updated based on historical data to improve the accuracy of accident prediction.

[0083] Example 2

[0084] Example 1

[0085] The UAV-based highway traffic condition monitoring system based on a dynamic charging network is characterized by comprising a closed-loop system consisting of a UAV swarm, a dynamic charging network, edge computing nodes, and an information distribution platform. The UAV swarm is equipped with multispectral sensors, visible light sensors, thermal imaging sensors, and LiDAR, and is divided into responsibility areas according to a grid. The dynamic charging network deploys magnetic resonance charging piles every 5km along the highway, supporting 30-second fast charging and low-temperature high-efficiency charging. The edge computing nodes run multimodal data fusion algorithms and event hierarchical models. The information distribution platform triggers 5G broadcasts, BeiDou short messages, RSUs, and navigation APP multi-channel push notifications based on event priorities.

[0086] The dynamic charging network includes a magnetic resonance coupling device, a supercapacitor energy storage system, and a fuzzy PID controller.

[0087] Optimal placement algorithm: Calculate the charging pile spacing based on the drone's cruising speed v (12m / s) and battery capacity Q (6000mAh).

[0088]

[0089] Where η is the charging efficiency and ≥85%, and θ is the highway gradient angle.

[0090] Magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters.

[0091] Weather-adaptive design: The top of the charging pile integrates a mini weather station for wind speed and precipitation detection, which triggers the dehumidification and heating device;

[0092] The multimodal perception fusion technology includes deformable convolution to align visible light, thermal imaging, and millimeter-wave radar feature maps and an adaptive weighting mechanism. The adaptive weighting mechanism dynamically allocates feature weights based on the signal-to-noise ratio of each modality.

[0093] Develop a dynamic feature weight calculation model:

[0094]

[0095] Where S(I) i ) represents the signal-to-noise ratio evaluation function for each modality of data.

[0096] The information distribution platform's distribution strategy is as follows: P0 level events are synchronously triggered by 5G broadcast and Beidou short message with 100% coverage; P1 level events are linked with the guidance screen through RSU with a coverage radius of 200m; and P2 level events are asynchronously pushed to the navigation APP through Protobuf compressed data.

[0097] Real-time information distribution system

[0098]

[0099]

[0100] Path planning algorithm:

[0101] def dynamic_astar(nav_map,open_l ist=None,goal=None,traffic_weight=None)whi le open_l ist.not_empty():

[0102] current=open_list.pop_lowest_f()if current==goal:returnreconstruct_path()for neighbor in get neighbors(current):

[0103] tentative_g=current.g+movement_cost(current,neighbor)if tentative_g <neighbor.g:

[0104] neighbor.parent current

[0105] neighbor.g = tentative_g

[0106] neighbor.h = heuristic(neighbor,goal) * traffic_weight # Dynamic traffic weight neighbor.f = neighbor.g + neighbor.h.

[0107] Step 1: Multimodal traffic data collaborative collection, deploying drone formations equipped with multispectral sensors, dividing responsibility areas according to highway grids, extracting lane line texture features through ResNet-50, identifying vehicles crossing lane lines and littering, using PointNet++ to model vehicle 3D geometric features, detecting abnormal parking and low-temperature objects such as ice, and using 16-line LiDAR to scan road surface smoothness to generate millimeter-level elevation difference heatmaps;

[0108] Step 1, multimodal traffic data collaborative acquisition, includes a multimodal traffic state recognition algorithm. The multimodal traffic state recognition algorithm is as follows: a data fusion mechanism that simultaneously processes visible light images, thermal imaging data, and millimeter-wave radar point clouds; an improved YOLOv5 algorithm that introduces the attention mechanism module CBAM to improve the recognition rate of vehicle occlusion scenes; and the design of a weather robustness training strategy and a congestion index calculation model.

[0109] Robust training strategies:

[0110] def data_augmentation(img):

[0111] img = add rain filter(img, orm(0.1.0.5)) intens1 t # Add rain and fog noise img = adjust_contrast(img, orm(0.8.1.2) # Dynamic contrast adjustment Camma = ran return img.

[0112] Congestion Index Calculation Model:

[0113]

[0114] Where: N is the number of vehicles, L is the road segment length, v_i is the speed of a single vehicle, and σ_v is the speed variance.

[0115] Innovation Point Two: Dynamic Charging Network Design

[0116] Optimal placement algorithm: Calculate the charging pile spacing based on the drone's cruising speed v (12m / s) and battery capacity Q (6000mAh).

[0117]

[0118] Where η is the charging efficiency (≥85%), and θ is the highway gradient angle.

[0119] Magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters.

[0120] Weather-adaptive design: The top of the charging pile integrates a mini weather station (wind speed and precipitation detection) to trigger the dehumidification and heating device;

[0121] Step 2: Dynamic Priority Time Discrimination and Data Compression: Run a multi-dimensional discrimination model, input traffic flow disruption index, speed anomaly coefficient, and accident risk probability, and output P0-P2 event classification. For P0 level data, improve Huffman coding is used to retain full-resolution images. For P1 / P2 level data, AV1 video coding is enabled, and the keyframe interval is reduced to 0.5 seconds.

[0122] Step 2, dynamic priority time discrimination and data compression, includes dynamic charging network design, which includes an optimal deployment algorithm: calculating the distance between charging piles based on the drone's cruising speed v and battery capacity Q; a magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters; and a weather adaptability design: integrating a micro weather station on the top of the charging pile to trigger a dehumidification and heating device.

[0123] Dynamic charging network construction method: An optimal point placement algorithm based on an energy consumption model is proposed to achieve seamless switching between "charging" and "flight" states for UAVs.

[0124] 3D point layout optimization model:

[0125] Establish an energy consumption equation that takes into account topographic relief:

[0126]

[0127] Where h(t) is the altitude function, k1 = 0.012, k2 = 9.8 m / s 2 Fast charging technology for wind tunnel test calibration coefficients:

[0128] Developing a magnetic resonance-supercapacitor composite energy storage system:

[0129] The charging station's output power can be dynamically adjusted from 50W to 500W (matching the drone's remaining battery power). It employs a dual-coil LCCL resonant topology (transmitter: L1 = 15μH, C1 = 120nF; receiver: L2 = 12μH, C2 = 150nF), with a stable operating frequency of 85kHz. Power output is controlled by real-time Q-value adjustment (range 0.5-8) via DSP. It features six 2.7V / 3000F lithium-ion supercapacitors connected in series, providing a total energy storage of 243J, supporting continuous discharge at 500W peak power for 0.5 seconds, thus resolving instantaneous power fluctuations during drone charging. Based on SOC (remaining battery power) data from the drone's BMS, a fuzzy PID controller (proportional coefficient Kp = 0.8, integral time Ti = 0.2s) is used to achieve linear power adjustment.

[0130] SOC < 20%: 500W constant current fast charging (efficiency ≥ 90%)

[0131] 20% ≤ SOC < 80%: 200-350W constant voltage charging (92% efficiency)

[0132] SOC≥80%: 50W trickle charge (efficiency 95%).

[0133] Step 3: Multi-channel adaptive distribution and feedback control: P0 event: synchronously trigger 5G broadcast and Beidou short message, coverage rate 100%; P1 event: link with RSU roadside unit and guidance screen, coverage radius 200m; P2 event: WeChat mini program / navigation APP asynchronous push; feedback closed loop: vehicle terminal returns ACK signal, exponential backoff retransmission is enabled when packet loss occurs.

[0134] Step 4: Magnetic resonance supercapacitor charging scheduling: Magnetic resonance charging piles are deployed every 5km along the highway, covering a radius of 0.8-1.2m. The drone reports the SOC in real time, and the charging pile matches the power according to the remaining power. The receiving end is equipped with 6 sets of 2.7V / 3000F capacitors, which can store 243J of energy in 30 seconds of fast charging, support 15km of emergency range, and enable pulse heating in low temperature environment to maintain charging efficiency of more than 85%.

[0135] Step 5: Drone swarm collaborative path planning: Cruise along a preset grid path at an altitude of 50m and a speed of 10m / s. The A* algorithm plans the shortest path and avoids other drones in real time. When SOC < 15%, a charging request is triggered, and the scheduling system allocates the nearest available charging pile with higher priority than P2 level monitoring tasks.

[0136] Step 6: System self-test and fault-tolerant recovery: Self-test is performed every 30 minutes: sensor calibration, checkerboard calibration and temperature compensation, communication link stress test, fault switching strategy: when a single drone fails, the adjacent drone automatically expands its responsibility area, increasing the coverage radius by 50%; when the charging pile is offline, the supercapacitor is used to store power and return to the backup base station.

[0137] Step 7: Data Fusion and Decision Output: Visible light / thermal imaging / millimeter wave feature map fusion is achieved through deformable convolution, event visualization is generated to create a 3D traffic situation sand table, and Bayesian network parameters are updated based on historical data to improve the accuracy of accident prediction.

[0138] Cross-modal feature alignment:

[0139] Establish a two-stream neural network architecture, in which:

[0140] The visible light channel uses ResNet-50 to extract texture features and is fine-tuned on open-source datasets such as Cityscapes and BDD100K to enhance the texture sensitivity of lane lines, traffic signs, and vehicle outlines (mIoU reaches 78.5%). TensorRT is used to perform FP16 quantization and layer fusion on the model, achieving a single-frame processing time of <30ms on a Jetson AGX Xavier device. An SE attention module (compression ratio r=16) is inserted after the Stage3 residual block to increase the feature weights of key regions (such as accident vehicles and road cracks) by 2-3 times.

[0141] The thermal imaging / millimeter-wave channel uses PointNet++ to extract geometric features. Based on farthest point sampling (FPS), a local point set is constructed through a spherical neighborhood with a radius of 0.5m. MLP (32→64→128) is used to extract geometric parameters such as the surface curvature and normal vector of obstacles. In the decoding stage, inverse distance weighted interpolation (IDW) is used to fuse the global features of 1024 points with the original point cloud coordinates to achieve centimeter-level spatial positioning (error <15cm). To address multipath interference from millimeter-wave radar, a DBSCAN-based spatial clustering filter (ε=0.3m, MinPts=5) is added to the input layer, reducing the false detection rate to 3.2%.

[0142] Multi-scale feature map alignment is achieved through deformable convolution. A lightweight regression head (3×3Conv→ReLU→3×3Conv) is connected after the backbone network to output 18 channels of offset (corresponding to the x / y direction offset of the 3×3 convolution kernel). In the boundary region of traffic targets (such as the junction of trucks and guardrails), the convolution kernel automatically expands the receptive field to 2.5 times the original size, improving the feature continuity in occluded scenes. The L1 alignment loss + feature similarity loss (SSIM) are jointly optimized to make the cosine similarity of the visible light / thermal imaging feature map at the FPN-P3 layer reach 0.87.

[0143] Fast charging technology:

[0144] Developing a magnetic resonance-supercapacitor composite energy storage system:

[0145] The charging station's output power can be dynamically adjusted from 50W to 500W (matching the drone's remaining battery power). It employs a dual-coil LCCL resonant topology (transmitter: L1 = 15μH, C1 = 120nF; receiver: L2 = 12μH, C2 = 150nF), with a stable operating frequency of 85kHz. Power output is controlled by real-time Q-value adjustment (range 0.5-8) via DSP. It features six 2.7V / 3000F lithium-ion supercapacitors connected in series, providing a total energy storage of 243J, supporting continuous discharge at 500W peak power for 0.5 seconds, thus resolving instantaneous power fluctuations during drone charging. Based on SOC (remaining battery power) data from the drone's BMS, a fuzzy PID controller (proportional coefficient Kp = 0.8, integral time Ti = 0.2s) is used to achieve linear power adjustment.

[0146] SOC < 20%: 500W constant current fast charging (efficiency ≥ 90%)

[0147] 20% ≤ SOC < 80%: 200-350W constant voltage charging (92% efficiency)

[0148] SOC≥80%: 50W trickle charge (efficiency 95%).

[0149] The drone achieves hovering positioning accuracy <2cm (based on UWB + visual QR code composite positioning). Four Decawave DW3000 base stations are deployed, and 3D positioning is achieved based on the TOF (Time of Flight) algorithm, with a single ranging error <3cm. Dynamic positioning frequency is increased to 100Hz through TDoA (Time Difference of Arrival) optimization. It is equipped with an OV9281 global shutter camera (1280×800@120fps), recognizes ArUco QR codes (10cm×10cm, encoding capacity 4096), and uses the PnP algorithm to calculate 6DoF pose (translation error <0.5cm, angle error <0.3°). An extended Kalman filter (EKF) is designed to fuse UWB and visual data.

[0150] Prediction phase: UWB provides prior location estimates (covariance matrix Q = diag[0.01, 0.01, 0.02]).

[0151] Update phase: Visual QR code correction of accumulated error (observation noise matrix R = diag[0.001, 0.001, 0.005])

[0152] A 30-second fast charge can replenish the flight range by 15km. It employs a bidirectional DC / DC topology (98% efficiency). During fast charging, the supercapacitor first absorbs 500W peak power, then transfers energy to the drone's 4S lithium polymer battery (14.8V / 6000mAh) at a constant power of 200W. Before charging, a 10A pulse current (30% duty cycle, 1kHz frequency) is applied to raise the cell temperature from -20℃ to 25℃, ensuring a charging efficiency >85% in low-temperature environments. Based on the drone's aerodynamic model (thrust-to-power ratio 4.2g / W), a 30-second charge injects 4166J of energy (approximately 1.16Wh). Combined with a lightweight body (total weight <800g) and a high-efficiency brushless motor (KV value 1200), it achieves a 15km flight range (measured average power consumption 75Wh / km).

[0153] 3. Tiered real-time distribution mechanism

[0154] Establish an event-priority-driven, multi-channel information push system, and combine it with dynamic path planning algorithms to achieve precise guidance.

[0155] Multi-channel concurrent push technology:

[0156] High Priority (P0): 5G broadcast + BeiDou short message (100% coverage), based on the 3GPP R17 NR Multicast standard, using the Sub-6GHz band (n78: 3.3-3.8GHz), supporting 100ms-level low-latency broadcast within a 500m radius, with a single cell concurrent capacity of up to 1000 terminals. Dual-frequency redundant transmission (B1 I (1561.098MHz) + B2a (1207.14MHz) is enabled, using convolutional coding (code rate 1 / 2) and CRC check to ensure 99.2% information integrity in mountainous / oceanic areas without public network coverage. Dual-channel parallel backup allows for emergency command issuance at 6 bytes / second via BeiDou even in base station failure scenarios, meeting the rigid requirement of 100% coverage in disaster relief scenarios.

[0157] Medium Priority (P1): Roadside Unit (RSU) + Traffic Guidance Screen, based on the IEEE 802.11p protocol, employing MIMO 2×2 antennas and OFDM modulation (subcarrier spacing 156.25kHz), achieving a transmission rate of 54Mbps within a 200m communication distance, and supporting 256QAM high-order modulation. It complies with the JT / T 607-2004 standard, receives commands via RS485 bus, supports GB2312 / Unicode dual encoding, and has a single-screen refresh latency of <1 second. The RSU and guidance screen protocol is compatible with 90% of existing vehicle-mounted OBUs, and its multipath interference resistance is 3 times better than 4G, making it suitable for the complex electromagnetic environment of urban roads.

[0158] Low Priority (P2): Push notifications from WeChat Mini Programs / Navigation Apps, using Protobuf protocol to compress data (60% smaller than JSON), asynchronous distribution via long connections + WebSocket, and supporting offline message caching (up to 72 hours). Integrates with Amap / Baidu SDKs, triggering push notifications by calling the onTrafficEvent interface, and filtering irrelevant information based on user location (geofencing accuracy ±10m). Leveraging a mobile internet user reach rate >95%, it supports millions of concurrent requests through dynamic load balancing (Alibaba Cloud SLB + CDN), with a push success rate >99.5%.

[0159] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A UAV-based highway traffic condition monitoring system based on a dynamic charging network, characterized in that: The system comprises a closed-loop system consisting of a drone swarm, a dynamic charging network, edge computing nodes, and an information distribution platform. The drone swarm is equipped with multispectral sensors, visible light sensors, thermal imaging sensors, and LiDAR, and its responsibility areas are divided into grids. The dynamic charging network deploys magnetic resonance charging piles every 5km along highways, supporting 30-second fast charging and low-temperature high-efficiency charging. The edge computing nodes run multimodal perception fusion technology and an event classification model. The information distribution platform triggers 5G broadcasts, BeiDou short messages, RSUs, and navigation apps through multiple channels based on event priorities.

2. The UAV highway traffic condition monitoring system based on a dynamic charging network according to claim 1, characterized in that: The dynamic charging network includes a magnetic resonance coupling device, a supercapacitor energy storage system, and a fuzzy PID controller.

3. The UAV highway traffic condition monitoring system based on a dynamic charging network according to claim 1, characterized in that: The multimodal perception fusion technology includes deformable convolution to align visible light, thermal imaging, and millimeter-wave radar feature maps, and an adaptive weighting mechanism. The adaptive weighting mechanism dynamically allocates feature weights based on the signal-to-noise ratio of each modality.

4. The UAV highway traffic condition monitoring system based on a dynamic charging network according to claim 1, characterized in that: The information distribution platform's distribution strategy is as follows: P0 level events are synchronously triggered by 5G broadcast and BeiDou short message with 100% coverage; P1 level events are linked with the guidance screen through RSU with a coverage radius of 200m; and P2 level events are asynchronously pushed to the navigation APP through Protobuf compressed data.

5. A method for monitoring highway traffic conditions using unmanned aerial vehicles (UAVs) based on a dynamic charging network, characterized in that: The steps include: Step 1: Collaborative collection of multimodal traffic data, deployment of drone formations equipped with multispectral sensors, division of responsibility areas according to highway grids, extraction of lane line texture features through ResNet-50, identification of vehicles crossing lane lines and littering, modeling of vehicle 3D geometric features using PointNet++, detection of abnormal parking and low-temperature objects such as ice, and scanning of road surface smoothness using 16-line LiDAR to generate millimeter-level elevation difference heat map; Step 2: Dynamic Priority Time Discrimination and Data Compression: Run a multi-dimensional discrimination model, input traffic flow disruption index, speed anomaly coefficient, and accident risk probability, and output P0-P2 event classification. For P0 level data, improve Huffman coding is used to retain full-resolution images. For P1 / P2 level data, AV1 video coding is enabled, and the keyframe interval is reduced to 0.5 seconds. Step 3: Multi-channel adaptive distribution and feedback control: P0 event: synchronously trigger 5G broadcast and Beidou short message, coverage rate 100%; P1 event: link with RSU roadside unit and guidance screen, coverage radius 200m; P2 event: WeChat mini program / navigation APP asynchronous push; feedback closed loop: vehicle terminal returns ACK signal, exponential backoff retransmission is enabled when packet loss occurs. Step 4: Magnetic resonance supercapacitor charging scheduling: Magnetic resonance charging piles are deployed every 5km along the highway, covering a radius of 0.8-1.2m. The drone reports the SOC in real time, and the charging pile matches the power according to the remaining power. The receiving end is equipped with 6 sets of 2.7V / 3000F capacitors, which can store 243J of energy in 30 seconds of fast charging, support 15km of emergency range, and enable pulse heating in low temperature environment to maintain charging efficiency of more than 85%. Step 5: Drone swarm collaborative path planning: Cruise along a preset grid path at an altitude of 50m and a speed of 10m / s. The A* algorithm plans the shortest path and avoids other drones in real time. When SOC < 15%, a charging request is triggered, and the scheduling system allocates the nearest available charging pile with higher priority than P2 level monitoring tasks. Step 6: System self-test and fault-tolerant recovery: Self-test is performed every 30 minutes: sensor calibration, checkerboard calibration and temperature compensation, communication link stress test, fault switching strategy: when a single drone fails, the adjacent drone automatically expands its responsibility area, increasing the coverage radius by 50%; when the charging pile is offline, the supercapacitor is used to store power and return to the backup base station. Step 7: Data Fusion and Decision Output: Visible light / thermal imaging / millimeter wave feature map fusion is achieved through deformable convolution, event visualization is generated to create a 3D traffic situation sand table, and Bayesian network parameters are updated based on historical data to improve the accuracy of accident prediction.

6. The method for monitoring highway traffic conditions using an unmanned aerial vehicle (UAV) based on a dynamic charging network according to claim 5, characterized in that: The multimodal traffic data collaborative acquisition in step 1 includes a multimodal traffic state recognition algorithm, which is as follows: a data fusion mechanism that simultaneously processes visible light images, thermal imaging data, and millimeter-wave radar point clouds; an improved YOLOv5 algorithm that introduces the attention mechanism module CBAM to improve the recognition rate of vehicle occlusion scenes; and the design of a weather robustness training strategy and a congestion index calculation model.

7. The method for monitoring highway traffic conditions using an unmanned aerial vehicle (UAV) based on a dynamic charging network according to claim 5, characterized in that: Step 2, dynamic priority time discrimination and data compression, includes dynamic charging network design, which includes an optimal deployment algorithm: calculating the distance between charging piles based on the drone's cruising speed v and battery capacity Q; a magnetic resonance coupling device: operating frequency 85kHz, transmission power 300W, effective charging distance 0.8-1.2 meters; and a weather adaptability design: integrating a micro weather station on the top of the charging pile to trigger a dehumidification and heating device.

Images