An unmanned aerial vehicle-based traffic anomaly detection method, device, equipment and medium

Abstract

The application discloses a kind of traffic anomaly detection method, device, equipment and medium based on unmanned plane, it is related to intelligent traffic monitoring and unmanned plane inspection technical field, the application is by obtaining the geographic position coordinates of traffic anomaly event, determine the obstacle information of influence unmanned plane flight height in preset area;First flight path is planned based on obstacle information and geographic position coordinates, control unmanned plane flies to preset position;Unmanned plane collects the picture of event at preset height, extracts the feature set of abnormal event and combines expected feature set to calculate the completeness of abnormal event, according to completeness, the preset position is determined again.Relative to prior art, the closed-loop iteration mechanism of information acquisition, path planning, image acquisition, completeness determination, shooting position dynamic adjustment makes unmanned plane adjust shooting position according to event actual development trend, eliminates the influence of obstacle shelter and shooting dead angle, improves the monitoring effect of unmanned plane inspection to traffic anomaly event.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic monitoring and drone inspection technology, and in particular to a method, device, equipment and medium for detecting traffic anomalies based on drones. Background Technology

[0002] Existing methods for detecting traffic anomalies using drones include: using drones for fixed-orbit navigation to collect road traffic video data from different altitudes and perspectives, constructing a traffic event information dataset; building a coarse-grained anomaly detection model based on the traffic event information dataset; obtaining anomaly event datasets based on the coarse-grained anomaly detection model; building a fine-grained anomaly detection model based on the anomaly event dataset; deploying both the coarse-grained and fine-grained anomaly detection models; and using these models to detect traffic anomalies. By dividing the task into coarse-grained event detection at the edge of the drone and fine-grained anomaly detection deployed in the cloud, efficient perception of road traffic events is achieved, providing comprehensive coverage of the application road segment and improving detection effectiveness. However, existing technologies have limitations: because the drone's flight path on the road is fixed, the drone's shooting position is also fixed, leading to blind spots in the onboard camera's view and obstruction from obstacles, resulting in poor monitoring effectiveness of drone patrols for traffic anomalies. Summary of the Invention

[0003] This invention provides a method, apparatus, device, and medium for detecting traffic anomalies based on unmanned aerial vehicles (UAVs), to solve the problem in the prior art where the fixed flight trajectory of UAVs on traffic roads results in a fixed shooting position for the UAVs, leading to blind spots in the onboard camera's shooting and obstruction by obstacles, resulting in poor monitoring effect of UAV inspections for traffic anomalies.

[0004] In a first aspect, the present invention provides a traffic anomaly detection method based on unmanned aerial vehicles (UAVs), the UAV-based traffic anomaly detection method comprising: Step 100: Obtain the geographic coordinates, final event type label, timestamp, and preset credibility of the abnormal event occurring on the traffic road; Step 200: Based on the geographical coordinates of the abnormal event, determine the geographical coordinates, ground height, altitude, and status label of the obstacles affecting the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event. Step 300: Based on the geographical coordinates of the obstacle, the ground elevation of the obstacle's location, the obstacle's own height, and the geographical coordinates of the abnormal event, determine the first flight path of the drone, and control the drone to fly to the preset position according to the first flight path; Step 400: Based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, obtain the feature set of the abnormal event, and calculate the completeness of the abnormal event based on the feature set of the abnormal event and the preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and traffic density. When the completeness meets the preset conditions, output the current feature set of the abnormal event. Step 500: When the completeness does not meet the preset condition, determine the target position of the drone based on the completeness of the abnormal event, control the drone to fly to the target position, take the target position as the preset position, and return to step 400 above.

[0005] Secondly, the present invention provides a traffic anomaly detection device based on unmanned aerial vehicles (UAVs), the UAV-based traffic anomaly detection device comprising: The abnormal event information acquisition module is used to acquire the geographical coordinates, final event type label, timestamp, and preset credibility of abnormal events occurring on traffic roads; The obstacle information acquisition module is used to determine, based on the geographical coordinates of the abnormal event, the geographical coordinates of the obstacles that affect the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event, the ground height of their location, their own height, and their own status labels. The first flight path determination module is used to determine the first flight path of the UAV based on the geographical coordinates of the obstacle, the ground height of the obstacle's location, the height of the obstacle itself, and the geographical coordinates of the abnormal event, and to control the UAV to fly to a preset position according to the first flight path. The abnormal event feature set output module is used to obtain the feature set of the abnormal event based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, and to calculate the completeness of the abnormal event based on the feature set of the abnormal event and a preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and the traffic density. When the completeness meets the preset conditions, the current feature set of the abnormal event is output. The target location determination module is used to determine the target location of the UAV based on the completeness of the abnormal event when the completeness does not meet the preset condition, control the UAV to fly to the target location, use the target location as the preset location, and return the feature set output module of the above-mentioned abnormal event.

[0006] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the traffic anomaly detection method based on unmanned aerial vehicles as described in the first aspect.

[0007] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the traffic anomaly detection method based on unmanned aerial vehicles as described in the first aspect.

[0008] The aforementioned UAV-based traffic anomaly detection method, device, equipment, and medium acquire the geographic coordinates of an anomaly occurring on a traffic road. They then determine the geographic coordinates, ground altitude, altitude, and status label of obstacles affecting the UAV's flight altitude within a preset area of ​​the anomaly's geographic coordinates, providing data support for subsequent initial flight path planning. Based on the obstacle's geographic coordinates, ground altitude, altitude, status label, and the anomaly's geographic coordinates, the UAV's initial flight path is determined, and the UAV is controlled to fly along this path to a preset location. By planning an optimal flight path that satisfies safety constraints, the system effectively avoids... By eliminating obstacle interference, the risk of drone collisions is reduced, ensuring the safety and stability of drone inspection operations. Based on initial images of abnormal events captured by the drone at a preset altitude and location, a feature set of the abnormal event is obtained. The completeness of the abnormal event is calculated by comparing this feature set with a preset expected feature set. The drone's preset position is then redefined based on the completeness of the abnormal event. When the completeness meets preset conditions, the drone's current position remains unchanged, and the current feature set of the abnormal event is output. When the completeness does not meet preset conditions, the drone's preset position is dynamically adjusted, and data is re-captured, improving the accuracy and completeness of abnormal event detection. Compared to existing technologies, this invention employs a closed-loop iterative mechanism encompassing abnormal event information acquisition, obstacle information acquisition, flight path planning, abnormal event image acquisition, abnormal event completeness determination, and dynamic adjustment of the drone's shooting position. This mechanism enables the drone to adaptively adjust its shooting position according to the actual development trend of the abnormal event, eliminating the impact of obstacle obstruction and blind spots in the onboard camera's shooting, and improving the monitoring effectiveness of drone inspections for traffic anomalies. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of an application environment for the traffic anomaly detection method based on unmanned aerial vehicles (UAVs) in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a traffic anomaly detection method based on unmanned aerial vehicles (UAVs) in Embodiment 1 of the present invention; Figure 3 This is a diagram of an apparatus for a traffic anomaly detection method based on unmanned aerial vehicles (UAVs) in Embodiment 8 of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device provided in Embodiment 9 of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) provided in this invention can be applied to, for example... Figure 1 The application environment is shown. Specifically, this UAV-based traffic anomaly detection method is applied in a traffic anomaly detection system, which includes, as shown in the example... Figure 1 The diagram shows a drone and controller. The drone communicates with the controller via a network to achieve real-time traffic anomaly detection. The drone is an unmanned aerial vehicle equipped with a high-resolution visible light camera, a high-precision positioning module, and a communication unit. It executes tasks such as following the flight path planned by the controller, adjusting the shooting angle and gimbal parameters, acquiring images of the target area, and performing environmental perception of obstacles, transmitting the acquired data to the controller. The controller can be implemented as a standalone controller or a controller cluster composed of multiple controllers. It receives specific traffic anomaly detection task descriptions from the user and, based on the data transmitted by the drone, performs tasks such as anomaly event information fusion, environmental obstacle perception modeling, event completeness calculation and judgment, drone flight path planning, and drone position optimization.

[0013] In Example 1, as Figure 2 As shown, this embodiment provides a traffic anomaly detection method based on unmanned aerial vehicles (UAVs), which is applied to... Figure 1 Taking the controller in the example, the traffic anomaly detection method based on UAVs includes: Step 100: Obtain the geographic coordinates, final event type label, timestamp, and preset credibility of the abnormal event occurring on the traffic road; Here, "abnormal event" refers to a target event occurring on a traffic road to be monitored, identified by the drone based on a specific task description given by the user (e.g., detecting abnormal parking events on highways, identifying fire events on traffic roads, identifying smoke evaporation events on traffic roads, or monitoring pedestrians entering traffic lanes). The geographic coordinates of the abnormal event refer to the latitude and longitude coordinates of the location where the abnormal event occurred, used to accurately pinpoint the scene. The final event type label refers to the standardized final classification identifier for the abnormal event (e.g., accident, fire, congestion, pedestrian intrusion, abnormal parking, etc.), used to characterize the type of abnormal event. The timestamp refers to the standard time when the abnormal event was first detected by the system or reported by the user, used to record the start time of the abnormal event. The preset credibility refers to a pre-set evaluation value of the reliability of the information source for the abnormal event, ranging from 0 to 1: the higher the credibility value, the higher the reliability of the information source for the abnormal event; the lower the credibility value, the lower the reliability of the information source for the abnormal event.

[0014] In this embodiment, the controller synchronously acquires multi-source initial information of abnormal events occurring on the traffic road to be monitored from multiple third-party channels (such as pedestrian alarms and fixed camera data collection). The acquired multi-source initial information is then processed for deduplication, noise reduction, information extraction, and standardization to obtain the geographical coordinates of the abnormal event, the final event type label, the timestamp, and the preset credibility.

[0015] Step 200: Based on the geographical coordinates of the abnormal event, determine the geographical coordinates, ground height, altitude, and status label of the obstacles affecting the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event. The preset area refers to the geographical area within a pre-defined detection radius of the abnormal event, centered on its geographical coordinates. An obstacle is an object whose height is sufficient to limit the minimum safe flight altitude of the drone (e.g., static obstacles such as overpasses, tall buildings, high-voltage lines, towers, lampposts, or large billboards, as well as dynamic obstacles such as vehicles or crowds). The geographical coordinates of an obstacle refer to the latitude and longitude coordinates of its center point, reference point, or highest point. The ground elevation refers to the absolute altitude (in meters) of the ground where the obstacle's base is located. The obstacle's height (in meters) refers to the vertical height (in meters) of the obstacle extending from its base (ground) to its top. The obstacle's status label is a classification identifier for its static or dynamic attributes: static obstacles include fixed facilities such as buildings, overpasses, high-voltage lines, and lampposts; dynamic obstacles include moving objects such as vehicles and crowds.

[0016] In this embodiment, the controller obtains information on obstacles affecting the flight altitude of the UAV within a preset detection radius of the geographic location coordinates of the abnormal event from a high-precision map or road network information (structured data related to road traffic networks in a geographic information system).

[0017] Step 300: Based on the geographical coordinates of the obstacle, the ground elevation of the obstacle's location, the obstacle's own height, and the geographical coordinates of the abnormal event, determine the first flight path of the drone, and control the drone to fly to the preset position according to the first flight path; The first flight path refers to the improved A * The algorithm plans an initial optimal 3D flight path in 3D space from the drone's current location to the geographic location of the anomalous event. This flight path consists of a series of path points (p1, p2, ..., p...). n The structure consists of p1, p1, and p1, where p1 is the starting position (current position) of the drone. n The preset location for the drone (the geographical location where the abnormal event occurred: p) n =p0). The preset location refers to the observation point where the UAV hovers after flying along the first flight path to the destination (the geographical location where the abnormal event occurred) in order to collect images of the abnormal event.

[0018] In this embodiment, the improved A *An algorithm is a heuristic path search algorithm used to plan the flight path of a UAV, which minimizes the estimated total cost (the sum of the actual cost from the starting position to the intermediate position and the heuristically estimated cost from the intermediate position to the preset position) of the first flight path.

[0019] Step 400: Based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, obtain the feature set of the abnormal event, and calculate the completeness of the abnormal event based on the feature set of the abnormal event and the preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and traffic density. When the completeness meets the preset conditions, output the current feature set of the abnormal event. The preset altitude refers to the standard altitude (in meters) at which the drone hovers to collect images of the abnormal event, pre-set based on the final event type label. The initial image of the abnormal event refers to the original image (single image or sequence of frames) collected by the drone at a preset position, preset altitude, and initial gimbal angle, using an onboard camera (such as a high-definition visible light camera), covering the entire area where the abnormal event occurred. The feature set of the abnormal event refers to the set of features extracted from the initial image of the abnormal event after target detection and image analysis, including: number of injured persons, number of vehicles involved, traffic density, fire indicator (value 1 if there is a fire, value 0 if there is no fire), and smoke indicator (value 1 if there is smoke, value 0 if there is no smoke). The preset expected feature set refers to the standard feature set pre-set based on the final event type label, which is expected to be detected for the abnormal event, used to evaluate the completeness of the information collected about the abnormal event. The completeness of the abnormal event refers to an index used to quantify the degree of matching between the currently extracted features of the abnormal event and the expected features, used to characterize the comprehensiveness of the information collected about the abnormal event. The preset condition refers to the preset integrity threshold.

[0020] In this embodiment, when the final event type label is a fire event or an explosion event, the preset height is 60m; when the final event type label is a major traffic accident (such as multi-vehicle collision and personal injury), the preset height is 50m; when the final event type label is a traffic jam event, the preset height is 50m; and when the final event type label is an abnormal parking event or a pedestrian entering an accident scene event, the preset height is 40m.

[0021] In this embodiment, when the calculated completeness of the abnormal event is greater than or equal to the preset completeness threshold, there is no need to adjust the current preset position of the drone, and the feature set of the current abnormal event is output.

[0022] Step 500: When the completeness does not meet the preset condition, determine the target position of the drone based on the completeness of the abnormal event, control the drone to fly to the target position, take the target position as the preset position, and return to step 400 above.

[0023] The target location of the UAV refers to the three-dimensional spatial coordinates of the new observation point calculated based on the current completeness. By supplementing the acquisition of new images of the abnormal event at the target location, the completeness of the abnormal event is maximized.

[0024] In this embodiment, when the calculated completeness of the abnormal event is less than the preset completeness threshold, the drone's preset position is adjusted based on the current surrounding information of the abnormal event (such as crowds, traffic jams, ambulance locations, etc.), and the drone returns to step 400 to recalculate the completeness of the abnormal event until the completeness of the abnormal event is greater than or equal to the preset completeness threshold.

[0025] This embodiment of the UAV-based traffic anomaly detection method acquires the geographic coordinates of an anomaly occurring on a traffic road, determines the geographic coordinates, ground altitude, altitude, and status label of obstacles affecting the UAV's flight altitude within a preset area of ​​the anomaly's geographic coordinates, providing data support for subsequent first flight path planning. Based on the obstacle's geographic coordinates, ground altitude, altitude, status label, and the anomaly's geographic coordinates, the method determines the UAV's first flight path and controls the UAV to fly to a preset location along this path. By planning an optimal flight path that meets safety constraints, the method effectively avoids obstacles. Obstacles and interference are eliminated, reducing the risk of drone collisions and ensuring the safety and stability of drone inspection operations. Based on initial images of abnormal events captured by the drone at a preset altitude and location, a feature set of the abnormal event is obtained. The completeness of the abnormal event is calculated by comparing this feature set with a preset expected feature set. The drone's preset position is then redefined based on the completeness of the abnormal event. If the completeness meets preset conditions, the drone's current position remains unchanged, and the current feature set of the abnormal event is output. If the completeness does not meet preset conditions, the drone's preset position is dynamically adjusted, and data is re-captured, improving the accuracy and completeness of abnormal event detection. Compared to existing technologies, this invention employs a closed-loop iterative mechanism encompassing abnormal event information acquisition, obstacle information acquisition, flight path planning, abnormal event image acquisition, abnormal event completeness determination, and dynamic adjustment of the drone's shooting position. This mechanism enables the drone to adaptively adjust its shooting position according to the actual development trend of the abnormal event, eliminating the effects of obstacle obstruction and blind spots in the onboard camera's shooting, and improving the monitoring effectiveness of drone inspections for traffic anomalies.

[0026] In Embodiment 2, step 100 includes: Step 101: Obtain initial information of the abnormal event occurring on the traffic road based on a preset first channel, a preset second channel, a preset third channel, and a preset fourth channel, respectively. The initial information includes first initial information obtained based on the first channel, second initial information obtained based on the second channel, third initial information obtained based on the third channel, and fourth initial information obtained based on the fourth channel. The four preset channels refer to the traffic management center, fixed surveillance cameras on roads, pedestrian alarm terminals, and social media channels, respectively, which are pre-defined channels for obtaining initial information on abnormal events. The first initial information refers to the initial data information of abnormal events occurring on roads reported in real time by the traffic management center. The second initial information refers to the initial information of abnormal events occurring on roads captured by fixed surveillance cameras. The third initial information refers to the initial data information of abnormal events occurring on roads uploaded by pedestrian alarm terminals. The fourth initial information refers to the initial information of abnormal events occurring on roads publicly appealed for help on social media.

[0027] Step 102: Perform deduplication, noise reduction, information extraction, and standardization on the first initial information, the second initial information, the third initial information, and the fourth initial information respectively to obtain the geographic location coordinates, initial event type label, timestamp, and preset credibility of the abnormal event. The initial event type label includes a first event type label obtained based on the first initial information, a second event type label obtained based on the second initial information, a third event type label obtained based on the third initial information, and a fourth event type label obtained based on the fourth initial information. The process involves several steps: Deduplication, Noise Reduction, and Initial Event Type Labeling. Deduplication involves removing duplicate reports of the same anomalous event from the first, second, third, and fourth initial information sources. Noise Reduction involves filtering background noise unrelated to the anomalous event from the first, second, third, and fourth initial information sources. Information Extraction involves extracting core anomalous event elements from the data obtained after deduplication and noise reduction. Standardization involves unifying the core anomalous event elements with different formats and expressions obtained after information extraction into a fixed data structure that the system can recognize. Initial Event Type Labels are preliminary extractions of event category identifiers (e.g., accident events, fire events, congestion events, pedestrian intrusion events, or abnormal parking events) from information obtained from various channels after deduplication, noise reduction, information extraction, and standardization. These labels include first, second, third, and fourth event type labels.

[0028] In this embodiment, the geographical coordinates of the abnormal event are latitude and longitude coordinates: P0=(x0,y0), where P0 is the geographical coordinate of the abnormal event, x0 is the longitude coordinate of the abnormal event, and y0 is the latitude coordinate of the abnormal event. The precision of both x0 and y0 is retained to 6 decimal places. The timestamp is T0, in the format of year-month-day, hour:minute:second. The preset confidence level is C0. The preset confidence level for the first channel is 0.9, for the second channel it is 0.8, for the third channel it is 0.6, and for the fourth channel it is 0.6.

[0029] Step 103: When the first event type label, the second event type label, the third event type label, and the fourth event type label are consistent, the initial event type label is matched with a preset set of event type labels to obtain the final event type label; The preset event type label set refers to the standard abnormal event category library pre-defined by the system.

[0030] In this embodiment, when the first event type label, the second event type label, the third event type label, and the fourth event type label are the same (do not conflict), the initial event type label is matched with the preset event type label set to obtain the final event type label.

[0031] Step 104: When the first event type label, the second event type label, the third event type label, and the fourth event type label are inconsistent, the first event type label, the second event type label, the third event type label, and the fourth event type label are weighted and summed according to the preset confidence level to obtain the final event type label.

[0032] In this embodiment, the formula for obtaining the maximum value of the weighted sum of the final event type labels is: i = 1, 2, 3, 4 Where L is the final event type label; l is all the labels in the preset event type label set; i is the channel number; s i This is the preset i-th channel. When i=1, it corresponds to the preset first channel; when i=2, it corresponds to the preset second channel; when i=3, it corresponds to the preset third channel; and when i=4, it corresponds to the preset fourth channel. The preset credibility level corresponding to the i-th channel: , , , ; δ is the indicator function (when When δ=1; otherwise δ=0). The initial event type label is the preset label for the i-th channel. When i=1, it corresponds to the first event type label; when i=2, it corresponds to the second event type label; when i=3, it corresponds to the third event type label; and when i=4, it corresponds to the fourth event type label.

[0033] This embodiment of the UAV-based traffic anomaly detection method simultaneously acquires initial information about anomalies through four preset channels: a first, a second, a third, and a fourth. This achieves mutual supplementation between official data, monitoring data, and publicly reported data, significantly improving the detection coverage of anomalies and effectively reducing missed and false detections caused by single-channel failures or delays. When the event type labels from all channels are consistent, they are directly matched with a preset set of event type labels to determine the final event type label, simplifying the anomaly detection process and improving the response speed of the detection system. When event type labels from different channels conflict, a weighted voting system based on credibility is used for decision-making, enhancing the reliability of the anomaly detection results. This lays the foundation for subsequent adjustments to the anomaly detection radius based on the final event type label, calculation of the anomaly completeness, and dynamic adjustment of the UAV's shooting position based on the anomaly completeness, thereby improving the monitoring effectiveness of UAV patrols for traffic anomalies.

[0034] In Embodiment 3, step 200 includes: Step 201: Based on the final event type label and the mapping relationship between the final event type label and the preset base radius, obtain the target base radius, which is used to quantify the preset area of ​​the geographic location coordinates of the abnormal event; The target base radius refers to the pre-defined perception range and detection radius of the initial abnormal event based on the final event type label, which is a preset area used to quantify the geographical coordinates of the abnormal event.

[0035] In this embodiment, the mapping relationship between the final event type label and the preset base radius is as follows: When the final event type label is a fire or explosion, the preset base radius (target base radius) is 500m (to adapt to the range of heat radiation and smoke diffusion); when the final event type label is a major traffic accident (such as multi-vehicle collisions and personal injuries), the preset base radius (target base radius) is 300m (to cover the accident scene and rescue access); when the final event type label is a traffic congestion, the preset base radius (target base radius) is 200m (to adapt to the needs of road queue length assessment); when the final event type label is an abnormal parking incident or a pedestrian entering an accident scene, the preset base radius (target base radius) is 100m (to cover only the local affected area); when the final event type label is ambiguous, the preset base radius (target base radius) is 500m (taking the maximum value of the preset base radius for each category to ensure complete coverage).

[0036] Step 202: Correct the target base radius according to the preset severity coefficient of the abnormal event, the preset credibility coefficient of the abnormal event, and the preset environmental complexity coefficient to obtain the corrected radius; The preset severity coefficient of the abnormal event refers to a pre-set correction factor used to quantify the severity of the abnormal event based on the number of injured and damaged vehicles initially reported by multiple third-party channels. The preset credibility coefficient of the abnormal event refers to a pre-set correction factor used to adjust the target's basic radius in reverse, based on the preset credibility of the abnormal event (the lower the preset credibility, the lower the reliability of the abnormal event information, and the more necessary it is to expand the detection range to verify the authenticity of the abnormal event and detect its full picture; the higher the preset credibility, the higher the reliability of the abnormal event information, and the more necessary it is to narrow the detection range to improve the system's detection and calculation efficiency and focus on the core area of ​​the abnormal event). The preset environmental complexity coefficient refers to a pre-set correction factor used to adjust the target's basic radius based on the building density and building height within the detection radius range of the UAV during flight.

[0037] In this embodiment, the calculation formula for correcting the target base radius is as follows: R t =R base ×α severity ×β confidence ×γ env Among them, R t R is the corrected radius; base α is the target base radius (preset base radius); severity α is a preset severity coefficient for abnormal events, ranging from 1.0 to 2.0 (α is set when either the number of injured or the number of damaged vehicles reported initially by multiple third-party channels is greater than or equal to 3). severity Take 2.0; when either the number of injured or the number of damaged vehicles in the initial feedback from multiple third-party channels is 1 or 2, α severity Take 1.5; when either the number of injured or the number of damaged vehicles in the initial feedback from multiple third-party channels regarding the abnormal event is 0, α severity (Take 1.0); β confidence β is the preset confidence coefficient for abnormal events, ranging from 0.8 to 1.5 (when the preset confidence is less than 0.5, β...). confidence Set β to 1.5; when the preset confidence level is greater than or equal to 0.5 and less than or equal to 0.9, β confidence Set to 1.0; when the preset confidence level is greater than 0.9, β confidence Take 0.8); γ env γ is a preset environmental complexity coefficient, ranging from 0.8 to 1.2 (in densely populated urban areas with a building density greater than or equal to 0.6, γ...). env Take 1.2; in open areas with a building density of less than 0.3, γ env Take 0.8; in the intermediate zone where the building density is greater than or equal to 0.3 and less than 0.6, γ env Take 1.0).

[0038] The environmental complexity coefficient is also affected by the building height within the corrected radius: when the building height within the corrected radius is low (less than or equal to a preset building height threshold, e.g., 15m), it has no impact on environmental complexity, i.e., the environmental complexity coefficient is 0; when the building height within the corrected radius is high (greater than the preset building height threshold, e.g., 15m), the formula for calculating the environmental complexity coefficient affected by the building height within the corrected radius is as follows: γ ’ env =1+(γ max -1)×(1-e -λP ), Where, γ’ env γ is the environmental complexity coefficient affected by the building height within the modified radius. max The maximum value of the preset environmental complexity coefficient (i.e., γ) max =1.2); e is the natural constant; λ is the preset sensitivity coefficient (e.g., λ=2.0); P is the building height influence factor within the corrected radius, calculated using the following formula: , where N ref The number of reference buildings within the preset corrected radius (e.g., N) ref =10), N is the actual number of buildings within the corrected radius, j is the index of the buildings within the corrected radius (ranging from 1 to N), h j eff The effective height is the j-th building within the corrected radius (when the building height within the corrected radius is greater than the preset building height threshold, the effective height is the difference between the building height and the building height threshold; when the building height within the corrected radius is less than or equal to the preset building height threshold, the effective height is 0), and β is a preset shape parameter, which takes the value of 1 (linear effect) or 2 (square magnification effect), preferably 1.

[0039] Step 203: Update the corrected radius based on the radius of the influence range of the abnormal event observed in real time by the UAV during flight to obtain the updated radius; Among them, the radius of the impact range of the abnormal event observed in real time refers to the maximum radius of the actual impact range of the abnormal event detected in real time by the UAV through the airborne camera or airborne sensor during flight, which is used to reflect the true degree of spread of the abnormal event.

[0040] In this embodiment, the UAV collects environmental information about abnormal events in real time during flight. If the affected area of ​​the observed abnormal event expands (e.g., smoke spreads or traffic congestion worsens), the radius is updated in real time. The formula for updating the corrected radius is as follows: R t+1 =max(R t ,ρ×R observed,extent，max ), Among them, R t+1 For the updated radius, R t R is the corrected radius, ρ is the preset redundancy coefficient (e.g., ρ=1.2), and R observed,extent,max The maximum radius of influence of anomalies observed in real time by the UAV during flight: when R observed,extent,max When R is greater than or equal to a preset threshold radius of 1500m for the impact range of the abnormal event,observed,extent,max =1500m, when R observed,extent,max When the radius of the abnormal event's impact range is less than the preset threshold of 1500m, R observed,extent,max The calculation formula is: R observed,extent,max =R base ×K severity ×K dynamic ×K spread , Among them, K severity The comprehensive severity coefficient for the number of injured and the vehicles damaged ranges from 1.0 to 2.5, and is calculated using the following formula: K severity =1+δ inj ×ln(1+N injured )+δ veh ×ln(1+N vehicles ), where δ inj The influence coefficient of the number of injured persons (e.g., δ) is a preset value. inj =0.2); δ veh The preset impact coefficient of the number of damaged vehicles (e.g., δ) veh =0.15); N injured N represents the number of injured; vehicles The number of damaged vehicles; when N injured Greater than or equal to 5 or N vehicles When K is greater than or equal to 5 severity It needs to be increased by 1.2 times.

[0041] K dynamic The dynamic propagation coefficient of the abnormal event ranges from 1.0 to 2.2, and is calculated using the following formula: K dynamic =1+0.5×fire+0.3×smoke+0.4×traffic, where fire is the preset fire propagation coefficient, which takes a value of 1 or 0 (1 when there is a fire; 0 when there is no fire); smoke is the preset smoke propagation coefficient, which takes a value of 1 or 0 (1 when there is smoke; 0 when there is no smoke); traffic is the preset traffic density influence coefficient, which takes a value of 0 to 1.

[0042] K spread The time propagation impact factor of the abnormal event is initially set to 1.0, and the calculation formula is as follows: , Where, k maxThe maximum radius expansion factor is set to 3.0 (meaning the maximum radius of the affected area does not exceed three times the base radius); η is the preset propagation rate coefficient (η=0.15 when the final event type label is a fire or explosion; η=0.08 when the final event type label is a major traffic accident; η=0.05 when the final event type label is a traffic congestion; η=0.03 when the final event type label is an abnormal parking incident or a pedestrian entering an accident scene; η=0.03 when the final event type label is ambiguous), T elapsed This is the difference between the current time and the timestamp of the abnormal event.

[0043] Step 204: Using the geographic coordinates of the abnormal event as the center and the updated radius as the range, perform information extraction processing on the abnormal event to obtain the geographic coordinates, ground height, height, and status label of each obstacle affecting the flight altitude of the UAV within the updated preset area of ​​the geographic coordinates of the abnormal event.

[0044] The updated preset area refers to the geographical area within the range centered on the geographical coordinates of the abnormal event and the updated detection radius of the abnormal event.

[0045] In this embodiment, the geographical coordinates of the obstacle are latitude and longitude coordinates: P m =(x m ,y m ), where P m x represents the geographical coordinates of the obstacle. m Here are the longitude coordinates of the highest point of the obstacle, y m Let z be the latitude coordinates of the highest point of the obstacle. The ground elevation at the obstacle's location is z. m The obstacle's height is h. m The obstacle's own status label is either a static obstacle or a dynamic obstacle.

[0046] This embodiment of the UAV-based traffic anomaly detection method matches the detection baseline radius of anomaly events based on the final event type label. It adaptively adjusts the baseline radius by considering three factors: the scale of the anomaly, the reliability of the information source, and the influence of the surrounding geographical environment. During the UAV's flight, the radius is updated in real time as the anomaly spreads, ensuring that the detection range always covers the actual situation at the scene of the anomaly, thus improving the accuracy of the system's perception of the surrounding environment. Using the updated radius as the detection range, information on surrounding obstacles is accurately extracted, providing data support for the UAV's path planning and improving the monitoring effectiveness of UAV inspections for traffic anomalies.

[0047] In Embodiment 4, step 300 includes: Step 301: Based on the geographical coordinates of the obstacle and the geographical coordinates of the abnormal event, obtain the preset path points, and calculate the length of the preset flight path based on the sum of the distances between all adjacent path points in the preset path points; Among them, the preset path points refer to the flight trajectory nodes on the preset flight path of a number of discrete UAVs that are planned in advance, taking into account the positional constraints of obstacles and the positional constraints of abnormal events.

[0048] In this embodiment, the formula for calculating the length of the preset flight path is: , Where length is the length of the preset flight path; k is the kth preset path point; n is the total number of preset path points; p k+1 Let p be the three-dimensional spatial coordinates of the (k+1)th path point. k+1 =(x k+1 ,y k+1 ,z k+1 ), x k+1 y k+1 and z k+1 These represent the instantaneous displacement components of the (k+1)th path point in the X, Y, and Z dimensions of the three-dimensional space; p k Let p be the three-dimensional spatial coordinates of the k-th path point. k =(x k ,y k ,z k ), x k y k and z k These are the instantaneous displacement components of the k-th path point in the X, Y, and Z dimensions of the three-dimensional space, respectively. The Euclidean distance between the (k+1)th path point and the kth path point is calculated using the following formula: .

[0049] Step 302: Calculate the safety risk value of the preset flight path based on the distance between the three-dimensional spatial coordinates of each preset path point and the three-dimensional spatial coordinates of each obstacle. In this context, the three-dimensional spatial coordinates of each obstacle refer to the coordinates in the X-axis of the three-dimensional space, based on the longitude coordinates of the obstacle's geographical location; the coordinates in the Y-axis of the three-dimensional space, based on the latitude coordinates of the obstacle's geographical location; and the coordinates in the Z-axis of the three-dimensional space, based on the sum of the ground height at the obstacle's location and the obstacle's own height. The safety risk value is a quantitative assessment indicator used to reflect the degree of risk of collision between the UAV and an obstacle when flying close to it along a preset flight path.

[0050] In this embodiment, the formula for calculating the safety risk value of the preset flight path is: , Where risk is the safety risk value of the preset flight path; O m Let O be the three-dimensional spatial coordinates of the m-th obstacle within the updated radius. m =(x m ,y m ,z m ), where m is the m-th obstacle within the updated radius, x m y m and z m Let X and Y be the coordinate components of the m-th obstacle within the updated radius in the three-dimensional space along the X (longitude), Y (latitude), and Z (height) dimensions, respectively, and let O be the set of three-dimensional spatial coordinates of all obstacles within the updated radius. The Euclidean distance between the k-th path point and the m-th obstacle is calculated using the following formula: .

[0051] Step 303: Use a preset first objective function to perform a weighted summation of the length of the preset flight path and the safety risk value of the preset flight path to obtain a first flight path of the UAV that satisfies the preset first constraint condition, and control the UAV to fly to the preset position according to the first flight path.

[0052] The preset first objective function refers to a flight path optimization function pre-constructed based on the preset flight path length and the preset flight path safety risk value, used to quantitatively evaluate the overall merits of each flight path in the path planning. The preset first constraint refers to the mandatory safety constraints that all waypoints on the path must satisfy when the UAV flies along the first flight path.

[0053] In this embodiment, the calculation formula for the first objective function is preset as follows: aim1=min(α×length+β×risk), Where aim1 is the preset first objective function, α is the weight coefficient of the preset flight path length (with a value of α=0.6), length is the length of the preset flight path, β is the safety risk weight coefficient of the preset flight path (with a value of β=0.4), and risk is the safety risk value of the preset flight path.

[0054] Furthermore, all path points must satisfy the following preset first constraint condition (taking path point p as an example). k (For example) zk >z ground (x k ,y k )+h max (x k ,y k )+Q, Where, x k y k and z k These represent the instantaneous displacement components of the k-th path point in the X, Y, and Z dimensions of the three-dimensional space, respectively, z. ground (x k ,y k ) represents path point p k The ground elevation at the location, h max (x k ,y k ) represents path point p k The maximum height of obstacles within the updated radius of the location, where Q is the preset safe height, with a value of 5m (to avoid collisions).

[0055] Furthermore, an initial path point sequence (p1, p2, ..., p) is generated based on the obtained first flight path. n ), where p1 is the starting position (current position) of the drone, p n The preset location for the drone (the geographical location where the abnormal event occurred: p) n =p0). To further optimize the flight path and avoid sharp turns or sudden speed changes at path inflection points, the system employs a B-spline curve fitting algorithm to smooth the initial path point sequence: fitting the discrete path points into a continuous, smooth three-dimensional spatial curve. This curve, while passing near each path point, ensures the continuity of its first derivative (velocity) and second derivative (acceleration), thereby ensuring stable attitude, free from jitter or sharp turns during actual flight.

[0056] This embodiment of the UAV-based traffic anomaly detection method combines the geographical coordinates of obstacles and the geographical coordinates of anomalies to set preset path points. By accumulating the distances between adjacent path points, the length of the preset flight path is calculated, achieving a quantitative representation of the flight path. Based on the distance between the three-dimensional spatial coordinates of each path point and the three-dimensional spatial coordinates of the obstacle, the safety risk value of the preset flight path is calculated, accurately quantifying the degree of risk of the UAV approaching and colliding with obstacles during flight. A preset first objective function is used to weight and sum the length of the preset flight path and the safety risk value of the preset flight path, and a preset first constraint is used to select the optimal first flight path, balancing the dual requirements of short flight path and high flight safety. By smoothing the UAV flight trajectory, the stability of the UAV flight is improved. This path planning method can autonomously complete flight path selection in complex multi-obstacle traffic scenarios, effectively avoiding flight collision risks, improving the flight safety and inspection efficiency of the UAV, and providing reliable flight trajectory assurance for subsequent point-to-point detection and integrity assessment of traffic anomalies.

[0057] In Example 5, step 400 includes: Step 401: Determine the preset height and the expected feature set based on the final event type label; In this embodiment, the preset altitude is 60m when the final event type label is a fire or explosion; 50m when the final event type label is a major traffic accident (e.g., multi-vehicle collision and personal injury); 50m when the final event type label is a traffic congestion; and 40m when the final event type label is an abnormal parking incident or a pedestrian entering an accident scene. After the drone flies to the preset position, the GPS positioning confirms that the deviation of the preset hovering altitude is less than or equal to 1m.

[0058] Step 402: Preprocess and extract features from the initial images of the abnormal event captured by the UAV at the preset altitude to obtain a feature set of the abnormal event; Preprocessing refers to basic image optimization and quality improvement operations such as denoising and image enhancement on the initial images of the acquired anomalous events. Feature extraction refers to the process of mining and filtering the preprocessed images of anomalous events based on object detection algorithms to extract key information that can characterize the attributes of the anomalous events, thus obtaining a feature set of the anomalous events.

[0059] In this embodiment, when the UAV collects initial high-definition images of the abnormal event at a preset altitude, the initial gimbal angle of its onboard camera is set to 0 degrees horizontally and -30 degrees vertically. The image acquisition resolution is set to 4K (3840×2160), and the acquisition frame rate is set to 10 frames per second, continuously acquiring 30 frames of images. This is to avoid the impact of blurry single-frame images on feature extraction. Preprocessing involves using a Gaussian filtering algorithm to denoise the acquired initial high-definition images, removing noise interference from the images; and using a histogram equalization algorithm to improve image contrast, optimize the image visual effect, and ensure that the images are clear and distinguishable. The YOLOv8 algorithm (a new generation single-stage object detection algorithm) is used to perform feature extraction processing on the preprocessed images to obtain the feature set of the abnormal event.

[0060] In this embodiment, the feature set of abnormal events is F. t :F t ={N injured N vehicles fire flag ,smoke flag traffic density}, Where, N injured N represents the number of injured persons within the updated radius (an integer, taking the value 0 if unidentifiable). vehicles For the updated radius number of vehicles involved (integer, including accident vehicles and congestion vehicles, etc.), fire flag The fire indicator (value 1 when there is a fire; value 0 when there is no fire), smoke flag For smoke indicators (value 1 when there is smoke; value 0 when there is no smoke), traffic density The updated traffic density within the radius (normalized value is 0 to 1: 0 when there are no vehicles; 1 when there is extreme congestion).

[0061] Step 403: Calculate the completeness of the abnormal event based on the ratio of the intersection of the feature set of the abnormal event and the preset expected feature set to the preset expected feature set. In this embodiment, the formula for calculating the completeness of an abnormal event is: , Where, η t F represents the completeness of the exception event (ranging from 0 to 1). t F is the feature set of abnormal events. expected For the expected feature set, This represents the number of intersections between the features of the abnormal events in the feature set of abnormal events and the expected features in the expected feature set. This represents the total number of expected features in the expected feature set.

[0062] Step 404: When the completeness of the abnormal event is greater than or equal to the preset completeness threshold, there is no need to adjust the preset position of the drone, and the images of the abnormal event are collected according to the preset monitoring duration and the preset image acquisition frequency. The images are preprocessed and feature extraction is performed, and the updated feature set of the abnormal event is output.

[0063] In this embodiment, the preset completeness threshold is a numerical range (ranging from 0.8 to 0.95) that is pre-set based on the final event type label and can be adjusted according to the urgency of the abnormal event: when the final event type label is an emergency event such as a fire, explosion, or major traffic accident (e.g., multi-vehicle collision and personal injury), the preset completeness threshold is 0.95 (it is necessary to fully obtain information about the abnormal event to support rescue decisions); when the final event type label is a regular event such as a congestion event, abnormal parking event, or pedestrian entering an accident scene, the preset completeness threshold is 0.8 (it is sufficient to meet the basic detection requirements of abnormal events).

[0064] In this embodiment, the preset monitoring duration is pre-set based on the final event type label: when the final event type label is an emergency event such as a fire, explosion, or major traffic accident (e.g., multi-vehicle collision and personal injury), the preset monitoring duration is 30 minutes; when the final event type label is a regular event such as a traffic jam, abnormal parking, or pedestrian entering an accident scene, the preset monitoring duration is 15 minutes.

[0065] In this embodiment, the preset image acquisition frequency is once every 30 seconds.

[0066] In this embodiment, when the completeness of the abnormal event is greater than or equal to the preset completeness threshold, there is no need to adjust the current preset position of the drone. Instead, the images of the abnormal event collected at the current preset position with a preset monitoring duration and a preset image acquisition frequency are preprocessed and feature extracted to output an updated feature set of the abnormal event.

[0067] This embodiment of the UAV-based traffic anomaly detection method determines the preset altitude and expected feature set for UAV capture based on the final event type label. It obtains the feature set of the anomaly event from the initial images of the anomaly event captured by the UAV at the preset altitude, and calculates the completeness of the anomaly event based on the feature set and the expected feature set. When the completeness of the anomaly event is greater than or equal to a preset completeness threshold, there is no need to adjust the preset position of the UAV. The images captured within a preset monitoring period at a preset image acquisition frequency are then processed to obtain an updated feature set of the anomaly event. By ensuring that the completeness of the anomaly event meets the requirements, the comprehensiveness of traffic anomaly detection and the monitoring effect of UAV patrols on traffic anomalies are improved.

[0068] In Embodiment Six, step 500, determining the target location of the UAV, includes: Step 501: When the completeness of the abnormal event is less than a preset completeness threshold, obtain the three-dimensional spatial coordinates of the current location of the drone, the remaining battery power of the drone, the crowd density at the current location, and the vehicle congestion index at the current location. The current location's population density refers to the number of people per unit area within the updated detection radius of the drone's current location, used to quantify the density of population gathering at the current location. The current location's vehicle congestion index refers to the number of vehicles per unit area within the updated detection radius of the drone's current location, used to quantify the degree of road traffic congestion at the current location.

[0069] Step 502: Determine the initial information gain urgency coefficient, the initial displacement cost tolerance coefficient, and the initial security risk sensitivity coefficient based on the final event type label; The initial information gain urgency coefficient is a quantitative coefficient, pre-defined based on the final event type label, used to characterize the urgency of supplementing the collection of information about abnormal events to improve the completeness of the abnormal events. The initial displacement cost tolerance coefficient is a quantitative coefficient, pre-defined based on the final event type label, used to measure the tolerance for costs such as displacement distance and energy consumption when the UAV moves its position to supplement the collection of information about abnormal events. The initial safety risk sensitivity coefficient is a quantitative coefficient, pre-defined based on the final event type label, used to characterize the sensitivity to flight safety risks during the UAV's position adjustment process.

[0070] In this embodiment, when the final event type label is a fire event or an explosion event, the initial information gain urgency coefficient γ1 = 0.3, the initial displacement cost tolerance coefficient γ2 = 0.2, and the initial safety risk sensitivity coefficient γ3 = 0.5 (prioritizing safety); when the final event type label is a congestion event, an abnormal parking event, or a pedestrian entering an accident scene event, the initial information gain urgency coefficient γ1 = 0.6, the initial displacement cost tolerance coefficient γ2 = 0.2, and the initial safety risk sensitivity coefficient γ3 = 0.2 (prioritizing improving information gain); when the final event type label is a major traffic accident or other event requiring high-precision evidence collection of accident details, the initial information gain urgency coefficient γ1 = 0.3, the initial displacement cost tolerance coefficient γ2 = 0.5, and the initial safety risk sensitivity coefficient γ3 = 0.2 (prioritizing controlling displacement cost).

[0071] Step 503: Calculate the current safety estimate of the drone based on the distance between the three-dimensional spatial coordinates of the drone's current location and the three-dimensional spatial coordinates of each obstacle, the crowd density at the current location, and the vehicle congestion index at the current location. The current safety assessment is a comprehensive quantitative evaluation of the flight safety level of the drone at its current location.

[0072] In this embodiment, the formula for calculating the current security assessment of the drone is: , Among them, safety score Current safety assessment of drones; O m Let O be the three-dimensional spatial coordinates of the m-th obstacle within the updated radius. m =(x m ,y m ,z m ), where m is the m-th obstacle within the updated radius, x m y m and z m Let be the coordinate components of the m-th obstacle within the updated radius in the X, Y, and Z dimensions of 3D space, respectively; and O be the set of 3D spatial coordinates of all obstacles within the updated radius. cur p represents the three-dimensional spatial coordinates of the drone's current position. cur =(x cur ,y cur ,z cur ), x cur y cur and z cur These are the instantaneous displacement components of the UAV's current position in the X, Y, and Z dimensions of three-dimensional space, respectively. Let be the Euclidean distance between the current position of the drone and the m-th obstacle, calculated using the following formula: ;d ref The preset reference safety distance is set at 10m; λ dynamic The preset risk weight is 0.5; D(p) cur ) is the dynamic risk comprehensive value of the drone's current location, with a value greater than 0 and less than or equal to 3.

[0073] D(p cur The formula for calculating D(p) is: cur )=ρ crowd (p cur )+τ traffic (p cur )+fire spread (p cur ) Wherein, D(p) cur ρ represents the dynamic risk composite value of the drone's current location. crowd (p cur τ represents the population density at the drone's current location. traffic (p cur () represents the vehicle congestion index at the drone's current location, fire spread (p cur ) represents the probability that the fire will spread to the drone's current location (calculated only in the event of a fire, otherwise 0).

[0074] Step 504: Based on the difference between the preset integrity threshold and the integrity of the abnormal event, the initial information gain urgency coefficient is corrected to obtain the corrected information gain urgency coefficient; based on the remaining battery power of the UAV, the initial displacement cost tolerance coefficient is corrected to obtain the corrected displacement cost tolerance coefficient; based on the current security estimate, the initial security risk sensitivity coefficient is corrected to obtain the corrected security risk sensitivity coefficient. In this embodiment, the calculation formula for correcting the initial information gain urgency coefficient γ1 is as follows: γ1 ’ =γ1×[1+μ×(θ complete -η t )], Where, γ1 ’ Here, γ1 is the initial information gain urgency coefficient, μ is the preset gain sensitivity coefficient (with a value of 1.0), and θ is the adjusted information gain urgency coefficient. complete η is the preset integrity threshold (ranging from 0.8 to 0.95). tThis represents the completeness of the abnormal event. While correcting the initial information gain urgency coefficient γ1, γ2 and γ3 are proportionally reduced to ensure that γ1 + γ2 + γ3 = 1.

[0075] In this embodiment, the calculation formula for correcting the initial displacement cost tolerance coefficient γ2 is as follows: γ2 ’ =γ2×[1+ν×(1-b)], Among them, γ2 ’ Let γ2 be the initial displacement cost tolerance coefficient, ν be the preset power sensitivity coefficient (valued at 0.8), and b be the remaining power of the drone. While correcting the initial displacement cost tolerance coefficient γ2, γ1 and γ3 are proportionally reduced to ensure γ1 + γ2 + γ3 = 1. If the remaining power of the drone is less than 0.3, the initial displacement cost tolerance coefficient is increased.

[0076] In this embodiment, the calculation formula for correcting the initial security risk sensitivity coefficient γ3 is as follows: γ3 ’ =γ3×[1+ε×(1-safety score ) Among them, γ3 ’ γ3 is the corrected initial safety risk sensitivity coefficient, ε is the preset adjustment coefficient (with a value of 0.5), and safety is the initial safety risk sensitivity coefficient. score The current security assessment is calculated. While correcting the initial security risk sensitivity coefficient γ3, γ1 and γ2 are proportionally reduced to ensure that γ1 + γ2 + γ3 = 1. If the current security assessment is less than 0.7, the initial security risk sensitivity coefficient is increased.

[0077] Step 505: Normalize the modified information gain urgency coefficient, the modified displacement cost tolerance coefficient, and the modified safety risk sensitivity coefficient respectively to obtain the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized safety risk sensitivity coefficient. In this embodiment, the calculation formula for normalizing the corrected information gain urgency coefficient is as follows: , In this embodiment, the calculation formula for normalizing the corrected displacement cost tolerance coefficient is as follows: , In this embodiment, the calculation formula for normalizing the modified security risk sensitivity coefficient is as follows: , Where, γ 1,final γ 2,final and γ 3,final These are the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized security risk sensitivity coefficient, respectively. ’ γ2 ’ and γ3 ’ These are the corrected information gain urgency coefficient, the corrected initial displacement cost tolerance coefficient, and the corrected initial security risk sensitivity coefficient, respectively.

[0078] Step 506: Determine each candidate location based on the three-dimensional spatial coordinates of the current location of the UAV and the geographical coordinates of the abnormal event; Among them, candidate locations refer to alternative hovering and shooting positions that can be adjusted by the drone to supplement the collection of images of abnormal events and improve the completeness of the abnormal events.

[0079] In this embodiment, the three-dimensional spatial coordinates of the current position of the UAV are taken as the center, and a range of 50m is taken to the left and right in the horizontal X dimension, a range of 50m is taken to the left and right in the horizontal Y dimension, and a range of 20m is taken to the top and bottom in the vertical Z dimension. The sampling step size is set based on the building density (5m in dense urban areas and 10m in open areas) to obtain discrete three-dimensional spatial sampling grid points. The geographical coordinates of the abnormal event are taken as the center, and a detection radius of 30m or 60m is set. The azimuth angle is 0 degrees to 360 degrees (sampling step size is set to 45 degrees), and the pitch angle is -45 degrees to -15 degrees (sampling step size is set to 15 degrees). The discrete three-dimensional spatial sampling grid points are initially screened to obtain various initial candidate points. Candidate points that meet the physical constraints such as the UAV's flight altitude being greater than or equal to the preset safe flight altitude, the initial candidate point not being in a no-fly zone, and the distance between the initial candidate point and the obstacle being greater than or equal to 5m are selected to obtain various candidate positions (the number is less than or equal to 20 to avoid redundant calculations).

[0080] Step 507: Calculate the expected integrity improvement value for each candidate location based on the image projection area of ​​the target detection region of the abnormal event at each candidate location, the image projection area of ​​the target detection region at the current position of the UAV, and the obstacle occlusion ratio at each candidate location; calculate the displacement cost value for each candidate location based on the three-dimensional spatial coordinates of each candidate location and the three-dimensional spatial coordinates of the current position of the UAV; calculate the safety risk estimate for each candidate location based on the three-dimensional spatial coordinates of each obstacle, the three-dimensional spatial coordinates of each candidate location, the crowd density at each candidate location, and the vehicle congestion index at each candidate location. The target detection area refers to the pre-defined key targets at the scene containing anomalous events (such as accident vehicles, injured persons, and fire points) and key targets in the surrounding area (such as congested vehicles), as well as the ground area that needs to be fully detected. The expected completeness improvement value refers to the expected increase in the completeness of anomalous events that can be obtained after the UAV moves to each candidate location. The displacement cost refers to the flight distance required for the UAV to move from its current location to each candidate location. The safety risk assessment refers to the safety risk value of the UAV when hovering and conducting inspections at each candidate location.

[0081] In this embodiment, the formula for calculating the expected integrity improvement value is: , Where Δη(p) is the expected integrity improvement value of the p-th candidate position; σ is the normalization function used to map the calculation result to the interval [0,1]; A proj (p) represents the image projection area of ​​the target detection region of the abnormal event at the p-th candidate location; A ref is the image projection area (reference image projection area) of the target detection region of the abnormal event at the current position of the UAV; occlusion(p) is the obstacle occlusion ratio at the p-th candidate position, which is calculated by the ratio of the area of ​​the target detection region of the abnormal event at the p-th candidate position that is occluded by obstacles to the image projection area of ​​the target detection region of the abnormal event at the p-th candidate position, and the value range is 0 to 1: occlusion(p)=0 means that there is no obstacle occlusion at the p-th candidate position, and occlusion(p)=1 means that the target detection region of the abnormal event is completely occluded by obstacles at the p-th candidate position.

[0082] In this embodiment, the formula for calculating the displacement cost value is: , in, Let p be the displacement cost of the p-th candidate position (the Euclidean distance between the p-th candidate position and the current position of the UAV); p Let p be the three-dimensional spatial coordinates of the p-th candidate position. p =(x p ,y p ,z p ), x p y p and z p Let p represent the instantaneous displacement components of the p-th candidate position in the X, Y, and Z dimensions of the three-dimensional space; p cur p represents the three-dimensional spatial coordinates of the drone's current position. cur =(x cur ,y cur ,zcur ), x cur y cur and z cur These represent the instantaneous displacement components of the UAV's current position in the X, Y, and Z dimensions of three-dimensional space.

[0083] In this embodiment, the formula for calculating the security risk assessment is as follows: , Among them, safety risk (p) represents the security risk estimate for the p-th candidate position; O m Let O be the three-dimensional spatial coordinates of the m-th obstacle within the updated radius. m =(x m ,y m ,z m ), where m is the m-th obstacle within the updated radius, x m y m and z m Let be the coordinate components of the m-th obstacle within the updated radius in the X, Y, and Z dimensions of 3D space, respectively; and O be the set of 3D spatial coordinates of all obstacles within the updated radius. p Let p be the three-dimensional spatial coordinates of the p-th candidate position. p =(x p ,y p ,z p ), x p y p and z p These are the instantaneous displacement components of the p-th candidate position in the X, Y, and Z dimensions of the three-dimensional space, respectively. λ is the Euclidean distance between the p-th candidate position and the m-th obstacle; dynamic The preset risk weight is 0.5; dynamic risk (p) represents the dynamic risk composite value of the p-th candidate position.

[0084] dynamic risk The formula for calculating (p) is: dynamic risk (p)=ρ crowd (p)+τ traffic (p)+fire spread (p) Where, dynamic risk (p) represents the dynamic risk composite value of the p-th candidate position, ρ crowd (p) represents the population density at the p-th candidate location, τ traffic (p) represents the vehicle congestion index for the p-th candidate location, firespread (p) is the probability that the fire will spread to the p-th candidate location (calculated only in the case of a fire, otherwise 0).

[0085] Step 508: Perform a weighted summation on the expected integrity improvement value of each candidate position and its corresponding normalized information gain urgency coefficient, the displacement cost value of each candidate position and its corresponding normalized displacement cost tolerance coefficient, and the security risk estimate of each candidate position and its corresponding normalized security risk sensitivity coefficient to obtain the comprehensive evaluation value of each candidate position. In this embodiment, the formula for calculating the comprehensive evaluation value is: , Where U(p) is the comprehensive evaluation value of the p-th candidate position, γ 1,final γ 2,final and γ 3,final These are the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized security risk sensitivity coefficient, respectively. Δη(p) is the expected integrity improvement value for the p-th candidate position. Let the displacement cost of the p-th candidate position (the Euclidean distance between the p-th candidate position and the current position of the UAV) be _safety_ risk (p) represents the security risk estimate for the p-th candidate position.

[0086] Step 509: Determine the target location of the UAV based on the comprehensive evaluation value of each candidate location.

[0087] In this embodiment, if there is a maximum positive value among the comprehensive evaluation values ​​of each candidate position, the candidate position with the largest comprehensive evaluation value is selected as the target position of the UAV; if the comprehensive evaluation values ​​of all candidate positions are negative, the current position of the UAV is maintained and used as the target position of the UAV.

[0088] The UAV-based traffic anomaly detection method in this embodiment first synchronously collects the three-dimensional spatial coordinates of the UAV's current location, the UAV's remaining battery power, the crowd density at the current location, and the vehicle congestion index when the completeness of an abnormal event is less than a preset completeness threshold. Then, it initializes the information gain urgency coefficient, displacement cost tolerance coefficient, and safety risk sensitivity coefficient based on the final event type label. Combining the three-dimensional spatial distance between the UAV's current location and each obstacle, the crowd density, and the vehicle congestion index, it calculates the current safety estimate. Based on the completeness deviation, the UAV's remaining battery power, and the current safety estimate, it corrects and normalizes each initialized coefficient. Based on the UAV's current location and the geographical location of the abnormal event, it determines each candidate location. From the three dimensions of information gain, movement cost, and flight safety, it calculates the expected completeness improvement value, displacement cost value, and safety risk estimate of each candidate location. Then, it obtains the comprehensive evaluation value of each candidate location through weighted summation. Based on the comprehensive evaluation value, it selects the target location of the UAV. By simultaneously considering the information completion needs of abnormal events, the energy consumption cost of drone flight, and the flight safety of drones, it can adapt to complex multi-obstacle traffic scenarios, realize the quantitative selection of drone observation positions, improve the completeness of abnormal events, the safety and environmental adaptability of drone inspection operations, and provide optimal observation point guarantees for subsequent re-collection of abnormal event images and iterative evaluation of the completeness of abnormal events.

[0089] In Embodiment Seven, step 509 includes: Step 5091: When there is a maximum positive value among the comprehensive evaluation values ​​of each candidate position, the candidate position corresponding to the maximum positive value among the comprehensive evaluation values ​​is taken as the target position of the UAV. In this embodiment, when there is a maximum positive value among the comprehensive evaluation values ​​of each candidate position, the candidate position with the largest comprehensive evaluation value is selected as the target position of the UAV.

[0090] Step 5092: Determine the second flight path of the UAV based on the target location of the UAV; The second flight path refers to the flight trajectory of the UAV from its current location to the target location.

[0091] In this embodiment, the Minimum Snap trajectory generation method is used to plan a smooth flight trajectory from the current position to the target position, and the feasibility of the planned flight trajectory is verified: on the one hand, it must meet the dynamic constraints such as the maximum flight speed of the UAV being less than or equal to 15 meters per second and the maximum tilt angle of the UAV being less than or equal to 30 degrees; on the other hand, it must ensure that the current flight trajectory avoids all obstacles along the way, so as to obtain a second flight path that meets the above feasibility verification.

[0092] Step 5093: When the second flight path does not meet the preset flight conditions, verify in turn whether each of the remaining candidate positions meets the flight conditions until the flight path of the UAV that meets the flight conditions is obtained as the second flight path of the UAV, and the candidate position corresponding to the second flight path that meets the flight conditions is taken as the target position of the UAV. The preset flight conditions refer to dynamic constraints such as the drone's maximum flight speed being less than or equal to 15 meters per second and the drone's maximum tilt angle being less than or equal to 30 degrees, as well as constraints on the current flight trajectory to avoid all obstacles along the way.

[0093] In this embodiment, when the second flight path does not meet the preset flight conditions, the second-best candidate positions are selected in sequence to regenerate the trajectory and verify its feasibility until the optimal feasible flight trajectory that meets the preset flight conditions is selected and used as the second flight path. The candidate position corresponding to the second flight path that meets the preset flight conditions is used as the target position of the UAV.

[0094] Furthermore, the drone flies along the optimal feasible flight path to the target location. Using the target location as a preset position, it returns to step 400 above to recalculate the completeness of the abnormal event. Based on the recalculated completeness of the abnormal event, the next target location of the drone is determined. If, after three consecutive adjustments to the drone's position, the completeness of the abnormal event is still less than the preset completeness threshold, the system automatically triggers the abnormality reporting mechanism, sends the abnormal event information to the traffic management center, and controls the drone to return to the take-off and landing point.

[0095] In this embodiment, if a sudden dangerous situation is detected during the drone's position adjustment process (such as an obstacle suddenly approaching or an ambient wind speed greater than or equal to 8 meters per second), the adjustment of the drone's position is immediately interrupted, and an emergency obstacle avoidance procedure is executed (the drone is quickly raised to a preset safe altitude and then hovers to wait for the system's control command); when the drone's remaining battery power is less than 0.2, the adjustment of the drone's position is forcibly terminated, and the drone is controlled to return to the take-off and landing point along the original flight path.

[0096] Step 5094: When the comprehensive evaluation value of each candidate position is negative or it is determined that adjusting the position of the UAV will affect the safety of the preset area of ​​the abnormal event, there is no need to adjust the current position of the UAV. The field coverage of the target detection area of ​​the abnormal event and the pixel clarity of the UAV are weighted and summed using a preset second objective function to obtain the gimbal angle and camera focal length of the UAV that satisfy the preset second constraint. The current image of the abnormal event is re-acquired according to the gimbal angle and the camera focal length, and the current completeness of the abnormal event is calculated based on the current image of the abnormal event. The pre-defined second objective function refers to a pre-constructed shooting parameter optimization function that aims to maximize the field-of-view coverage of the target detection area of ​​the abnormal event and the image clarity (pixel clarity) of the UAV. It quantifies the shooting effect under the combination of the UAV's gimbal angle and camera focal length. Field-of-view coverage refers to the percentage of the target detection area of ​​the abnormal event covered by the airborne camera's image at the current gimbal angle and camera focal length (a value greater than or equal to 80%), used to quantify the completeness of the visualization of the target detection area of ​​the abnormal event. Image clarity refers to the clarity of the imaging details and pixel resolution quality (a value greater than or equal to 0.8) of the abnormal event information in the image captured by the UAV's airborne camera at the current gimbal angle and camera focal length, used to quantify the recognizability of the target detection area of ​​the abnormal event. The pre-defined second constraint refers to the mandatory safety constraints that the UAV must meet when shooting at the current gimbal angle and camera focal length, used to quantify whether the shooting angle avoids dangerous areas (such as fire sources or high-voltage lines).

[0097] In this embodiment, the calculation formula for the preset second objective function is: aim2=max θ,f [visibility(I θ,f )], Where aim2 is the preset second objective function, θ is the gimbal angle (ranging from 0 to 360 degrees in azimuth and from -60 to 0 degrees in pitch), f is the camera focal length (ranging from 10 mm to 50 mm), and visibility(I θ,f I is the image visibility estimate. θ,f This is the weighted sum of the field-of-view coverage of the target detection area for abnormal events and the image clarity of the UAV when the gimbal angle is θ and the camera focal length is f.

[0098] In this embodiment, the calculation formula for the preset second constraint condition is as follows: safety(p cur ,θ cur )≥safety(p safe ,θ safe ), Among them, safety(p cur ,θ cur p represents the safe value for the current position and the current gimbal angle. cur Let θ be the current position of the drone. cur Given the current gimbal angle of the drone, safety(p) safe ,θ safe ) represents the preset safe threshold (valued at 0.8) for the safe position and safe shooting gimbal angle.safe θ is the preset safe position. safe This is the preset safe shooting angle for the gimbal.

[0099] safety(p safe ,θ safe The formula for calculating ) is: , Where 'a' is the sequence number of the hazard source within the preset area of ​​the abnormal event (ranging from 1 to b), 'b' is the total number of hazard sources within the preset area of ​​the abnormal event, and ω a φ is the weighting coefficient of the a-th hazard source within the preset area of ​​the abnormal event. a γ is the angle (in radians) between the line of sight of the drone to the a-th hazard and the current optical axis of the camera. FOV D is the camera's half field of view (the larger of half the horizontal field of view and half the vertical field of view). ref p is the preset reference distance. a Let p be the three-dimensional spatial coordinates of the a-th hazard source within the preset area of ​​the abnormal event. cur Here are the three-dimensional spatial coordinates of the drone's current position. Let be the Euclidean distance between the current location of the drone and the a-th hazard source within the preset area of ​​the abnormal event, and δ be an exponential function (when φ a ≤γ FOV When the time condition is met, the value is 1; otherwise, it is 0.

[0100] In this embodiment, when the comprehensive evaluation value of each candidate position is negative or it is determined that adjusting the position of the drone will affect the safety of the preset area of ​​the abnormal event (e.g., near a crowd, fire source, or high-voltage line), there is no need to adjust the current position of the drone. Instead, by adjusting the gimbal angle and camera focal length of the drone, the field of view coverage of the target detection area of ​​the abnormal event and the image clarity (pixel clarity) of the drone are maximized to maximize the image visibility estimate, thus obtaining the optimal gimbal angle and camera focal length.

[0101] Furthermore, using the current angle as a reference, the UAV adjusts the azimuth angle by 45 degrees and the pitch angle by 15 degrees each time until it reaches the optimal gimbal angle that meets the preset second constraint. Using the current focal length (10 mm) as a reference, it adjusts it by 10 mm each time, gradually increasing it to the optimal camera focal length that meets the preset second constraint. Under the optimal gimbal angle and camera focal length, the UAV re-acquires the current image of the abnormal event and calculates the current completeness of the abnormal event based on the current image.

[0102] Step 5095: When the current completeness of the abnormal event is greater than or equal to the completeness threshold, there is no need to adjust the current position of the drone; when the current completeness of the abnormal event is less than the completeness threshold, return to step 501 to redetermine the target position of the drone until the event completeness calculated based on the target position of the drone is greater than or equal to the completeness threshold.

[0103] In this embodiment, when the current completeness of the abnormal event is greater than or equal to the completeness threshold, there is no need to adjust the current position of the drone. The drone collects the current image of the abnormal event at the current position, with a preset monitoring duration and a preset image acquisition frequency. When the current completeness of the abnormal event is less than the completeness threshold, the process returns to step 501 to redetermine the target position of the drone until the event completeness calculated based on the image collected from the target position of the drone is greater than or equal to the completeness threshold.

[0104] The UAV-based traffic anomaly detection method in this embodiment, when the comprehensive evaluation value of each candidate location has a maximum positive value, filters and verifies each candidate location according to preset flight conditions, selects the optimal target location for the UAV, and plans a feasible flight path. When the comprehensive evaluation value of each candidate location is negative or it is determined that adjusting the UAV's position would affect the safety of a preset area of ​​the anomaly event, there is no need to adjust the UAV's current position. Instead, the optimal gimbal angle and camera focal length that satisfy the preset second constraint conditions are solved through a preset second objective function. Under the optimal gimbal angle and camera focal length, images of the anomaly event are re-acquired, the completeness of the anomaly event is updated, and the completeness threshold is determined. This method can select the optimal observation point and match a safe and feasible flight path in scenarios where the UAV can move freely, and can also achieve high-quality on-site image acquisition by optimizing the gimbal angle and camera focal length in scenarios where position adjustment is not advisable. This effectively improves the scene adaptability, path planning capability, image acquisition quality, and anomaly event completeness detection accuracy of UAV traffic anomaly detection, while reducing unnecessary UAV maneuvers, saving flight energy consumption, and comprehensively ensuring the continuous, stable, safe, and reliable operation of UAV inspection operations.

[0105] In embodiment eight, a traffic anomaly detection device 30 based on an unmanned aerial vehicle (UAV) is provided, such as... Figure 3 As shown, it includes: The abnormal event information acquisition module is used to acquire the geographical coordinates, final event type label, timestamp, and preset credibility of abnormal events occurring on traffic roads; The obstacle information acquisition module is used to determine, based on the geographical coordinates of the abnormal event, the geographical coordinates of the obstacles that affect the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event, the ground height of their location, their own height, and their own status labels. The first flight path determination module is used to determine the first flight path of the UAV based on the geographical coordinates of the obstacle, the ground height of the obstacle's location, the height of the obstacle itself, and the geographical coordinates of the abnormal event, and to control the UAV to fly to a preset position according to the first flight path. The abnormal event feature set output module is used to obtain the feature set of the abnormal event based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, and to calculate the completeness of the abnormal event based on the feature set of the abnormal event and a preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and the traffic density. When the completeness meets the preset conditions, the current feature set of the abnormal event is output. The target location determination module is used to determine the target location of the UAV based on the completeness of the abnormal event when the completeness does not meet the preset condition, control the UAV to fly to the target location, use the target location as the preset location, and return the feature set output module of the above-mentioned abnormal event.

[0106] Optionally, the above-mentioned abnormal event information acquisition module includes: The initial information acquisition module is used to acquire initial information of the abnormal event occurring on the traffic road based on a preset first channel, a preset second channel, a preset third channel and a preset fourth channel respectively. The initial information includes first initial information acquired based on the first channel, second initial information acquired based on the second channel, third initial information acquired based on the third channel and fourth initial information acquired based on the fourth channel. The initial information processing module is used to perform deduplication, noise reduction, information extraction, and standardization processing on the first initial information, the second initial information, the third initial information, and the fourth initial information, respectively, to obtain the geographic location coordinates, initial event type label, timestamp, and preset credibility of the abnormal event. The initial event type label includes a first event type label obtained based on the first initial information, a second event type label obtained based on the second initial information, a third event type label obtained based on the third initial information, and a fourth event type label obtained based on the fourth initial information. The event type label matching module is used to match the initial event type label with a preset set of event type labels when the first event type label, the second event type label, the third event type label and the fourth event type label are consistent, so as to obtain the final event type label; The event type label weighting module is used to perform weighted summation and take the maximum value of the first event type label, the second event type label, the third event type label and the fourth event type label according to the preset confidence level when the first event type label, the second event type label, the third event type label and the fourth event type label are inconsistent, so as to obtain the final event type label.

[0107] Optionally, the obstacle information acquisition module mentioned above includes: The target base radius acquisition module is used to obtain the target base radius based on the final event type label and the mapping relationship between the final event type label and the preset base radius. The target base radius is used to quantify the preset area of ​​the geographic location coordinates of the abnormal event. The target base radius correction module is used to correct the target base radius based on a preset severity coefficient of the abnormal event, a preset credibility coefficient of the abnormal event, and a preset environmental complexity coefficient, so as to obtain the corrected radius. The target base radius update module is used to update the corrected radius based on the radius of the influence range of the abnormal event observed in real time by the UAV during the flight, so as to obtain the updated radius. The obstacle information extraction module is used to extract information about the abnormal event with the geographical coordinates of the abnormal event as the center and the updated radius as the range, to obtain the geographical coordinates, ground height, height and status label of each obstacle affecting the flight altitude of the UAV within the updated preset area of ​​the geographical coordinates of the abnormal event.

[0108] Optionally, the aforementioned first flight path determination module includes: The preset flight path length calculation module is used to obtain preset path points based on the geographical coordinates of the obstacles and the geographical coordinates of the abnormal events, and to calculate the length of the preset flight path based on the sum of the distances between all adjacent path points in the preset path. The safety risk value calculation module is used to calculate the safety risk value of the preset flight path based on the distance between the three-dimensional spatial coordinates of each preset path point and the three-dimensional spatial coordinates of each obstacle. The first flight path determination module is used to perform a weighted summation of the length of the preset flight path and the safety risk value of the preset flight path using a preset first objective function to obtain a first flight path of the UAV that satisfies a preset first constraint condition, and control the UAV to fly to the preset position according to the first flight path.

[0109] Optionally, the feature set output module for the aforementioned abnormal events includes: The expected feature set determination module is used to determine the preset height and the expected feature set based on the final event type label; The abnormal event feature set acquisition module is used to preprocess and extract features from the initial images of the abnormal events collected by the UAV at the preset altitude to obtain the feature set of the abnormal events. The completeness calculation module for abnormal events is used to calculate the completeness of the abnormal event based on the ratio of the intersection of the feature set of the abnormal event and the preset expected feature set to the preset expected feature set. The feature set update module for abnormal events is used to collect images of the abnormal event according to a preset monitoring duration and a preset image acquisition frequency when the completeness of the abnormal event is greater than or equal to a preset completeness threshold, without adjusting the preset position of the drone, and to perform preprocessing and feature extraction processing on the images, and output the updated feature set of the abnormal event.

[0110] Optionally, the target location determination module mentioned above includes: The drone's current location information acquisition module is used to acquire the three-dimensional spatial coordinates of the drone's current location, the drone's remaining battery power, the crowd density at the current location, and the vehicle congestion index at the current location when the completeness of the abnormal event is less than a preset completeness threshold. The initial coefficient determination module is used to determine the initial information gain urgency coefficient, the initial displacement cost tolerance coefficient, and the initial security risk sensitivity coefficient based on the final event type label. The current safety valuation calculation module is used to calculate the current safety valuation of the drone based on the distance between the three-dimensional spatial coordinates of the drone's current location and the three-dimensional spatial coordinates of each obstacle, the crowd density at the current location, and the vehicle congestion index at the current location. The initial coefficient correction module is used to correct the initial information gain urgency coefficient based on the difference between the preset integrity threshold and the integrity of the abnormal event, to obtain a corrected information gain urgency coefficient; to correct the initial displacement cost tolerance coefficient based on the remaining battery power of the UAV, to obtain a corrected displacement cost tolerance coefficient; and to correct the initial security risk sensitivity coefficient based on the current security estimate, to obtain a corrected security risk sensitivity coefficient. The initial coefficient normalization module is used to normalize the modified information gain urgency coefficient, the modified displacement cost tolerance coefficient, and the modified safety risk sensitivity coefficient respectively, to obtain the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized safety risk sensitivity coefficient. The candidate location determination module is used to determine each candidate location based on the three-dimensional spatial coordinates of the current location of the UAV and the geographical coordinates of the abnormal event. The candidate location evaluation module is used to calculate the expected integrity improvement value of each candidate location based on the image projection area of ​​the target detection region of the abnormal event at each candidate location, the image projection area of ​​the target detection region at the current position of the UAV, and the obstacle occlusion ratio at each candidate location; calculate the displacement cost of each candidate location based on the three-dimensional spatial coordinates of each candidate location and the three-dimensional spatial coordinates of the current position of the UAV; and calculate the safety risk estimate of each candidate location based on the three-dimensional spatial coordinates of each obstacle, the three-dimensional spatial coordinates of each candidate location, the crowd density at each candidate location, and the vehicle congestion index at each candidate location. The comprehensive evaluation value calculation module is used to perform weighted summation on the expected integrity improvement value of each candidate position and its corresponding normalized information gain urgency coefficient, the displacement cost value of each candidate position and its corresponding normalized displacement cost tolerance coefficient, and the security risk estimate of each candidate position and its corresponding normalized security risk sensitivity coefficient to obtain the comprehensive evaluation value of each candidate position. The target location filtering module is used to determine the target location of the UAV based on the comprehensive evaluation value of each candidate location.

[0111] Optionally, the target location filtering module mentioned above includes: The first target location filtering module is used to select the candidate location corresponding to the largest positive value in the comprehensive evaluation value of each candidate location as the target location of the UAV when there is a largest positive value in the comprehensive evaluation value of each candidate location. The second flight path determination module is used to determine the second flight path of the UAV based on the target position of the UAV; The second target location filtering module is used to sequentially verify whether each of the remaining candidate locations meets the flight conditions when the second flight path does not meet the preset flight conditions, until the flight path of the UAV that meets the flight conditions is obtained as the second flight path of the UAV, and the candidate location corresponding to the second flight path that meets the flight conditions is taken as the target location of the UAV. The shooting parameter optimization module is used to optimize the shooting parameters when the comprehensive evaluation value of each candidate position is negative or it is determined that adjusting the position of the drone will affect the safety of the preset area of ​​the abnormal event. Without adjusting the current position of the drone, the module uses a preset second objective function to perform a weighted summation of the field of view coverage of the target detection area of ​​the abnormal event and the pixel clarity of the drone to obtain the gimbal angle and camera focal length of the drone that satisfy the preset second constraint. The module then re-acquires the current image of the abnormal event according to the gimbal angle and the camera focal length, and calculates the current completeness of the abnormal event based on the current image of the abnormal event. The integrity threshold determination module is used to determine the drone's current position without adjusting it when the current integrity of the abnormal event is greater than or equal to the integrity threshold; and to return to the drone's current position information acquisition module to redetermine the drone's target position when the current integrity of the abnormal event is less than the integrity threshold, until the event integrity calculated based on the drone's target position is greater than or equal to the integrity threshold.

[0112] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0113] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0114] In embodiment nine, a computer device is provided, such as Figure 4 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the UAV-based traffic anomaly detection method described in any of the embodiments one to seven above, for example... Figure 2 Steps 100 to 500 shown are omitted here to avoid repetition.

[0115] In Embodiment 10, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the UAV-based traffic anomaly detection method described in any of Embodiments 1 to 7 above, for example... Figure 2 Steps 100 to 500 shown are omitted here to avoid repetition.

[0116] 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. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program 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.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0118] While specific embodiments of the invention have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the invention. The scope of this invention is defined by the appended claims.

Claims

1. A traffic anomaly detection method based on unmanned aerial vehicles (UAVs), characterized in that, The UAV-based traffic anomaly detection method includes: Step 100: Obtain the geographic coordinates, final event type label, timestamp, and preset credibility of the abnormal event occurring on the traffic road; Step 200: Based on the geographical coordinates of the abnormal event, determine the geographical coordinates, ground height, altitude, and status label of the obstacles affecting the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event. Step 300: Based on the geographical coordinates of the obstacle, the ground elevation of the obstacle's location, the obstacle's own height, and the geographical coordinates of the abnormal event, determine the first flight path of the drone, and control the drone to fly to the preset position according to the first flight path; Step 400: Based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, obtain the feature set of the abnormal event, and calculate the completeness of the abnormal event based on the feature set of the abnormal event and the preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and traffic density. When the completeness meets the preset conditions, output the current feature set of the abnormal event. Step 500: When the completeness does not meet the preset condition, determine the target position of the drone based on the completeness of the abnormal event, control the drone to fly to the target position, take the target position as the preset position, and return to step 400 above. Step 500, the step of determining the target location of the drone, includes: Step 501: When the completeness of the abnormal event is less than a preset completeness threshold, obtain the three-dimensional spatial coordinates of the current location of the drone, the remaining battery power of the drone, the crowd density at the current location, and the vehicle congestion index at the current location. Step 502: Determine the initial information gain urgency coefficient, the initial displacement cost tolerance coefficient, and the initial security risk sensitivity coefficient based on the final event type label; Step 503: Calculate the current safety estimate of the drone based on the distance between the three-dimensional spatial coordinates of the drone's current location and the three-dimensional spatial coordinates of each obstacle, the crowd density at the current location, and the vehicle congestion index at the current location. Step 504: Based on the difference between the preset integrity threshold and the integrity of the abnormal event, the initial information gain urgency coefficient is corrected to obtain the corrected information gain urgency coefficient; based on the remaining battery power of the UAV, the initial displacement cost tolerance coefficient is corrected to obtain the corrected displacement cost tolerance coefficient; based on the current security estimate, the initial security risk sensitivity coefficient is corrected to obtain the corrected security risk sensitivity coefficient. Step 505: Normalize the modified information gain urgency coefficient, the modified displacement cost tolerance coefficient, and the modified safety risk sensitivity coefficient respectively to obtain the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized safety risk sensitivity coefficient. Step 506: Determine each candidate location based on the three-dimensional spatial coordinates of the current location of the UAV and the geographical coordinates of the abnormal event; Step 507: Calculate the expected integrity improvement value for each candidate location based on the image projection area of ​​the target detection region of the abnormal event at each candidate location, the image projection area of ​​the target detection region at the current position of the UAV, and the obstacle occlusion ratio at each candidate location; calculate the displacement cost value for each candidate location based on the three-dimensional spatial coordinates of each candidate location and the three-dimensional spatial coordinates of the current position of the UAV; calculate the safety risk estimate for each candidate location based on the three-dimensional spatial coordinates of each obstacle, the three-dimensional spatial coordinates of each candidate location, the crowd density at each candidate location, and the vehicle congestion index at each candidate location. Step 508: Perform a weighted summation on the expected integrity improvement value of each candidate position and its corresponding normalized information gain urgency coefficient, the displacement cost value of each candidate position and its corresponding normalized displacement cost tolerance coefficient, and the security risk estimate of each candidate position and its corresponding normalized security risk sensitivity coefficient to obtain the comprehensive evaluation value of each candidate position. Step 509: Determine the target location of the UAV based on the comprehensive evaluation value of each candidate location.

2. The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Step 100 includes: Step 101: Obtain initial information of the abnormal event occurring on the traffic road based on a preset first channel, a preset second channel, a preset third channel, and a preset fourth channel, respectively. The initial information includes first initial information obtained based on the first channel, second initial information obtained based on the second channel, third initial information obtained based on the third channel, and fourth initial information obtained based on the fourth channel. Step 102: Perform deduplication, noise reduction, information extraction, and standardization on the first initial information, the second initial information, the third initial information, and the fourth initial information respectively to obtain the geographic location coordinates, initial event type label, timestamp, and preset credibility of the abnormal event. The initial event type label includes a first event type label obtained based on the first initial information, a second event type label obtained based on the second initial information, a third event type label obtained based on the third initial information, and a fourth event type label obtained based on the fourth initial information. Step 103: When the first event type label, the second event type label, the third event type label, and the fourth event type label are consistent, the initial event type label is matched with a preset set of event type labels to obtain the final event type label; Step 104: When the first event type label, the second event type label, the third event type label, and the fourth event type label are inconsistent, the first event type label, the second event type label, the third event type label, and the fourth event type label are weighted and summed according to the preset confidence level to obtain the final event type label.

3. The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Step 200 includes: Step 201: Based on the final event type label and the mapping relationship between the final event type label and the preset base radius, obtain the target base radius, which is used to quantify the preset area of ​​the geographic location coordinates of the abnormal event; Step 202: Correct the target base radius according to the preset severity coefficient of the abnormal event, the preset credibility coefficient of the abnormal event, and the preset environmental complexity coefficient to obtain the corrected radius; Step 203: Update the corrected radius based on the radius of the influence range of the abnormal event observed in real time by the UAV during flight to obtain the updated radius; Step 204: Using the geographic coordinates of the abnormal event as the center and the updated radius as the range, perform information extraction processing on the abnormal event to obtain the geographic coordinates, ground height, height, and status label of each obstacle affecting the flight altitude of the UAV within the updated preset area of ​​the geographic coordinates of the abnormal event.

4. The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Step 300 includes: Step 301: Obtain preset path points based on the geographical coordinates of the obstacles and the geographical coordinates of the abnormal events, and calculate the length of the preset flight path based on the sum of the distances between all adjacent path points in the preset path points; Step 302: Calculate the safety risk value of the preset flight path based on the distance between the three-dimensional spatial coordinates of each preset path point and the three-dimensional spatial coordinates of each obstacle. Step 303: Use a preset first objective function to perform a weighted summation of the length of the preset flight path and the safety risk value of the preset flight path to obtain a first flight path of the UAV that satisfies the preset first constraint condition, and control the UAV to fly to the preset position according to the first flight path.

5. The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Step 400 includes: Step 401: Determine the preset height and the expected feature set based on the final event type label; Step 402: Preprocess and extract features from the initial images of the abnormal event captured by the UAV at the preset altitude to obtain a feature set of the abnormal event; Step 403: Calculate the completeness of the abnormal event based on the ratio of the intersection of the feature set of the abnormal event and the preset expected feature set to the preset expected feature set. Step 404: When the completeness of the abnormal event is greater than or equal to the preset completeness threshold, there is no need to adjust the preset position of the drone, and the images of the abnormal event are collected according to the preset monitoring duration and the preset image acquisition frequency. The images are preprocessed and feature extraction is performed, and the updated feature set of the abnormal event is output.

6. The traffic anomaly detection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, Step 509 includes: Step 5091: When there is a maximum positive value among the comprehensive evaluation values ​​of each candidate position, the candidate position corresponding to the maximum positive value among the comprehensive evaluation values ​​is taken as the target position of the UAV. Step 5092: Determine the second flight path of the UAV based on the target location of the UAV; Step 5093: When the second flight path does not meet the preset flight conditions, verify in turn whether each of the remaining candidate positions meets the flight conditions until the flight path of the UAV that meets the flight conditions is obtained as the second flight path of the UAV, and the candidate position corresponding to the second flight path that meets the flight conditions is taken as the target position of the UAV. Step 5094: When the comprehensive evaluation value of each candidate position is negative or it is determined that adjusting the position of the UAV will affect the safety of the preset area of ​​the abnormal event, there is no need to adjust the current position of the UAV. The field coverage of the target detection area of ​​the abnormal event and the pixel clarity of the UAV are weighted and summed using a preset second objective function to obtain the gimbal angle and camera focal length of the UAV that satisfy the preset second constraint. The current image of the abnormal event is re-acquired according to the gimbal angle and the camera focal length, and the current completeness of the abnormal event is calculated based on the current image of the abnormal event. Step 5095: When the current completeness of the abnormal event is greater than or equal to the completeness threshold, there is no need to adjust the current position of the drone; when the current completeness of the abnormal event is less than the completeness threshold, return to step 501 to redetermine the target position of the drone until the event completeness calculated based on the target position of the drone is greater than or equal to the completeness threshold.

7. A traffic anomaly detection device based on unmanned aerial vehicles (UAVs), characterized in that, The UAV-based traffic anomaly detection device includes: The abnormal event information acquisition module is used to acquire the geographical coordinates, final event type label, timestamp, and preset credibility of abnormal events occurring on traffic roads; The obstacle information acquisition module is used to determine, based on the geographical coordinates of the abnormal event, the geographical coordinates of the obstacles that affect the flight altitude of the UAV within a preset area of ​​the geographical coordinates of the abnormal event, the ground height of their location, their own height, and their own status labels. The first flight path determination module is used to determine the first flight path of the UAV based on the geographical coordinates of the obstacle, the ground height of the obstacle's location, the height of the obstacle itself, and the geographical coordinates of the abnormal event, and to control the UAV to fly to a preset position according to the first flight path. The abnormal event feature set output module is used to obtain the feature set of the abnormal event based on the initial image of the abnormal event collected by the UAV at a preset altitude at the preset location, and to calculate the completeness of the abnormal event based on the feature set of the abnormal event and a preset expected feature set. The feature set of the abnormal event includes: the number of injured persons, the number of vehicles involved, and the traffic density. When the completeness meets the preset conditions, the current feature set of the abnormal event is output. The target location determination module is used to determine the target location of the UAV based on the completeness of the abnormal event when the completeness does not meet the preset condition, control the UAV to fly to the target location, use the target location as the preset location, and return the feature set output module of the above-mentioned abnormal event; The aforementioned target location determination module includes: The drone's current location information acquisition module is used to acquire the three-dimensional spatial coordinates of the drone's current location, the drone's remaining battery power, the crowd density at the current location, and the vehicle congestion index at the current location when the completeness of the abnormal event is less than a preset completeness threshold. The initial coefficient determination module is used to determine the initial information gain urgency coefficient, the initial displacement cost tolerance coefficient, and the initial security risk sensitivity coefficient based on the final event type label. The current safety valuation calculation module is used to calculate the current safety valuation of the drone based on the distance between the three-dimensional spatial coordinates of the drone's current location and the three-dimensional spatial coordinates of each obstacle, the crowd density at the current location, and the vehicle congestion index at the current location. The initial coefficient correction module is used to correct the initial information gain urgency coefficient based on the difference between the preset integrity threshold and the integrity of the abnormal event, to obtain a corrected information gain urgency coefficient; to correct the initial displacement cost tolerance coefficient based on the remaining battery power of the UAV, to obtain a corrected displacement cost tolerance coefficient; and to correct the initial security risk sensitivity coefficient based on the current security estimate, to obtain a corrected security risk sensitivity coefficient. The initial coefficient normalization module is used to normalize the modified information gain urgency coefficient, the modified displacement cost tolerance coefficient, and the modified safety risk sensitivity coefficient respectively, to obtain the normalized information gain urgency coefficient, the normalized displacement cost tolerance coefficient, and the normalized safety risk sensitivity coefficient. The candidate location determination module is used to determine each candidate location based on the three-dimensional spatial coordinates of the current location of the UAV and the geographical coordinates of the abnormal event. The candidate location evaluation module is used to calculate the expected integrity improvement value of each candidate location based on the image projection area of ​​the target detection region of the abnormal event at each candidate location, the image projection area of ​​the target detection region at the current position of the UAV, and the obstacle occlusion ratio at each candidate location; calculate the displacement cost of each candidate location based on the three-dimensional spatial coordinates of each candidate location and the three-dimensional spatial coordinates of the current position of the UAV; and calculate the safety risk estimate of each candidate location based on the three-dimensional spatial coordinates of each obstacle, the three-dimensional spatial coordinates of each candidate location, the crowd density at each candidate location, and the vehicle congestion index at each candidate location. The comprehensive evaluation value calculation module is used to perform weighted summation on the expected integrity improvement value of each candidate position and its corresponding normalized information gain urgency coefficient, the displacement cost value of each candidate position and its corresponding normalized displacement cost tolerance coefficient, and the security risk estimate of each candidate position and its corresponding normalized security risk sensitivity coefficient to obtain the comprehensive evaluation value of each candidate position. The target location filtering module is used to determine the target location of the UAV based on the comprehensive evaluation value of each candidate location.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the traffic anomaly detection method based on any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the traffic anomaly detection method based on any one of claims 1 to 6.

Images