Intersection turning radius measurement and evaluation method based on refined trajectory

Abstract

The invention discloses an intersection turning radius measurement and evaluation method based on a refined motor vehicle track by adopting a method of combining video data of an unmanned aerial vehicle with deep learning, and belongs to the technical field of intelligent traffic. Intersection road section traffic flow video data are obtained through aerial photography of an unmanned aerial vehicle, a basic data set is constructed, and a vehicle detection tracking model is constructed to obtain motor vehicle trajectory data; a refined motor vehicle track is output by using a data denoising and missing value processing method based on Kalman filtering, interpolation supplement, mean translation and other methods; fitting motor vehicle track coordinate points by using a Bezier curve, obtaining a fitted track line, calculating the curvature of each point location, and determining the curvature of each point location of the track curve; establishing a space-time motion model of the vehicle in the intersection, determining a motion time midpoint and a space midpoint according to the model, and determining the turning radius of the boundary arc track through the three points; and based on the turning radius detection result, evaluating whether the intersection turning radius design is reasonable.

Description

Technical Field

[0001] The invention relates to the field of intelligent transportation technology, and in particular to a method for measuring and evaluating an intersection turning radius based on a refined trajectory. Background Art

[0002] The vehicle driving environment on urban roads is complex. Intersections are locations where traffic accidents and traffic congestion frequently occur, and their accident rate and congestion level are much higher than those of ordinary roads. As an important technical indicator of intersections, turning radius directly affects the size of the intersection, the safety of vehicle traffic, the efficiency of pedestrian crossing, etc., and thus affects the traffic safety and traffic capacity of the intersection. The design of the turning radius of the intersection needs to comply with urban planning and road design specifications. Conventional intersection designs are mostly based on simple geometric models, and often fail to analyze the turning trajectory of the intersection. In the current era of rapid traffic development, the rationality of its design needs to be further verified. Therefore, it is necessary to evaluate the turning radius design based on actual traffic flow, and further use it to solve the problems of safety hazards and traffic efficiency at intersections.

[0003] With the rapid development of deep learning technology and drone technology, clear and accurate vehicle trajectory data can be obtained based on drone aerial videos. Through motion model construction and refined processing of trajectory data, automatic calculation of intersection turning radius is realized, which greatly saves labor costs while ensuring detection accuracy and efficiency to optimize intersection design. Summary of the invention

[0004] The present invention adopts the method of combining UAV video data with deep learning to realize the automatic calculation of the turning radius of the intersection. The purpose is to realize the design evaluation of the turning radius of the intersection, solve the problems of safety hazards and traffic efficiency of the intersection, and improve the overall safety of the intersection design.

[0005] In order to achieve the above content, the present invention proposes a method for measuring and evaluating the intersection turning radius based on refined vehicle trajectory data, and the specific steps are as follows:

[0006] S1: Construct a vehicle trajectory dataset and obtain traffic flow videos collected by drones as the basic dataset;

[0007] S2: Build a vehicle detection and tracking model to track the target vehicle and obtain the original vehicle trajectory data;

[0008] S3: Refined processing of vehicle trajectories, using a combination of methods based on Kalman filtering, interpolation supplementation, mean shift, etc. to perform trajectory data denoising and missing value processing, and verifying trajectory accuracy and completeness with the help of annotation software;

[0009] S4: Calculate the turning radius, use Bezier curve to fit the fine trajectory coordinate points, get the fitted trajectory line and calculate the curvature of each point; build a spatiotemporal model of vehicle movement inside the intersection, determine the coordinates of the time midpoint and space midpoint of the movement trajectory, and determine the turning radius of the limit circle by three feature points;

[0010] S5: Construct an intersection evaluation method to evaluate whether the intersection design is reasonable through the numerical results of turning radius detection.

[0011] Furthermore, the step S1 includes:

[0012] S101. Collect data using a drone to obtain traffic flow video and image information at an intersection.

[0013] S102. Use labelimg software to calibrate the traffic flow data image, mainly distinguishing vehicle types, as the basic data set.

[0014] Furthermore, the step S2 includes:

[0015] S201, preferably YOLOv8 model is used for vehicle detection, and the calibrated images and corresponding labels are used as training sets to train the YOLOv8 model. The traffic flow images of the intersection to be detected are used as test sets, and the object detection generates bounding boxes and type information.

[0016] S202, optimizing the DeepSORT algorithm to achieve stable tracking and identity maintenance of vehicle targets between multiple frames, and outputting original vehicle trajectory information.

[0017] Furthermore, the step S3 includes:

[0018] S301, a trajectory data denoising method based on classification and recognition of noise point features, using a mean shift method to solve the problem of trajectory detection frame drift.

[0019] S302, based on the trajectory completion method of identifying missing trajectory features, the linear interpolation and Kalman filter synthesis method is used to solve the identity jump problem.

[0020] Furthermore, the step S4 includes:

[0021] S401. Using the results of detection and data cleaning, the detailed trajectory coordinates of each vehicle can be obtained, and then the turning area of ​​the intersection can be determined. Based on research experience, the instantaneous turning radius of 70 meters is used as the basis for determination. Assuming that the actual distance between the two targets is L1 (meters), and there is a length of L2 (pixels) in the image, the starting point coordinates P0 (x0, y0) of the vehicle entering the intersection are determined by the following formula:

[0022] L1*d 2=L2*9800*(1-cosθ)

[0023]

[0024] Among them, (x1, y1) are the coordinates of adjacent frames, d is the pixel distance between adjacent frames, θ is the instantaneous turning angle of the starting trajectory in the intersection,

[0025] Determine the end point coordinates P S (x s ,y s ):

[0026]

[0027] Among them, (x s-1 ,y s-1 ) are the adjacent frame coordinates, d s is the pixel distance between adjacent frames, θ s is the instantaneous turning angle of the terminal trajectory in the intersection,

[0028] S402, a method for defining a prominent turning area is proposed, which is composed of an area surrounded by three trajectory feature points: a time midpoint, a space midpoint, and a curvature extreme point inside the intersection:

[0029] Define the time midpoint P n , corresponding to the frame number T n That is, the time when the vehicle enters the intersection T0 and the time when the vehicle leaves the intersection T S The mean of

[0030] Define the midpoint P in space k , which is the starting point P0 of the vehicle entering the intersection and the ending point P S The track coordinate point that the connecting line is closest to. Point P k (x k (t),y k The formula for determining the parameter t value in (t)) is as follows:

[0031]

[0032] Among them, (x1, y1), (x2, y2), and (x3, y3) are trajectory control points, which are determined by trajectory data.

[0033] S403, using the three trajectory feature points in the turning protruding area, build a bounding circle model and calculate the turning radius R:

[0034]

[0035] Among them, d1, d2, and d3 are the side lengths of the triangle formed by the three points.

[0036] The beneficial effects of the present invention are:

[0037] 1. Drones can quickly cover a large area of ​​road surface. Using drones to collect vehicle trajectory data is more flexible and efficient, and the plane traffic flow data is very intuitive.

[0038] 2. The model structure used in the present invention is lightweight and has a fast calculation speed. At the same time, it adopts an intelligent raw data refinement processing method to quickly obtain accurate calculation results.

[0039] 3. The present invention detects the turning radius of intersections based on field traffic flow data and evaluates the rationality of intersection design. The research results can be used to optimize regional traffic planning standards, such as the recommended value of turning radius, which is beneficial for traffic managers to carry out regional management planning.

[0040] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0042] Figure 1 It is a flow chart of intersection turning radius measurement and evaluation based on refined trajectory of the present invention;

[0043] Figure 2 It is a schematic diagram of characteristic points based on the intersection turning vehicle motion model proposed by the present invention.

[0044] Figure 3 It is a schematic diagram of a turning radius calculation method based on vehicle motion characteristics in a prominent turning area of ​​an intersection proposed by the present invention. DETAILED DESCRIPTION

[0045] The technical scheme in the embodiment of the present invention is described clearly and completely below in conjunction with the drawings in the embodiment of the present invention. It should be noted that the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0046] The present invention designs a method for measuring and evaluating the turning radius of an intersection based on refined vehicle trajectory data, and the specific implementation method is as follows:

[0047] Step 1: Build a basic data set. Obtain drone aerial videos, including but not limited to traffic flow videos of various sections of the road at intersections, and use labelimg software to calibrate the traffic flow data set, mainly distinguishing vehicle types, as the basic data set.

[0048] Step 2: Build a vehicle detection and tracking model. Build a YOLOv8 model for vehicle detection, and use calibrated images and corresponding labels as training sets to train the YOLOv8 model. Use the traffic flow image of the intersection to be detected as the test set, and use the model to extract vehicle features in the traffic flow image, including important visual features such as shape and texture. The detection model identifies the target position in the image and generates the target bounding box and type information. Through the DeepSORT algorithm, for each detected vehicle target, the extracted features are matched with the known vehicle features in the previous frame, and by estimating the vehicle's position and speed, stable tracking and identity maintenance of vehicle targets between multiple frames are achieved, and the original vehicle trajectory information of the intersection to be detected is obtained, including determining the coordinate points of the target bounding box and the target type.

[0049] Step 3: Refined processing of vehicle trajectories. Affected by factors such as model detection accuracy, the original traffic flow data obtained has certain errors. The goal of refined processing of vehicle trajectories is to achieve trajectory data denoising and missing value completion, so that the vehicle trajectory data is continuous and error-free.

[0050] Trajectory data denoising method based on classification and identification of noise point features: Trajectory noise points are divided into two categories.

[0051] The first category is the trajectory drift point. Since the trajectory point is determined as the center position of the detection bounding box, the trajectory drift is mainly caused by the jitter of the detection box. The distance difference between the target frame trajectory point and the adjacent frame is checked. If the distance difference between the two adjacent frames is too large, it is judged as the drift of the detection bounding box. After locating the noise point, the coordinates of the noise point are removed from the coordinate file. The mean shift method is used to calculate the coordinate mean of the local frame segment of each null value point, and the null value is filled with the mean value to achieve the purpose of smoothing the trajectory. It is suitable for removing small-amplitude trajectory detection frame jitter.

[0052] The second type of noise is background noise. The intersection environment is complex. In addition to traffic flow, there is a lot of background noise, such as ground signs, parked vehicles, roadside facilities, etc. The processing method is to detect the starting frame and end frame coordinates of each identity in the coordinate file and eliminate targets with small displacement range.

[0053] Trajectory completion method based on identifying missing trajectory features: Missing values ​​are divided into two categories. The first category is the missing trajectory values ​​of the detected target trajectory in some frame segments, which is caused by the model detection accuracy and the target being blocked or moving quickly in the actual scene. The method for detecting missing values ​​is to extract the corresponding frames of vehicle data for each identity in the coordinate file, sort the frames, and detect the integrity of the frames. In the case of frame jumps, the missing values ​​are supplemented by linear interpolation. The coordinate data supplement method is the same as the mean shift method.

[0054] The second type of missing values ​​is the missing of some trajectory values ​​caused by the identity jump of the detection target. Affected by the detection accuracy, the same detection target has identity jumps in different frame segments, that is, the same vehicle has different identity identifications at different time periods, and the trajectory detection result of a vehicle is manifested as the trajectory connection or intermittent connection of two different vehicles. After the first type of missing value completion processing, the frame number information of each identical target is extracted in the coordinate file, and the targets with incomplete trajectories are filtered out, that is, the starting frame or end frame of the target does not meet the boundary area of ​​the image screen. For a 1920×1080 pixel image, taking the vehicle target appearing on the left side of the image and leaving the top of the image as an example, the starting point x of the target trajectory (x, y) should be close to 0, and the ending point y should be close to 0. Considering the value range of different types of vehicle lengths, a margin of 200 to 400 is left for the initial and final coordinate values. Based on this benchmark, multiple trajectories with identity jumps are screened out, and the screened trajectories are put into a new coordinate file. Starting from the end frame of the previous trajectory, the Kalman filter is applied to predict the state and error covariance of the next frame rate based on the past trajectory state, and the coordinates of the starting frame of the next trajectory that match the Kalman filter estimated position are searched within a certain frame number range in the new coordinate file. If the predicted coordinates and the coordinates of the starting frame of a certain trajectory meet the error range, an associated match is performed to replace the same identity information.

[0055] Verify trajectory accuracy and completeness with labeling software: With labelimg software, users can easily select objects and add and modify labels through an intuitive graphical user interface, and verify the refined trajectory in combination with the trajectory coordinate file.

[0056] Step 4: Calculate the turning radius of the intersection. According to the complete traffic flow trajectory data obtained, fit the trajectory curve and calculate the curvature of each point. The vehicle turning motion at the plane intersection should be fitted with a third-order Bezier curve. The Bezier curve parameter equation is as follows: B(t) = (1-t) 3 *P0+3t*(1-t) 2 *P1+3t 2 *(1-t) 2 *P2+t 3 *P3

[0057] Where t∈(0,1), P0, P1, P2, and P3 are four trajectory control points, which are selected as the internal trajectory points of the intersection. The curvature calculation formula is as follows:

[0058]

[0059] Where B′(t) and B"(t) represent the first-order derivative and the second-order derivative of B(t), respectively, and × represents the vector product.

[0060] The curvature extreme point is determined by K'(t) = 0, and the pixel coordinate point P is converted by the corresponding t value m (x m ,y m ), the formula is as follows:

[0061] x(t)=(1-t) 3 *x0+3t(1-t) 2 *x1+3t 2 (1-t)*x2+t 3 *x3

[0062] y(t)=(1-t) 3 *y0+3t(1-t) 2 *y1+3t 2 (1-t)*y2+t 3 *y3

[0063] In each image frame, assume that the actual distance between the two targets is L1 (meters), and the length in the image is L2 (pixels). Let (x0, y0) represent the first trajectory point of the extracted target, (x p ,y p ) represents the target trajectory point extracted in the next p-th frame, then the target is from (x0, y0) to (x p ,y p The average speed V (unit: km / h) of the interval is calculated by the following formula:

[0064]

[0065] where d p Indicates the distance between two points in pixels. F represents the frame rate.

[0066] Then the target instantaneous velocity V at point i is i (Unit: km / h) can be calculated by the following formula:

[0067]

[0068] where d i Indicates the pixel distance between two adjacent points.

[0069] Due to the speed limit for vehicles entering the intersection, according to research experience, it is practical when the vehicle turning radius is less than 70 meters. Assuming that the vehicle turns left at the flat intersection, the coordinates of the starting point P0 (x0, y0) of the vehicle entering the intersection are determined by the following formula:

[0070] L1*d 2 =L2*9800*(1-cosθ)

[0071]

[0072] Among them, (x1, y1) are the coordinates of adjacent frames, d is the pixel distance between adjacent frames, θ is the instantaneous turning angle of the starting trajectory in the intersection,

[0073] Determine the end point coordinates P S (x s ,y s ):

[0074]

[0075]

[0076] Among them, (x s-1 ,y s-1 ) are the adjacent frame coordinates, d s is the pixel distance between adjacent frames, θ s is the instantaneous turning angle of the terminal trajectory in the intersection,

[0077] After determining the starting and ending points of the trajectory inside the intersection, the frame numbers T0 and T1 corresponding to the starting and ending point coordinates can also be determined. S .

[0078] In the study of vehicle turning characteristics, the turning protrusion area is an important interval to describe the safety of vehicle turning. Starting from the actual traffic flow trajectory, according to the geometric characteristics of vehicle turning, the turning protrusion area involved in calculating the turning radius in this example is defined. Build a spatiotemporal model of vehicle movement inside the intersection, such as Figure 2 As shown, the coordinates of the midpoint in time and the midpoint in space of the motion trajectory are determined.

[0079] Determine the frame number T corresponding to the midpoint of time n , that is, the time when the vehicle enters the intersection T0 and the time when the vehicle leaves the intersection T S The mean value of T0, T S Determine the midpoint T of the vehicle's movement time in the intersection n , Corresponding frame number T n Determine the trajectory coordinates P n .

[0080] Determine the midpoint P in space k ,like Figure 2 As shown, it is the starting point P0 of the vehicle entering the intersection and the ending point P of the vehicle exiting the intersection. S The closest track coordinate point. For arc track, the midpoint of the straight line from the start to the end is P' k The corresponding arc point P k To confirm, click P k (x k (t),y k The formula for determining the parameter t value in (t)) is as follows:

[0081]

[0082] The turning prominent area is composed of the area surrounded by three trajectory feature points: the time midpoint, space midpoint, and curvature extreme point inside the intersection. The turning radius is determined by using three trajectory feature points, such as Figure 3 As shown, d1, d2, and d3 are the side lengths of the triangle formed by the three points, and the radius R (pixel) of the bounding circle is determined by the following formula:

[0083]

[0084] The turning radius r (meters) is determined as:

[0085]

[0086] Step 5: Design and evaluation of intersection turning radius. The value range of intersection turning radius has clear specifications and recommended values. The design speed of the grade intersection is given in the range of the main road speed of different intersecting roads. The speed of the section is between 20km / h and 60km / h. The design speed in the grade intersection should be 0.5 to 0.7 times that of the section.

[0087] Taking into account the minimum turning radius that needs to be considered when planning fire lanes, the minimum turning radius evaluation is set to 12m to ensure that fire trucks can turn and operate safely and efficiently.

[0088] The system stores a recommended table of turning radius and design speed (as shown in the following table) as a basis for rationality verification. According to step 4, the instantaneous speed of various types of vehicles entering the intersection can be obtained using the vehicle trajectory calculation results. The calculation results of the turning radius are compared with the ideal turning radius to output the value evaluation.

[0089] Design turning speed (km / h) 30 25 20 15 Recommended radius (m) 25 20 15 12

[0090] The turning radius evaluation can solve two problems for existing intersections: one is to obtain the reasonable turning radius required for various types of motor vehicles to travel within the intersection; the other is to verify the turning capacity of various types of vehicles at the current intersection.

[0091] Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for measuring and evaluating intersection turning radius based on refined trajectory, characterized in that: The steps include: S1: Construct a vehicle trajectory dataset and obtain traffic flow videos collected by drones as the basic dataset; S2: Build a vehicle detection and tracking model to track the target vehicle and obtain the original vehicle trajectory data; S3: Refined processing of vehicle trajectories, using a combination of methods based on Kalman filtering, interpolation supplementation, mean shift, etc. to perform trajectory data denoising and missing value processing, and verifying trajectory accuracy and completeness with the help of annotation software; S4: Calculate the turning radius, use Bezier curve to fit the fine trajectory coordinate points, get the fitted trajectory line and calculate the curvature of each point; build a spatiotemporal model of vehicle movement inside the intersection, determine the coordinates of the time midpoint and space midpoint of the movement trajectory, and determine the turning radius of the limit circle by three feature points; S5: Construct an intersection evaluation method to evaluate whether the intersection design is reasonable through the numerical results of turning radius detection.

2. The method for measuring and evaluating intersection turning radius based on refined trajectory according to claim 1, characterized in that: In step S3, a trajectory data denoising method based on classification and identification of noise point features is proposed, and the mean shift method is applied to fill in the null values ​​to achieve the purpose of smoothing the trajectory, effectively removing background noise and small amplitude trajectory detection frame jitter problems.

3. The method for measuring and evaluating intersection turning radius based on refined trajectory according to claim 1, characterized in that: In step S3, a trajectory completion method based on identifying missing trajectory features is proposed. The trajectory state of future frames is predicted by combining linear interpolation with Kalman filtering, and identity matching is achieved by associating matching coordinate errors, which effectively solves the problem of trajectory identity jump.

4. The method for measuring and evaluating intersection turning radius based on refined trajectory according to claim 1, characterized in that: In step S4, the detailed trajectory coordinates of each vehicle can be obtained by using the results of detection and data refinement and cleaning, and then the turning area of ​​the intersection can be determined. Assuming that the actual distance between the two targets is L1 (meters), there is a length L2 (pixels) in the image, and the starting point coordinates P0 (x0, y0) of the vehicle entering the intersection are determined by the following formula: L1*d 2 =L2*9800*(1-cosθ) Among them, (x1, y1) are the coordinates of adjacent frames, d is the pixel distance between adjacent frames, θ is the instantaneous turning angle of the starting trajectory in the intersection, Determine the end point coordinates P S (x s ,y s ): Among them, (x s-1 ,y s-1 ) are the adjacent frame coordinates, d s is the pixel distance between adjacent frames, θ s is the instantaneous turning angle of the terminal trajectory in the intersection, 5. The method for measuring and evaluating intersection turning radius based on refined trajectory according to claim 1, characterized in that: In the step S4, a prominent turning area is defined, which is composed of an area surrounded by three trajectory feature points, namely, a time midpoint, a space midpoint, and a curvature extreme point inside the intersection. The limit turning radius is determined by a three-point formula, thereby ensuring the accuracy of the detection result.

6. The method for measuring and evaluating intersection turning radius based on refined trajectory according to claim 1, characterized in that: The system stores the recommended table of turning radius and design speed. As a basis for rationality verification, the vehicle trajectory calculation results are compared with the ideal turning radius. Continuous traffic flow trajectory detection and turning radius evaluation can be achieved.

Images