A highway driving unmanned aerial vehicle inspection method and system

Abstract

The application relates to the technical field of image recognition, in particular to a highway driving unmanned aerial vehicle inspection method and system. The method comprises the following steps: acquiring a high-speed inspection image; performing time sequence division on all frames of the high-speed inspection image according to the numerical consistency of the speed, to obtain an inspection image sequence; determining a speed feature vector according to the speed and the driving direction of the vehicle; determining a driving speed risk coefficient of the vehicle in combination with the consistency and the average value of the speed feature vector; determining a direction change uncertainty coefficient in the driving process; determining a driving risk possibility in combination with the driving speed risk coefficient and the direction change uncertainty coefficient; determining a road threat degree according to the distance between the vehicle and other vehicles and the driving risk possibility at the current moment, and performing collision risk vehicle identification according to the road threat degree. The application can effectively improve the timeliness of the risk vehicle identification while ensuring the accuracy.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, specifically to a method and system for unmanned aerial vehicle (UAV) inspection of highway traffic. Background Technology

[0002] With the rapid development of drone technology and the application of artificial intelligence in image recognition, drone inspections are widely used in various industries. Particularly in highway inspections, using drones for real-time video monitoring and image analysis can effectively improve inspection efficiency and reduce labor costs. Currently, highways mainly rely on fixed cameras for monitoring, which have blind spots (such as curves and ramps) and are costly to deploy, making it difficult to achieve full coverage. With surging traffic volume, violations and accident rates are rising year by year; drones, with their high mobility, flexible viewing angles, and high throughput, can compensate for the shortcomings of fixed monitoring, enabling aerial surveillance, hazard tracking, and rapid accident response.

[0003] Because drone inspections rely heavily on single-target detection algorithms and pre-trained hazard prediction models, and in the process of highway inspections under complex road conditions, there are dynamic scenarios such as most vehicles traveling at high speeds, exhibiting abnormal low speeds, or making sudden turns, the computational load for predicting vehicle driving status in highway inspection videos is relatively large, which in turn affects the analysis time. This makes it impossible to assess the danger of vehicles in complex environments in a timely manner, reducing the timeliness and accuracy of drone inspections in identifying dangerous vehicles on highways. Summary of the Invention

[0004] To address the technical problem of the inability to timely assess the danger of vehicles in complex environments and the resulting reduction in the timeliness and accuracy of drone inspections in identifying dangerous vehicles on highways, this invention provides a drone inspection method and system for highway driving. The specific technical solution adopted is as follows: This invention proposes a method for unmanned aerial vehicle (UAV) inspection of highway traffic, the method comprising: Acquire high-speed inspection images of different frames during the uniform speed inspection process of the UAV; determine the vehicle's speed in different frames based on the displacement of the same vehicle in two adjacent high-speed inspection images; Based on the consistency of speed values, all frames of high-speed inspection images are divided into temporal sequences to obtain inspection image sequences. In each high-speed inspection image sequence, the speed feature vector of the inspection image sequence is determined according to the vehicle's speed and direction of travel. Combining the consistency of the speed feature vectors of all inspection image sequences and the average speed of all frames, the vehicle's speed hazard factor is determined. The positions of vehicles in high-speed inspection images are fitted with trajectories. Based on the distribution of instantaneous directions in different frames on the fitted trajectory, the uncertainty coefficient of direction change during driving is determined. Combining the driving speed hazard coefficient and the uncertainty coefficient of direction change, the driving hazard probability of each vehicle is determined. Based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the probability of driving danger, the road threat level of the vehicle is determined, and collision hazard vehicles are identified based on the road threat level.

[0005] Furthermore, determining the vehicle's speed in different frames based on the displacement of the same vehicle in two adjacent high-speed inspection images includes: Calculate the displacement distance of the same vehicle in two adjacent frames, and use the ratio of the displacement distance to the time interval between the two frames as the velocity of the next frame.

[0006] Furthermore, based on the numerical consistency of the velocity, all frames of high-speed inspection images are temporally divided to obtain an inspection image sequence, including: Calculate the absolute value of the velocity difference between adjacent frames to obtain the velocity difference; combine adjacent frames with velocity differences less than a preset difference threshold into a sequence, and traverse all adjacent frames of high-speed inspection images to obtain different inspection image sequences.

[0007] Furthermore, determining the velocity feature vector of the inspection image sequence based on the vehicle's speed and direction of travel in different frames of the high-speed inspection image sequence includes: The least squares method is used to perform curve fitting on the positions in all frames of high-speed inspection images in the inspection image sequence to obtain the fitted curve; In the fitted curve, the velocity vector of each frame of high-speed inspection image is obtained by combining the tangent direction and velocity value corresponding to the position of each frame of high-speed inspection image. The velocity feature vector of the inspection image sequence is obtained by averaging the velocity vectors of all frames of the high-speed inspection image sequence.

[0008] Furthermore, the determination of the vehicle's speed hazard factor by combining the consistency of the speed feature vectors of all inspection image sequences and the average speed of all frames includes: Discreteness analysis was performed on the velocity feature vectors of all inspection image sequences, and the discreteness index of velocity features was obtained after normalization. Calculate the average velocity of all frames of high-speed inspection images, and normalize it as a numerical feature index of velocity. The product of the numerical characteristic index and the discrete index is normalized and used as the driving speed hazard coefficient.

[0009] Furthermore, determining the uncertainty coefficient of directional change during driving based on the distribution of instantaneous directions corresponding to different frames on the fitted trajectory includes: On the fitted curve, the angle between the instantaneous direction and the road extension direction in each frame of the high-speed inspection image is used as the analysis angle of the instantaneous direction. The variance of the analysis angle of all frames of high-speed inspection images is used as the direction change uncertainty coefficient.

[0010] Furthermore, the determination of the driving hazard probability of each vehicle by combining the driving speed hazard coefficient and the direction change uncertainty coefficient includes: The sum of the driving speed hazard coefficient and the direction change uncertainty coefficient is calculated and normalized to represent the driving hazard probability.

[0011] Furthermore, determining the road threat level of a vehicle based on the distance and driving hazard probability between each vehicle and other vehicles in the high-speed inspection image at the current moment includes: Determine the dangerous distance for a collision between two vehicles, calculate the ratio of the Euclidean distance between the vehicle and other vehicles in the high-speed inspection image to the dangerous distance, and normalize the reciprocal of the ratio as the vehicle distance danger index for a collision between the two vehicles. The average of the driving hazard probability of the two vehicles is used as the driving hazard index. Calculate the product of the distance hazard index and the driving hazard index, and normalize it to obtain the collision probability value of the two vehicles colliding. The maximum probability of collision between any vehicle and all other vehicles is taken as the level of road threat posed by the vehicle.

[0012] Furthermore, based on the road threat level, collision hazard vehicle identification includes: Vehicles whose road threat level exceeds a preset threshold are identified as collision hazard vehicles.

[0013] On the other hand, a drone inspection system for highway driving is also provided, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any of the foregoing.

[0014] The present invention has the following beneficial effects: In this embodiment of the invention, the speed of vehicles in different frames is analyzed during the uniform-speed inspection process of the UAV. Similarity temporal clustering is performed based on the speed values ​​to obtain an inspection image sequence. Then, based on the speed and driving direction, the speed feature vector of the inspection image sequence is determined. Based on the consistency of the speed feature vector and the speed mean of all frames, the driving speed hazard coefficient of the vehicle is determined. The driving speed hazard coefficient represents the degree of danger at driving speed, that is, hazard analysis is performed in the vehicle speed dimension. The position of the vehicle in the high-speed inspection image is trajectory fitted, and the direction change uncertainty coefficient is determined based on the direction change on the fitted trajectory. The direction change uncertainty coefficient is the hazard analysis in the direction change dimension. Combining the driving speed hazard coefficient and the direction change uncertainty coefficient, the driving hazard probability of each vehicle is determined. Based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the driving hazard probability, the road threat level of the vehicle is determined, and collision hazard risk vehicle identification is performed based on the road threat level. Compared to existing technologies, this invention utilizes drone inspections to achieve multi-dimensional factor analysis, such as vehicle movement trajectory, speed consistency, and driving direction fluctuations. This allows for analysis based on multi-dimensional historical data, avoiding extensive data calculations while accurately identifying the road threat level of vehicles. This makes threat assessment more timely and reliable, adaptable to complex traffic scenarios, and improves the timeliness of dangerous vehicle identification while ensuring accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0016] Figure 1 This is a flowchart of a method for unmanned aerial vehicle (UAV) inspection of highway traffic, provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a drone inspection method and system for highway traffic according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific scheme for a drone inspection method for highway driving provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a drone inspection method for highway traffic according to an embodiment of the present invention, the method comprising: S101: Acquire high-speed inspection images of different frames during the uniform speed inspection process of the UAV; determine the speed of the vehicle in different frames based on the displacement of the same vehicle in two adjacent high-speed inspection images.

[0021] With the rapid development of drone technology and the application of artificial intelligence in image recognition, drone inspections are widely used in various industries. Particularly in highway inspections, using drones for real-time video monitoring and image analysis can effectively improve inspection efficiency and reduce labor costs. Currently, highways mainly rely on fixed cameras for monitoring, which have blind spots (such as curves and ramps) and are costly to deploy, making it difficult to achieve full coverage. With surging traffic volume, violations and accident rates are rising year by year; drones, with their high mobility, flexible viewing angles, and high throughput, can compensate for the shortcomings of fixed monitoring, enabling aerial surveillance, hazard tracking, and rapid accident response.

[0022] Because drone inspections rely heavily on single-target detection algorithms and pre-trained hazard prediction models, and in the process of highway inspections under complex road conditions, there are dynamic scenarios such as most vehicles traveling at high speeds, exhibiting abnormal low speeds, or making sudden turns, the computational load for predicting vehicle driving status in highway inspection videos is relatively large, which in turn affects the analysis time. This makes it impossible to assess the danger of vehicles in complex environments in a timely manner, reducing the timeliness and accuracy of drone inspections in identifying dangerous vehicles on highways.

[0023] On highways, some vehicles may exhibit abnormal behavior (such as rapid acceleration, frequent lane changes, etc.) due to driver fatigue, illegal driving, or other reasons. Understandably, pre-trained hazard prediction models cannot analyze sudden anomalies in a timely manner. Therefore, in this embodiment of the invention, by tracking and analyzing the movement trajectory of vehicles, calculating the consistency of speed and the uncertainty of directional changes during the driving process of each vehicle, and combining the distance between different vehicles, a comprehensive assessment of the vehicle's hazard is made. Based on historical information, hazard identification can be performed in a timely manner, and suspected dangerous vehicles can be obtained from highway patrol videos.

[0024] The drone, equipped with a high-resolution camera, GPS positioning, and speed sensor, will fly along the direction of traffic on the highway according to a predetermined flight path, and collect highway inspection videos through the high-resolution camera to obtain highway inspection images.

[0025] For any frame of high-speed inspection image, the morphological center position of each vehicle region in the frame of high-speed inspection image is taken as the coordinate position of each vehicle in the frame of high-speed inspection image.

[0026] The speed and direction of a vehicle are crucial indicators affecting vehicle safety. Therefore, speed is analyzed first. In this embodiment of the invention, the vehicle's speed in different frames is determined based on the displacement of the same vehicle in two adjacent high-speed inspection images. This includes calculating the displacement distance of the same vehicle in two adjacent frames and using the ratio of the displacement distance to the time interval between the two frames as the speed of the next frame. The speed of the next frame is determined based on the displacement change between two adjacent frames to facilitate subsequent data analysis.

[0027] Since drone inspections occur at a constant speed, the analysis reference needs to be positioned on fixed reference objects within the image. In some embodiments of this invention, displacement analysis can be performed directly based on fixed reference objects on highways, such as road signs, milestones, and lane markings. By comparing the vehicle's coordinates between two frames with the fixed reference objects, the vehicle's displacement distance can be effectively analyzed, thereby determining the vehicle's speed based on the displacement distance and the time interval between the two frames.

[0028] S102: Based on the consistency of speed values, perform time-series division on all frames of high-speed inspection images to obtain an inspection image sequence; in different frames of high-speed inspection images within each inspection image sequence, determine the speed feature vector of the inspection image sequence based on the vehicle's speed and driving direction; combine the consistency of the speed feature vectors of all inspection image sequences and the average speed of all frames to determine the vehicle's driving speed hazard factor.

[0029] Vehicles driving safely on highways typically exhibit relatively stable speed characteristics. Therefore, when a vehicle changes speed or lanes on a highway, its speed will change significantly, but after the change, it will still maintain a relatively small change in speed and direction. Therefore, by tracking and analyzing the speed changes of vehicles, calculating the speed consistency of each vehicle within similar inspection time periods at different speeds, and then combining the overall speed characteristics and speed consistency, the driving speed hazard factor can be determined.

[0030] Based on the consistency of velocity values, all frames of high-speed inspection images are temporally divided to obtain inspection image sequences, including: calculating the absolute value of the velocity difference between adjacent frames to obtain velocity differences; combining adjacent frames with velocity differences less than a preset difference threshold into a sequence; and traversing all adjacent frames of high-speed inspection images to obtain different inspection image sequences.

[0031] The preset difference threshold is a threshold value for speed difference. In this embodiment of the invention, the preset difference threshold can be, for example, 1 meter per second. That is, when the speed difference between two adjacent frames is less than 1 meter per second, they are regarded as the same inspection image sequence.

[0032] It is understandable that the inspection image sequence represents a time series with similar vehicle speeds. Since vehicles on highways are usually in a constant speed or cruise state, when a vehicle undergoes abnormal behavior such as sudden acceleration / sudden braking (such as collision risk) or lane change (sudden change in trajectory) on the highway, its speed will change significantly. Therefore, segmentation based on speed changes is used to obtain inspection image sequences with similar speeds.

[0033] In each high-speed inspection image sequence, based on the vehicle's speed and direction of travel, the velocity feature vector of the inspection image sequence is determined. This includes: performing curve fitting on the positions of all high-speed inspection images in the inspection image sequence using the least squares method to obtain a fitted curve; combining the tangent direction and velocity value corresponding to the position of each high-speed inspection image in the fitted curve to obtain the velocity vector of each high-speed inspection image; and averaging the velocity vectors of all high-speed inspection images in the inspection image sequence to obtain the velocity feature vector of the inspection image sequence.

[0034] Among them, the least squares method is an algorithm well known to those skilled in the art. Based on the least squares method, curve fitting can be achieved, that is, the position of the vehicle in all frames of high-speed inspection images is fitted to a curve to obtain a fitted curve. The fitted curve represents the vehicle's driving trajectory. The driving direction can be determined based on the driving trajectory, that is, the tangent direction on the driving trajectory. The tangent direction and the velocity are combined to form a velocity vector.

[0035] In this embodiment of the invention, the velocity feature vector represents the overall velocity value and direction characteristics of the sequence. The velocity feature vector can be obtained by integrating the velocity vectors of all frames of high-speed inspection images in the inspection image sequence.

[0036] Furthermore, in some embodiments of the present invention, the vehicle's driving speed hazard factor is determined by combining the consistency of the speed feature vectors of all inspection image sequences and the speed mean of all frames. This includes: performing discrete analysis on the speed feature vectors of all inspection image sequences, and obtaining a discrete index of speed features after normalization; calculating the speed mean of all frames of high-speed inspection images, and normalizing it as a numerical feature index of speed; and normalizing the product of the numerical feature index and the discrete index as the driving speed hazard factor.

[0037] The discreteness analysis in this embodiment of the invention can specifically be, for example, calculating the standard deviation of all speed feature vectors, or using the coefficient of variation for specific calculation. There are no restrictions on this. After discreteness calculation and normalization, a discreteness index of speed features is obtained. The larger the value of the discreteness index, the more discrete the speed change, that is, the more unstable the overall driving process.

[0038] Among them, the larger the numerical characteristic index, the larger the average speed value, indicating that the vehicle driving speed is faster and the risk factor is higher. Therefore, the embodiments of the present invention directly calculate the product value of the numerical characteristic index and the discrete index, and normalize it as the driving speed risk factor, which represents the degree of danger caused by the speed value during driving.

[0039] S103: Perform trajectory fitting on the vehicle positions in the high-speed inspection images, and determine the direction change uncertainty coefficient during driving based on the distribution of instantaneous directions corresponding to different frames on the fitted trajectory; combine the driving speed hazard coefficient and the direction change uncertainty coefficient to determine the driving hazard probability of each vehicle.

[0040] In this embodiment of the invention, the least squares method can be used to fit the position in all frames of high-speed inspection images in the inspection image sequence, that is, the fitting curve in the above embodiment is used as the result of trajectory fitting.

[0041] Based on the distribution of instantaneous directions corresponding to different frames on the fitted trajectory, the uncertainty coefficient of direction change during driving is determined, including: on the fitted curve, the angle between the instantaneous direction and the road extension direction in each frame of high-speed inspection image is taken as the analysis angle of instantaneous direction; the variance of the analysis angles of all frames of high-speed inspection images is taken as the uncertainty coefficient of direction change.

[0042] The road extension direction specifically refers to the actual extension direction of the physical lanes of the highway. Under normal circumstances, vehicles should travel steadily in this direction. However, abnormal behavior due to driver fatigue, illegal driving, or other reasons (such as loss of vehicle control, frequent lane changes, etc.) will cause the vehicle's actual instantaneous direction of travel to deviate from the road's forward direction, generating certain angular information. In such cases, specific calculations are required to conduct a hazard analysis.

[0043] In this embodiment of the invention, the variance of all analysis angles is directly calculated as the uncertainty coefficient of direction change. The larger the variance, the more unstable the overall direction change, the less stable the vehicle can move along the road direction, and the greater the possibility of dangerous driving.

[0044] Furthermore, in some embodiments of the present invention, the driving hazard probability of each vehicle is determined by combining the driving speed hazard coefficient and the direction change uncertainty coefficient, including: calculating the sum of the driving speed hazard coefficient and the direction change uncertainty coefficient, and normalizing it as the driving hazard probability.

[0045] After normalizing the variance of the analysis angles of all frames of high-speed inspection images, the result is used as the direction change uncertainty coefficient. The normalization process involved in this invention refers to: based on the preset global statistical maxima and minima in the system's historical inspection database, using a maximum-minimum normalization method to linearly map the corresponding values ​​to the [0,1] interval. The driving speed hazard coefficient represents the degree of danger in the vehicle's driving speed dimension, while the direction change uncertainty coefficient represents the degree of danger in the vehicle's overall driving direction. Combining the vehicle's state analysis in both dimensions, the sum of the driving speed hazard coefficient and the direction change uncertainty coefficient is directly calculated and normalized as the driving hazard probability.

[0046] The higher the probability of driving danger, the more unstable the vehicle's speed and direction are during driving, and the more likely it is to cause driving danger.

[0047] S104: Based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the probability of driving danger, determine the road threat level of the vehicle, and identify the vehicle at risk of collision based on the road threat level.

[0048] The advantage of drone inspection is that, compared to analyzing driving hazards directly from inside the vehicle, drones can perform full-process tracking analysis from an objective perspective. This tracking analysis is not only for a single vehicle, but for the operation of all vehicles that can be observed.

[0049] In this embodiment of the invention, the driving hazard probability of all vehicles in the images captured during the drone inspection process, as well as the straight-line distance between different vehicles, are obtained, and collision hazard analysis between vehicles is realized based on this.

[0050] Furthermore, in some embodiments of the present invention, the road threat level of a vehicle is determined based on the distance and driving hazard probability between each vehicle and other vehicles in the high-speed inspection image at the current moment, including: determining the danger distance for a collision between two vehicles; calculating the ratio of the Euclidean distance between the vehicle and other vehicles in the high-speed inspection image to the danger distance; normalizing the reciprocal of the ratio as the distance danger index for a collision between the two vehicles; using the mean of the driving hazard probability of the two vehicles as the driving hazard index; calculating the product of the distance danger index and the driving hazard index, normalizing it as the collision probability value for a collision between the two vehicles; and taking the maximum value of the collision probability values ​​of any vehicle with all other vehicles as the road threat level of the vehicle.

[0051] The dangerous distance between the two vehicles is a preset value. This value can be a fixed value, such as 100 meters, or it can be analyzed based on the speed of the two vehicles. For example, the faster the speed, the longer the required braking buffer distance, and the larger the corresponding dangerous distance value. The dangerous distance is obtained through linear calculation, and there are no restrictions on this.

[0052] Drivers have a certain reaction time when driving a vehicle. Similarly, after operating the vehicle, a certain buffer distance is needed to avoid danger. Therefore, by determining the danger distance, it is possible to make a judgment on the distance between vehicles to avoid danger.

[0053] First, the ratio of the Euclidean distance to the danger distance between the vehicle and other vehicles in the high-speed inspection image is calculated. The ratio of the Euclidean distance to the danger distance represents the safety factor between the two vehicles. The larger the value, the safer the two vehicles are. Therefore, the reciprocal of the ratio is normalized and used as the vehicle distance danger index for the collision between the two vehicles, which characterizes the danger of the distance between the two vehicles.

[0054] Then, the average of the driving hazard probability of the two vehicles is used as the driving hazard index, that is, the instability of the two vehicles in terms of driving speed and driving direction is analyzed to obtain the driving hazard index between the two vehicles.

[0055] The product of the distance hazard index and the driving hazard index is calculated and normalized to obtain the collision probability value for the two vehicles. It should be noted that this collision probability value characterizes the likelihood of a collision and is obtained by analyzing the distance and driving status between the two vehicles. The collision probability value allows for hazard identification; that is, the maximum collision probability value of any vehicle with all other vehicles is taken as the vehicle's road threat level. A higher road threat level value indicates a greater probability of the vehicle colliding with other vehicles.

[0056] Furthermore, in some embodiments of the present invention, identifying vehicles at risk of collision based on the degree of road threat includes: identifying vehicles whose road threat level is greater than a preset threshold as vehicles at risk of collision.

[0057] The preset threshold is a threshold value for the road threat level. In this embodiment of the invention, the preset threshold can be, for example, 0.8. That is, when the road threat level is greater than 0.8, at least one of the corresponding vehicles and other vehicles has poor driving stability and the distance between the vehicles is more dangerous. Therefore, it is identified as a vehicle with a collision risk, which is convenient for timely tracking, recording and reminders.

[0058] In this embodiment of the invention, the speed of vehicles in different frames is analyzed during the uniform-speed inspection process of the UAV. Similarity temporal clustering is performed based on the speed values ​​to obtain an inspection image sequence. Then, based on the speed and driving direction, the speed feature vector of the inspection image sequence is determined. Based on the consistency of the speed feature vector and the speed mean of all frames, the driving speed hazard coefficient of the vehicle is determined. The driving speed hazard coefficient represents the degree of danger at driving speed, that is, hazard analysis is performed in the vehicle speed dimension. The position of the vehicle in the high-speed inspection image is fitted with a trajectory. The direction change uncertainty coefficient is determined based on the direction change on the fitted trajectory. The direction change uncertainty coefficient is then used for hazard analysis in the direction change dimension. Combining the driving speed hazard coefficient and the direction change uncertainty coefficient, the driving hazard probability of each vehicle is determined. Based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the driving hazard probability, the road threat level of the vehicle is determined. Collision hazard vehicle identification is performed based on the road threat level. Compared to existing technologies, this invention utilizes drone inspections to achieve multi-dimensional factor analysis, such as vehicle movement trajectory, speed consistency, and driving direction fluctuations. This allows for analysis based on multi-dimensional historical data, avoiding extensive data calculations while accurately identifying the road threat level of vehicles. This makes threat assessment more timely and reliable, adaptable to complex traffic scenarios, and improves the timeliness of dangerous vehicle identification while ensuring accuracy.

[0059] On the other hand, a drone inspection system for highway driving is also provided, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any of the foregoing.

[0060] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for unmanned aerial vehicle (UAV) inspection of highway traffic, characterized in that, The method includes: Acquire high-speed inspection images of different frames during the uniform speed inspection process of the UAV; determine the vehicle's speed in different frames based on the displacement of the same vehicle in two adjacent high-speed inspection images; Based on the consistency of speed values, all frames of high-speed inspection images are divided into temporal sequences to obtain inspection image sequences. In each high-speed inspection image sequence, the speed feature vector of the inspection image sequence is determined according to the vehicle's speed and direction of travel. Combining the consistency of the speed feature vectors of all inspection image sequences and the average speed of all frames, the vehicle's speed hazard factor is determined. The positions of vehicles in high-speed inspection images are fitted with trajectories. Based on the distribution of instantaneous directions in different frames on the fitted trajectory, the uncertainty coefficient of direction change during driving is determined. Combining the driving speed hazard coefficient and the uncertainty coefficient of direction change, the driving hazard probability of each vehicle is determined. Based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the probability of driving danger, the road threat level of the vehicle is determined, and collision hazard vehicles are identified based on the road threat level.

2. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, The step of determining the vehicle's speed in different frames based on the displacement of the same vehicle in two adjacent high-speed inspection images includes: Calculate the displacement distance of the same vehicle in two adjacent frames, and use the ratio of the displacement distance to the time interval between the two frames as the velocity of the next frame.

3. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, Based on the consistency of velocity values, all frames of high-speed inspection images are temporally divided to obtain an inspection image sequence, including: Calculate the absolute value of the velocity difference between adjacent frames to obtain the velocity difference; combine adjacent frames with velocity differences less than a preset difference threshold into a sequence, and traverse all adjacent frames of high-speed inspection images to obtain different inspection image sequences.

4. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, The step of determining the velocity feature vector of the inspection image sequence based on the vehicle's speed and direction of travel in different frames of the high-speed inspection image sequence includes: The least squares method is used to perform curve fitting on the positions in all frames of high-speed inspection images in the inspection image sequence to obtain the fitted curve; In the fitted curve, the velocity vector of each frame of high-speed inspection image is obtained by combining the tangent direction and velocity value corresponding to the position of each frame of high-speed inspection image. The velocity feature vector of the inspection image sequence is obtained by averaging the velocity vectors of all frames of the high-speed inspection image sequence.

5. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, The determination of the vehicle's speed hazard factor by combining the consistency of the speed feature vectors of all inspection image sequences and the average speed of all frames includes: Discreteness analysis was performed on the velocity feature vectors of all inspection image sequences, and the discreteness index of velocity features was obtained after normalization. Calculate the average velocity of all frames of high-speed inspection images, and normalize it as a numerical feature index of velocity. The product of the numerical characteristic index and the discrete index is normalized and used as the driving speed hazard coefficient.

6. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 4, characterized in that, The step of determining the direction change uncertainty coefficient during driving based on the distribution of instantaneous directions corresponding to different frames on the fitted trajectory includes: On the fitted curve, the angle between the instantaneous direction and the road extension direction in each frame of the high-speed inspection image is used as the analysis angle of the instantaneous direction. The variance of the analysis angle of all frames of high-speed inspection images is used as the direction change uncertainty coefficient.

7. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, The determination of the driving hazard probability for each vehicle by combining the driving speed hazard coefficient and the direction change uncertainty coefficient includes: The sum of the driving speed hazard coefficient and the direction change uncertainty coefficient is calculated and normalized to represent the driving hazard probability.

8. The method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, The determination of a vehicle's road threat level based on the distance between each vehicle and other vehicles in the high-speed inspection image at the current moment and the probability of driving danger includes: Determine the dangerous distance for a collision between two vehicles, calculate the ratio of the Euclidean distance between the vehicle and other vehicles in the high-speed inspection image to the dangerous distance, and normalize the reciprocal of the ratio as the vehicle distance danger index for a collision between the two vehicles. The average of the driving hazard probability of the two vehicles is used as the driving hazard index. Calculate the product of the distance hazard index and the driving hazard index, and normalize it to obtain the collision probability value of the two vehicles colliding. The maximum probability of collision between any vehicle and all other vehicles is taken as the level of road threat posed by the vehicle.

9. A method for unmanned aerial vehicle (UAV) inspection of highway traffic as described in claim 1, characterized in that, Based on the road threat level, collision hazard vehicle identification includes: Vehicles whose road threat level exceeds a preset threshold are identified as collision hazard vehicles.

10. A drone inspection system for highway driving, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.

Images