CN-121999616-A - Traffic monitoring method and traffic monitoring system based on unmanned aerial vehicle image recognition

CN121999616ACN 121999616 ACN121999616 ACN 121999616ACN-121999616-A

Abstract

The invention belongs to the technical field of traffic monitoring, in particular to a traffic monitoring method and a traffic monitoring system based on unmanned aerial vehicle image recognition, which are characterized in that real-time traffic data are combined, an optimal path is planned by an A-algorithm, and dynamic adjustment is performed; the invention combines the unmanned aerial vehicle cluster cooperative networking and dynamic position supplementing mechanism and the area priority resource allocation strategy, thereby having the advantages of realizing the comprehensive monitoring of the coverage blind areas of the traditional fixed equipment such as suburban roads, urban and rural junction sections, mountain roads and the like and eliminating the traffic monitoring blank area.

Inventors

ZHANG BAICHUN
LIU JINGBO
ZHANG YU
ZHAO HAIMING
ZHANG JIANCHAO

Assignees

北京中控交安科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260311

Claims (10)

1. The traffic monitoring method based on unmanned aerial vehicle image recognition is characterized by comprising the following specific steps: The method comprises the following steps of S1, firstly, combining real-time traffic data, planning an optimal path by an A algorithm and dynamically adjusting, then, adopting a Mesh ad hoc network technology to form an air monitoring network, supporting elastic expansion and automatic position compensation, avoiding blind areas, simultaneously, monitoring the state of an unmanned aerial vehicle in real time, automatically returning to the air and scheduling a standby machine when the power is low, and ensuring stable image acquisition; S2, firstly utilizing a visible light camera, an infrared thermal imaging sensor and a millimeter wave radar to acquire multi-source data, then preprocessing images of different environments by adopting a proprietary algorithm, and then eliminating invalid areas and compressing data to reduce transmission quantity; S3, identifying information by adopting a lightweight transducer algorithm, fusing images and radar data, accurately calculating parameters, early warning in time after identifying abnormality, tracing historical data and linking emergency departments; And S4, firstly, building a four-level transmission network, then displaying the monitoring data on a large screen of a command center in a multi-dimensional manner, facilitating quick positioning, and then providing a solution suggestion based on the data.
2. The traffic monitoring method based on unmanned aerial vehicle image recognition according to claim 1, wherein the specific steps of S1 are as follows: S11, planning an optimal flight path of the unmanned aerial vehicle through an A-algorithm according to a monitoring area preset by a traffic management department and combining real-time traffic data, supporting a dynamic adjustment path, and automatically generating a detour or focusing monitoring path when a new abnormal event is found; s12, realizing real-time data interaction among a plurality of unmanned aerial vehicles by adopting a Mesh ad hoc network technology, and forming an air monitoring network, so that when a certain unmanned aerial vehicle exits due to insufficient electric quantity or signal interruption, surrounding unmanned aerial vehicle position compensation is automatically scheduled, and monitoring blind areas are avoided; and S13, acquiring the electric quantity, the residual endurance time and the sensor state of the unmanned aerial vehicle in real time, automatically triggering a return instruction and scheduling a standby unmanned aerial vehicle to take over tasks when the electric quantity is lower than a threshold value, and simultaneously, calibrating the flight attitude of the unmanned aerial vehicle in real time to ensure the stability of image acquisition.
3. The traffic monitoring method based on unmanned aerial vehicle image recognition according to claim 1, wherein the specific steps of S2 are as follows: S21, each unmanned aerial vehicle is provided with a visible light camera, an infrared thermal imaging sensor and millimeter wave radar three-mode equipment to acquire multi-source data, and the sensor time sequence synchronization technology is used for realizing the alignment of the three-mode data at the same time and the same position and providing multi-dimensional data support for subsequent identification; S22, constructing an environment self-adaptive preprocessing model to automatically match an optimal preprocessing algorithm through image features, and needing no manual intervention; s23, carrying out region clipping on the preprocessed image, only reserving a region containing a traffic target, removing an invalid region, and reducing data transmission quantity.
4. The traffic monitoring method based on unmanned aerial vehicle image recognition according to claim 1, wherein the specific steps of S3 are as follows: S31, performing target recognition on the preprocessed image according to a lightweight transducer target detection algorithm, introducing small sample transfer learning, updating a model by only marking a small number of samples aiming at a special scene, and improving the adaptability of the system to a new scene; s32, after the vehicle speed data acquired by the millimeter wave radar are fused with the image recognition result, calculating traffic parameters of flow, speed and density; s33, automatically triggering early warning, tracing and linkage operation when an abnormal event is identified; The method comprises the steps of S34, firstly, pre-judging potential dangerous behaviors through a time sequence convolution network based on vehicle/pedestrian track data output by a multi-target accurate recognition unit and combining a traffic risk behavior feature library, then supplementing accurate recognition capability for special traffic scenes, recognizing illegal occupation of a fuel vehicle of a roadside charging pile parking space, overtime parking of the new energy vehicle, pushing a reminding by a charging pile management system, comparing a license plate recognition result with a suspected license plate database updated in real time by a traffic management department for license plate auxiliary verification, and if the license plate recognition result is matched with the suspected license plate database, automatically marking the suspected license plate as a high-risk vehicle, and synchronously calling the appearance details of the vehicle for manual secondary verification.
5. The traffic monitoring method based on unmanned aerial vehicle image recognition according to claim 1, wherein the specific steps of S4 are as follows: S41, constructing a four-level data transmission network of the unmanned aerial vehicle-edge node-cloud terminal so as to enable the key data to occupy the bandwidth preferentially by adopting a real-time data priority transmission mechanism, ensuring timeliness, and transmitting non-key data when the bandwidth is idle, so as to avoid network congestion; s42, on a large screen of the traffic control center, based on a map, realizing the visual display of real-time monitoring pictures, traffic parameter thermodynamic diagrams and abnormal event marks, supporting multi-dimensional screening, facilitating traffic control personnel to quickly locate the concerned area, and improving decision-making efficiency; and S43, providing congestion dispersion advice, road condition prediction and emergency scheme to generate decision support according to the historical traffic data and the real-time monitoring result, correlating the decision advice with the actual traffic effect, and continuously optimizing the model through a feedback mechanism to improve the effectiveness of the decision.
6. Traffic monitoring system based on unmanned aerial vehicle image recognition, characterized by comprising: The unmanned aerial vehicle cluster dynamic scheduling module is used for firstly combining real-time traffic data, planning an optimal path by an A-type algorithm and dynamically adjusting, then adopting a Mesh ad hoc network technology to form an air monitoring network, supporting elastic expansion and automatic bit filling, avoiding blind areas, simultaneously monitoring the unmanned aerial vehicle state in real time, automatically returning to the air and scheduling a standby machine when the power is low, and ensuring stable image acquisition; the multi-mode image acquisition and preprocessing module is used for firstly utilizing a visible light camera, an infrared thermal imaging sensor and a millimeter wave radar to acquire multi-source data, then preprocessing images of different environments by adopting a proprietary algorithm, and then eliminating invalid areas and compressing data to reduce transmission quantity; The intelligent image recognition and traffic analysis module is used for recognizing information by adopting a lightweight transducer algorithm, then fusing images and radar data, accurately calculating parameters, and then early warning in time after abnormal recognition, tracing historical data and linking emergency departments; The data interaction and decision support module is used for building a four-level transmission network, displaying monitoring data on a large screen of the command center in a multi-dimensional mode, facilitating quick positioning and providing a solution suggestion based on the data.
7. The traffic monitoring system based on unmanned aerial vehicle image recognition of claim 6, wherein the unmanned aerial vehicle cluster dynamic scheduling module comprises: The system comprises a task planning and path optimizing unit, a dynamic adjustment path supporting unit, a monitoring unit, a control unit and a control unit, wherein the task planning and path optimizing unit is used for planning an optimal flight path of the unmanned aerial vehicle through an A-algorithm according to a monitoring area preset by a traffic management department and combining real-time traffic data, and automatically generating a detour or focusing monitoring path when a new abnormal event is found; the multi-unmanned aerial vehicle cooperative networking unit is used for realizing real-time data interaction among a plurality of unmanned aerial vehicles by adopting a Mesh ad hoc network technology to form an air monitoring network, so that when a certain unmanned aerial vehicle exits due to insufficient electric quantity or signal interruption, peripheral unmanned aerial vehicle position compensation is automatically scheduled, and monitoring blind areas are avoided; The energy and state management unit is used for collecting the electric quantity, the remaining endurance time and the sensor state of the unmanned aerial vehicle in real time, automatically triggering a return instruction and scheduling a standby unmanned aerial vehicle to take over tasks when the electric quantity is lower than a threshold value, and simultaneously, carrying out real-time calibration on the flight attitude of the unmanned aerial vehicle to ensure the stability of image collection.
8. The unmanned aerial vehicle image recognition-based traffic monitoring system of claim 6, wherein the multi-modality image acquisition and preprocessing module comprises: The multi-sensor fusion acquisition unit is used for enabling each unmanned aerial vehicle to carry three-mode equipment of a visible light camera, an infrared thermal imaging sensor and a millimeter wave radar so as to acquire multi-source data, realizes the alignment of the three-mode data at the same time and the same position through a sensor time sequence synchronization technology, and provides multi-dimensional data support for subsequent identification; The image enhancement and denoising unit is used for constructing an environment self-adaptive preprocessing model so as to automatically match an optimal preprocessing algorithm through image features, and manual intervention is not needed; the image clipping and compressing unit is used for clipping the areas of the preprocessed images, only reserving the areas containing traffic targets, removing invalid areas and reducing the data transmission quantity.
9. The unmanned aerial vehicle image recognition-based traffic monitoring system of claim 6, wherein the intelligent image recognition and traffic analysis module comprises: Meanwhile, small sample transfer learning is introduced, and for special scenes, a model can be updated only by a small number of marked samples, so that the adaptability of the system to the new scenes is improved; The traffic parameter real-time calculation unit is used for calculating traffic parameters of flow, speed and density after the vehicle speed data acquired by the millimeter wave radar are fused with the image recognition result; The abnormal event early warning and tracing unit is used for automatically triggering early warning, tracing and linkage operation when the abnormal event is identified; The dynamic risk behavior pre-judging and special identifying unit is used for pre-judging potential dangerous behaviors through a time sequence convolution network based on vehicle/pedestrian track data output by the multi-target accurate identifying unit and combining with a traffic risk behavior feature library, then supplementing accurate identifying capability for special traffic scenes, identifying illegal occupation of fuel vehicles of roadside charging pile parking spaces for new energy vehicles, associating with overtime parking conditions of the new energy vehicles, pushing reminding by the charging pile management system, carrying out auxiliary verification on fake-licensed vehicles, comparing license plate identifying results with a suspected fake-licensed vehicle database updated in real time by a traffic management department, automatically marking the vehicles as high risk vehicles if the vehicles are matched, and synchronously calling the appearance details of the vehicles for manual secondary verification.
10. The unmanned aerial vehicle image recognition-based traffic monitoring system of claim 6, wherein the data interaction and decision support module comprises: the multi-terminal data transmission unit is used for constructing a four-level data transmission network of the unmanned aerial vehicle-edge node-cloud terminal so as to enable the key data to occupy the bandwidth preferentially by adopting a real-time data priority transmission mechanism, ensure timeliness and avoid network congestion when the non-key data is transmitted in the idle bandwidth; The traffic situation visualization unit is used for realizing the visual display of real-time monitoring pictures, traffic parameter thermodynamic diagrams and abnormal event marks on a large screen of the traffic management command center based on a map, supporting multi-dimensional screening, facilitating traffic management personnel to quickly locate a concerned area and improving decision-making efficiency; The traffic decision-making auxiliary unit is used for providing congestion dredging suggestions, road condition prediction and emergency schemes to generate decision support according to historical traffic data and real-time monitoring results, correlating the decision suggestions with actual traffic effects, continuously optimizing the model through a feedback mechanism and improving the effectiveness of decisions.

Description

Traffic monitoring method and traffic monitoring system based on unmanned aerial vehicle image recognition Technical Field The invention relates to the technical field of traffic monitoring, in particular to a traffic monitoring method and a traffic monitoring system based on unmanned aerial vehicle image recognition. Background With the acceleration of the urban process, the mileage of urban roads and the maintenance quantity of motor vehicles are continuously increased, the problems of traffic jam, traffic accidents, illegal driving and the like are increasingly prominent, and the real-time, comprehensive and accurate traffic monitoring needs are increasingly urgent. Traditional traffic monitoring mainly relies on fixed cameras, ground induction coils, microwave radars and other devices, and although partial monitoring functions are realized on core road sections, the problems of limited coverage, poor adaptability, low data timeliness and the like exist under complex road conditions (such as suburban roads, temporary construction road sections and river-crossing bridges), special weather (heavy rain, heavy fog and night) and emergency scenes (traffic accident sites and traffic control areas). Meanwhile, the current artificial intelligent algorithm (such as deep learning target detection and semantic segmentation) is mature, high-efficiency analysis capability is provided for traffic images acquired by the unmanned aerial vehicle, accurate recognition of targets such as vehicles, pedestrians and traffic marks and real-time calculation of traffic parameters (flow, speed and density) can be realized, and data support is provided for traffic management, emergency treatment and road condition prediction, so that the research and development of the traffic monitoring system based on unmanned aerial vehicle image recognition has important practical application value. Based on the above, the conventional traffic monitoring has the following problems: 1. Fixed monitoring points are insufficient in coverage and have multiple blind areas The traditional fixed cameras, the ground induction coils and other devices can only cover urban core roads or intersections, and monitoring blind areas exist in suburban roads, urban and rural junction road sections, mountain roads and other areas. For example, in suburban county roads in 2024, no fixed monitoring equipment is arranged, after a collision accident of a truck and a non-motor vehicle occurs, the accident responsibility is considered to take more than 72 hours due to lack of real-time image data, and the traffic flow change of a road section before the accident cannot be traced, so that the following road condition dredging scheme is delayed. 2. The image acquisition is greatly disturbed by the environment, and the data quality is low The existing unmanned aerial vehicle monitoring is dependent on a single visible light camera, and under complex environments such as heavy rain, heavy fog, night low illumination and the like, problems such as blurring, noise, low contrast and the like easily occur in images, so that the follow-up recognition accuracy is greatly reduced. For example, in the case of stormwater weather (the visibility is less than 50 meters) in a city in 2024, the vehicle contour is blurred in the city expressway image acquired by the unmanned aerial vehicle, and the traditional recognition algorithm misjudges 20% of small cars as non-motor vehicles, so that the traffic flow statistical error exceeds 30%, and an accurate congestion and dredging basis cannot be provided for traffic management departments. 3. Data processing and analysis are lagged, and real-time performance is poor The existing system mostly adopts a mode of unmanned aerial vehicle acquisition-cloud transmission-centralized processing, the data transmission and analysis flow is long, and particularly, data backlog is easy to occur in a traffic peak period or a multi-unmanned aerial vehicle collaborative monitoring scene. For example, in the early peak (7:30-8:30) period of a certain city, 5 unmanned aerial vehicles monitor 3 main roads of the city center at the same time, collected image data is required to be uploaded to a city grade traffic cloud platform through a 4G network, and the time required for generating traffic jam early warning from data collection is 28 minutes, so that traffic departments cannot allocate police officers in time to dredge, and the jammed road sections are continuously expanded to 2 branches at the periphery. 4. The multi-equipment has poor cooperativity and low emergency response efficiency In the existing system, a unified data interaction interface is lacking among equipment/terminals such as an unmanned aerial vehicle, an traffic control center, a traffic police patrol car, an emergency rescue vehicle and the like, and after the unmanned aerial vehicle finds an abnormal event (such as vehicle anchoring and road collapse), the unmanned aerial vehicle cannot