A method for constructing a small clearance velocity model for tunnel-interchange exits based on aerial photography identification

Abstract

The present invention provides a method for constructing a speed model for tunnel-interchange exits with small clearance distances based on aerial photography. The method comprises: collecting video of vehicle driving behavior on a small clearance distance section of a tunnel-interchange exit and performing registration processing; employing a target detection algorithm to identify and detect vehicles, and employing a multi-target tracking algorithm to continuously track them; demarcating lane lines within the video lanes and calculating and obtaining real-time vehicle speed and position information; and performing a single-factor analysis based on the vehicle speed and position data to determine the speed model's selected indicators as the circular curve radius, distance to the tunnel, and clearance length. After determining the nonlinear correlation between the three indicators, a speed model is fitted, and the best fitting equation is obtained as the vehicle speed model. This method is used to predict vehicle speeds and characterize speed variation patterns during the highway design phase, providing a reference for speed models in tunnel-interchange exit design.

Description

Technical Field

[0001] The present invention relates to the field of road traffic safety technology, and in particular to a method for constructing a tunnel-interchange exit small clearance speed model based on drone aerial photography identification. Background Art

[0002] Although high design standards are maintained in expressways, and efforts are made to minimize narrow clearances between tunnels and interchanges, expressways often encounter complex and challenging terrain and geology as they traverse mountainous terrain. This inevitably leads to interchanges being too close to other structures during road layout. Existing research shows that tunnel-interchange sections with narrow clearances are more prone to excessive speeds and increased vehicle dispersion than normal sections, leading to a higher risk of traffic accidents. Studying the speed patterns of vehicles on these special sections is a crucial consideration for rational expressway design. Previous studies have often used radar / laser velocimeters and pneumatic tube velocimeters for fixed-point speed measurement, but these methods suffer from discontinuous data collection. Furthermore, the availability of manual installation of lidar or pneumatic tube velocimeters is lacking on tunnel-interchange sections with narrow clearances. Therefore, accurately determining vehicle speeds on these sections and minimizing traffic safety hazards are urgent issues. Summary of the Invention

[0003] To solve the above problems, the present invention provides a method for constructing a small-clearance speed model of tunnel-interchange exits based on drone aerial photography identification. The method is used to predict vehicle operating speeds and characterize speed variation patterns during the highway design stage, providing a speed model reference for tunnel-interchange exit design.

[0004] To achieve the above objectives, the present invention provides the following technical solutions.

[0005] A method for constructing a tunnel-interchange exit small clearance velocity model based on aerial photography identification includes the following steps:

[0006] Collect vehicle driving behavior videos on the small clearance section between the tunnel and the interchange exit, perform frame-by-frame registration using the SIFT algorithm, and resynthesize the registered videos.

[0007] The target detection algorithm is used to detect vehicles in the re-synthesized video, and the multi-target tracking algorithm is used to continuously track vehicles;

[0008] Obtain tracking video images, calibrate lane lines within the video lane, and calculate the real-time speed and position information of the vehicle;

[0009] Based on the single factor analysis of vehicle speed and position information, the speed model selection indicators are determined to be the circular curve radius, distance to the tunnel position and clear distance length;

[0010] The collinearity diagnosis of factors is performed on the circular curve radius, distance to the tunnel position and clear distance length. After determining the nonlinear correlation between the three indicators, the running speed model is fitted and the fitting equation with the best fitting degree is obtained as the vehicle running speed model.

[0011] Preferably, the method of collecting a video of vehicle driving behavior on a small clearance section between a tunnel and an interchange exit and performing frame-by-frame registration processing using a SIFT algorithm comprises the following steps:

[0012] Take off the drone near the pre-determined measurement point, hover it in the air to take aerial photos and collect videos;

[0013] Use the SIFT algorithm to capture a clear image without any obstacles in the original video as the reference image. Let the reference image be f(x, y) and the image to be registered be g(x, y).

[0014] Let the point (x, y) on the reference image be (x, y) in the image to be registered, so that there is an affine relationship between the two. Using this logic, the feature points in the reference image are matched to the same position as the first frame in the video to be processed, and the registration process of the original video is completed by offsetting.

[0015] After the original video is registered, the cropped images are imported into Premiere Pro in time sequence, and the video is resynthesized to maintain the same frame rate as the original video for data analysis and output.

[0016] Preferably, the method of using a target detection algorithm to perform vehicle recognition detection on the resynthesized video and using a multi-target tracking algorithm to continuously track the vehicle includes the following steps:

[0017] The YOLO V5 algorithm is used to perform vehicle recognition and detection, and the DeepSort algorithm is used to continuously track the detected vehicles.

[0018] Preferably, the acquisition of tracking video images, calibration of lane lines within the video lane, and calculation of the real-time speed and position information of the vehicle include the following steps:

[0019] Cut out a picture from the aerial video, import it into the CAD software with the upper left corner of the picture at the origin of the coordinates, and scale it so that the coordinate values ​​of the length and width of the picture are equal to the pixel length. Make the coordinates of any point in the picture in the CAD software correspond to the pixel coordinates of the image, and the extracted vehicle X The coordinates are the same, Y The coordinates are opposite numbers;

[0020] Through the plane design function of the Weidi road design software, the design line that is the same as the lane line in the picture is designed according to the field survey situation. The pile-by-pile coordinates of the design line are output at fixed intervals. Y The axis coordinate value is negative, that is, the pixel coordinate calibration file of the lane line in the same coordinate system as the detection vehicle coordinate is obtained;

[0021] A program is written in Python to iterate over all vehicle numbers and obtain the continuous pixel coordinates and speed values ​​of each vehicle in the video. The program then iterates over each pile number of the marked lane line that is desired to be extracted, collecting data every 1 meter. The distance between each pile number and all the coordinates of the numbered vehicles extracted in the previous layer is calculated. This method iterates over each vehicle and each section to obtain the speed and acceleration information of each vehicle in each section.

[0022] Preferably, the single factor analysis is performed based on the data of vehicle speed and position information to determine that the speed model selection indicators are the circular curve radius, the distance to the tunnel position and the clear distance length, including the following steps:

[0023] When determining the indicators for the speed model, each independent variable and speed are analyzed and evaluated. The speed dispersion index is represented by the standard deviation SD, the average adjacent vehicle speed difference ASD, and the speed variation coefficient η. The speed variation coefficient η refers to the ratio of the speed standard deviation to the average speed.

[0024] According to the degree of discreteness of each indicator, the radius of the circular curve is selected R , distance from the tunnel location D , Clearance Length L as independent variables required to build the model.

[0025] Preferably, the method further comprises: performing a goodness of fit analysis on the fitted running speed model using the AIC criterion:

[0026] When constructing the speed model, the driving simulation data was supplemented and put into 1stopt software for multivariate nonlinear regression fitting analysis, with reference to the determination coefficient R 2 At the same time, AIC is introduced to measure the complexity of the model and the goodness of fitting data; the AIC indicator is based on the entropy value and uses the residual square sum SSE and sample size n And the number of parameters K Characterization: .

[0027] Beneficial effects of the present invention:

[0028] This paper proposes a method for constructing a small-clearance speed model for tunnel-interchange exits based on drone aerial photography. This method is used to predict vehicle operating speeds and characterize speed variations during highway design, providing a reference for tunnel-interchange exit design. The speed model constructed in this paper has an overall low average relative error, never exceeding 10%, a manageable error range, and good prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of model construction according to an embodiment of the present invention;

[0030] Figure 2 This is a screenshot comparing the video re-synthesized by Premiere Pro and the original video in an embodiment of the present invention;

[0031] Figure 3 In the embodiment of the present invention, the pixel position of each point is determined by the coordinate system in CAD;

[0032] Figure 4 In this embodiment of the present invention, the YOLO algorithm is used for video accuracy improvement training, and the vehicle frame is marked in the video used for training and analysis;

[0033] Figure 5 A vehicle speed and position detection diagram according to an embodiment of the present invention;

[0034] Figure 6 This is a logic flow chart of the DeepSort algorithm according to an embodiment of the present invention;

[0035] Figure 7 This is a schematic diagram of lane marking according to an embodiment of the present invention;

[0036] Figure 8 A schematic diagram of a characteristic cross-section vehicle speed extraction position according to an embodiment of the present invention;

[0037] Figure 9 A diagram illustrating the vehicle detection accuracy of an example embodiment of the present invention;

[0038] Figure 10 This is a comparison diagram of the vehicle speed of an example embodiment of the present invention and the speed predicted by the model. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] Example 1

[0041] The invention provides a method for constructing a tunnel-interchange exit small clearance speed model based on aerial photography identification, such as Figure 1-10 As shown, the following steps are included:

[0042] S1.1: Take off the drone near the pre-determined measurement point, hover it at an appropriate altitude in the air, and continuously record the video.

[0043] S1.2: Use the SIFT algorithm to capture a clear, unobstructed image from the original video as the reference image. Let the reference image be f(x, y) and the image to be registered be g(x, y). Let the point (x, y) on the reference image be (x, y) in the image to be registered, so that there is an affine relationship between the two. Using this logic, match the feature points in the reference image to the same position as in the first frame of the video to be processed, thereby completing the registration process of the original video.

[0044] S1.3: After the original video is registered, the cropped images are imported into Premiere Pro in time sequence, and the video is resynthesized to maintain the same frame rate as the original video for data analysis and output. Figure 2 shown.

[0045] S2.1: Use the YOLO algorithm to improve video recognition accuracy. Vehicles are marked in the training and analysis videos. The model is trained with a training set: validation set: test set ratio of 0.7:0.2:0.1. For a total of 100 rounds, the model with the best test results is used for vehicle detection.

[0046] S2.2: Use DeepSort algorithm to continuously track the detected vehicles. The specific steps are as follows: Figure 7 shown.

[0047] S3.1: Determine the method of pre-calibrating lane lines and then calculating the distance instead of the original YOLO method of real-time lane line recognition to stably obtain the speed and distance relationship between the vehicle and the lane line, such as Figure 4 Take a picture from the aerial video of each survey section, import the picture into the CAD software with the upper left corner at the coordinate origin, and scale it so that the coordinate values ​​of the length and width of the picture are equal to the pixel length, as shown in the figure below. Figure 3As shown. Under this condition, the coordinates of any point in the figure in the CAD software correspond to the pixel coordinates of the image, which are the same as the extracted vehicle X coordinates, and the Y coordinates are the opposite. Through the graphic design function of the Weidi road design software, the design line that is the same as the lane line in the picture is fitted and designed according to the field survey situation, and then the pile-by-pile coordinates of the design line are output according to a fixed interval, and can be output with different precisions according to experimental requirements, and the highest precision in the actual road can be achieved. Take the negative value of the Y-axis coordinate value of the pile-by-pile coordinate table, which is the pixel coordinate calibration file of the lane line. Its coordinates are in the same coordinate system as the vehicle coordinates detected and extracted in the previous chapter, and can be directly calculated to complete the design annotation.

[0048] S3.2: Use Python to write a program to traverse all vehicle numbers and obtain the continuous pixel coordinates and speed values ​​of each vehicle in the video; then traverse each pile number of the marked lane line that you want to extract, collect data every 1m according to the experimental requirements, and calculate the distance between each pile number and all the coordinates of the numbered vehicles extracted in the previous layer. By traversing each vehicle and each section in this way, you can get the speed of each vehicle in each section, such as Figure 5-8 Similarly, the speed, acceleration and other information of all vehicles at each pile number can also be obtained.

[0049] S4.1: Initially select the circular curve radius, longitudinal slope, clear distance length, curve-slope combination, distance to tunnel entrance, tunnel length, tunnel entrance orientation, and tunnel lighting as indicators. After single-factor analysis, determine the circular curve radius R, distance to tunnel position D, and clear distance length L as indicators of the vehicle speed model.

[0050] S4.2: Perform collinearity diagnosis on the three indicators of circular curve radius R, distance to tunnel position D, and clear distance length L. After confirming the nonlinear correlation between the three indicators, formulate multiple functional relationships in SPSS for fitting trial calculations. Use the multivariate nonlinear regression method to fit the collected data, and perform goodness of fit analysis on the fitting function to determine the best explanatory relationship between the dependent variable and the independent variable. Finally, obtain the vehicle speed prediction model related to the road geometry indicators. Finally, obtain the passenger car speed model with the best fit, as shown in the following formula and Table 1:

[0051] Passenger car running speed model:

[0052]

[0053] Table 1 Speed ​​model coefficients of small vehicles on exit sections with small clearance distances

[0054]

[0055] Notes:

[0056] In the table: f pD1 =17.28-3.44L / 1 / 2-166.80 / L2, f pD2 =0.036-10.93 / L-518.11eL

[0057] f pD1 zd =-9.995+0.48L+52.03 / L, f pD2 zd =-10.349 / L+0.10L-0.176

[0058] In the above formula, L∈[10,500]

[0059] S4.3: When constructing the large vehicle speed model, the same fitting equation as for passenger cars was used to analyze its applicability to large vehicles. It was found that the fitting equation for large vehicles was more applicable. The large vehicle speed model was then obtained, as shown in the following equation and Table 2.

[0060] Large vehicle speed model:

[0061]

[0062] Table 2. Model coefficients of large vehicle speed on exit sections with small clearance distances

[0063]

[0064] Notes:

[0065] In the table: f pD2 =-78.93+15.88Ln(L)+379.33 / L, f pD3 =e4.97+135.11 / L-56.13,f pR2 =1.88+2.58 / 1+2.54L;

[0066] f pD3 zd =22.14-0.37L-137.65 / L, f pR1 zd =15.39+0.16L-7.43Ln(L),

[0067] f pR2 zd =-0.96-eL / 55.02+0.248L

[0068] In the above formula, L∈[10,500]

[0069] S5.1: The speed model variance test was conducted on small and large vehicles on small clearance road sections. The residuals showed equal variance and approximately conformed to the normal distribution, indicating that the prediction model explained the actual situation well.

[0070] S5.2: Select engineering examples to conduct practical verification of the speed model. After the video recognition processing of the example road section, the data is sorted into large and small vehicle types. The cross-section data is tested together with the model prediction data to verify whether there are any anomalies in the speed model of large and small vehicles. The relative error and the average relative error are selected as error measurement indicators. The verification shows that the average relative error of the model is generally low, not exceeding 10%. The model error range is controllable and the model prediction accuracy is good. Figure 9-10 shown.

[0071] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a small clearance speed model of a tunnel-interchange exit based on aerial photography identification, characterized in that: The following steps are involved: Collect vehicle driving behavior videos on the small clearance section between the tunnel and the interchange exit, perform frame-by-frame registration using the SIFT algorithm, and resynthesize the registered videos. The target detection algorithm is used to detect vehicles in the re-synthesized video, and the multi-target tracking algorithm is used to continuously track vehicles; Obtain tracking video images, calibrate lane lines within the video lane, and calculate the real-time speed and position information of the vehicle; Based on the single factor analysis of vehicle speed and position information, the speed model selection indicators are determined to be the circular curve radius, distance to the tunnel position and clear distance length; Perform factor collinearity diagnosis on the circular curve radius, distance to tunnel position, and clear distance length. After confirming the nonlinear correlation between the three indicators, perform a fitting trial calculation of the running speed model and obtain the fitting equation with the best fit as the vehicle running speed model. Among them, the passenger car running speed model: Large vehicle speed model: Where R is the radius of the circular curve, D is the distance from the tunnel, and L is the clear distance length; 、 、 、 、 、 、 is the fitting parameter.

2. The method for constructing a tunnel-interchange exit small clearance velocity model based on aerial photography identification according to claim 1, characterized in that: The method of collecting a video of a vehicle driving behavior on a small clearance section between a tunnel and an interchange exit and performing frame-by-frame registration processing using a SIFT algorithm comprises the following steps: Take off the drone near the pre-determined measurement point, hover it in the air to take aerial photos and collect videos; Use the SIFT algorithm to capture a clear image without any obstacles in the original video as the reference image. Let the reference image be f(x, y) and the image to be registered be g(x, y). Let the point (x, y) on the reference image be (x, y) in the image to be registered, so that there is an affine relationship between the two. Using this logic, the feature points in the reference image are matched to the same position as the first frame in the video to be processed, and the registration process of the original video is completed by offsetting. After the original video is registered, the cropped images are imported into Premiere Pro in time sequence, and the video is resynthesized to maintain the same frame rate as the original video for data analysis and output.

3. The method for constructing a tunnel-interchange exit small clearance velocity model based on aerial photography identification according to claim 1, characterized in that: The method of using a target detection algorithm to perform vehicle recognition detection on the resynthesized video and using a multi-target tracking algorithm to continuously track the vehicle includes the following steps: The YOLO V5 algorithm is used for vehicle recognition and detection, and the DeepSort algorithm is used to continuously track the detected vehicles.

4. The method for constructing a tunnel-interchange exit small clearance velocity model based on aerial photography identification according to claim 1, characterized in that: The acquisition of tracking video images, calibration of lane lines within the video lane, and calculation of the real-time speed and position information of the vehicle include the following steps: Cut out a picture from the aerial video, import it into the CAD software with the upper left corner of the picture at the origin of the coordinates, and scale it so that the coordinate values ​​of the length and width of the picture are equal to the pixel length. Make the coordinates of any point in the picture in the CAD software correspond to the pixel coordinates of the image, and the extracted vehicle X The coordinates are the same, Y The coordinates are opposite numbers; Through the plane design function of Weidi road design software, the design line that is the same as the lane line in the picture is designed according to the field survey situation. The pile-by-pile coordinates of the design line are output at fixed intervals. Y The axis coordinate value is negative, that is, the pixel coordinate calibration file of the lane line in the same coordinate system as the detection vehicle coordinate is obtained; A program is written in Python to iterate over all vehicle numbers and obtain the continuous pixel coordinates and speed values ​​of each vehicle in the video. The program then iterates over each pile number of the marked lane line that is desired to be extracted, collecting data every 1 meter. The distance between each pile number and all the coordinates of the numbered vehicles extracted in the previous layer is calculated. This method iterates over each vehicle and each section to obtain the speed and acceleration information of each vehicle in each section.

5. The method for constructing a tunnel-interchange exit small clearance velocity model based on aerial photography identification according to claim 1, characterized in that: The single factor analysis is performed based on the data of vehicle speed and position information to determine that the speed model selection indicators are the circular curve radius, the distance to the tunnel position and the clear distance length, including the following steps: When determining the indicators for the speed model, each independent variable and speed are analyzed and evaluated. The speed dispersion index is represented by the standard deviation SD, the average adjacent vehicle speed difference ASD, and the speed variation coefficient η. The speed variation coefficient η refers to the ratio of the speed standard deviation to the average speed. According to the degree of discreteness of each indicator, the radius of the circular curve is selected R , distance from the tunnel location D , Clearance Length L as independent variables required to build the model.

Images