Method for evaluating highway alignment quality based on visual continuity

Abstract

The application provides a highway line quality evaluation method based on visual continuity, which comprises the following steps: (1) obtaining a two-dimensional design preliminary scheme of a road, including a horizontal section design file and a longitudinal section design file; (2) performing perspective transformation on the two-dimensional design preliminary scheme to obtain a driver's visual angle line; (3) inputting the driver's visual angle line into a variational autoencoder to obtain perspective characteristic parameters; and (4) performing visual continuity and expected speed coordination evaluation according to the perspective characteristic parameters to evaluate the highway line quality. The application evaluates the visual continuity and the running speed coordination of the highway line based on the horizontal visual parameters and the longitudinal visual parameters of the highway line, and evaluates the highway line quality, so that the defects of the single visual angle actual method are made up, and the overall design quality and level of the road geometry are improved.

Description

Technical Field

[0001] This invention relates to the field of road evaluation technology, and specifically to a method for evaluating the alignment quality of highways based on visual continuity. Background Technology

[0002] my country's transportation infrastructure is densely networked, and the quality of road construction has been improving year by year. Road geometry design is a crucial aspect of road construction. Whether road geometry meets driver expectations, satisfies vehicle kinematic requirements, and adapts to the surrounding environment directly affects driving safety and comfort. Reasonable road geometry not only effectively ensures traffic safety during road operation and reduces injuries and losses caused by traffic accidents, but also significantly reduces the cost of safety improvements on dangerous road sections.

[0003] Current highway alignment design methods decompose roads into basic element units such as horizontal, vertical, and cross-sectional planes, and then obtain the three-dimensional centerline of the highway through horizontal and vertical combination. Traditional two-dimensional road geometry design methods are clear in their approach, have complete relevant standards and specifications, and are convenient to operate. However, on the other hand, two-dimensional design methods weaken the consideration of the three-dimensional spatial characteristics of road geometry and pay insufficient attention to the relationship between driver-perceived geometric information and behavioral responses. This can lead to situations where even if the horizontal and vertical alignment elements meet design requirements, improper combination of alignment design may still result in discontinuities in the three-dimensional space of the road, or distortion or discontinuity of geometric information observed from the driver's perspective, ultimately manifesting as poor speed coordination and high driving risks. The key to this problem lies in the fact that traditional design methods only focus on two-dimensional road geometry design elements, lacking sufficient consideration of design indicators and requirements from different perspectives of road geometry.

[0004] Different perspectives of road geometry represent different descriptions of the same geometric object, and these descriptions are interconnected, reflecting varying design requirements. How to consider driver visual continuity, establish effective methods for evaluating highway alignment quality, and improve the quality and level of road geometry design presents a new challenge for road design.

[0005] Scholars have gradually recognized the shortcomings of the horizontal and vertical separation design method and proposed a series of road safety assessment methods and safeguards based on this method. Road geometry is inherently three-dimensional and continuous. Related literature studies the characteristics of road geometry under three-dimensional conditions to compensate for the deficiencies of traditional design methods, such as calculating three-dimensional sight distances and establishing three-dimensional road geometry models. These studies often focus on establishing accurate mathematical calculation models, lacking exploration of the quality of three-dimensional road geometry design and alignment design methods. Road geometry is presented in perspective from the driver's point of view. Whether the road information supply is coordinated with the driver's visual perception affects driving comfort and safety. Domestic and international research on road geometry from the driver's perspective mainly includes studies on the visual effects of combined alignments under traditional design methods, as well as studies on perspective alignment description methods. As road geometry is information directly perceived by the driver, how to comprehensively evaluate its coordination and establish effective design methods remains to be explored.

[0006] To address the issues of insufficient coordination in existing road geometry and inconsistencies between the actual alignment and the driver's visual environment, a road alignment quality evaluation method based on visual continuity is proposed. This method compensates for the shortcomings of single-view practical methods and improves the overall quality and level of road geometry design. Summary of the Invention

[0007] To address the problems of poor coordination between existing highway alignment design and operating speed, and high driving risks, this invention provides a highway alignment quality evaluation method based on visual continuity. Based on the lateral and longitudinal visual parameters of the highway alignment, the method assesses the visual continuity and operating speed coordination of the highway alignment, thereby evaluating the quality of the highway alignment.

[0008] The technical problem to be solved by this invention is achieved by the following technical solution:

[0009] A method for evaluating highway alignment quality based on visual continuity includes the following steps:

[0010] (1) Obtain preliminary two-dimensional design schemes for the road, including horizontal and vertical profile design documents;

[0011] (2) Perform perspective transformation on the preliminary two-dimensional design to obtain the driver's perspective line shape;

[0012] (3) Input the driver’s perspective line shape into the variational autoencoder to obtain perspective feature parameters;

[0013] (4) Based on the perspective feature parameters, conduct visual continuity and expected speed coordination assessment to evaluate the quality of highway alignment.

[0014] Further technology of the present invention:

[0015] Preferably, step (2): perform perspective transformation on the preliminary two-dimensional design scheme. Calculate the driver's perspective geometry from two-dimensional geometry. First, the two-dimensional geometry needs to be transformed into three-dimensional geometry, and then the perspective transformation is performed. The basic idea is to calculate the X and Y coordinates of each three-dimensional point through the planar design parameters, and calculate the Z coordinate of each point through the longitudinal section design parameters. Combine the two calculation results to obtain the three-dimensional coordinates (X, Y, Z) of each point. Perspective transformation is a linear transformation. To convert points on the road route to the driver's perspective, two coordinate system transformations are required. The first coordinate transformation converts the points from the three-dimensional world coordinate system to the three-dimensional camera coordinate system. This process mainly includes translation and rotation. The second coordinate transformation uses the pinhole imaging principle to convert the points in the three-dimensional camera coordinate system to the perspective coordinate system, which is a transformation from three-dimensional to two-dimensional. After perspective transformation, the driver's perspective linear shape is obtained.

[0016] Preferably, step (3) perspective feature parameter extraction: use unsupervised machine learning algorithm to analyze the geometric shape of the driver's perspective, use variational autoencoder to reduce the dimension of the driver's perspective line shape, and extract feature parameters. There are two parameters. The first parameter is the longitudinal visual parameter, which describes the longitudinal slope of the lane line and establishes a quantitative description method of the driver's perspective geometry. The second parameter is the lateral visual parameter, which describes the curvature of the lane.

[0017] Variational autoencoders (VAEs) are an extension of autoencoders. A VAE is an unsupervised neural network model consisting of an encoder and a decoder connected by a hidden layer. When data is input into the neural network, the encoder first compresses the data, outputting a lower-dimensional data distribution, commonly known as the hidden layer distribution. Then, a set of parameters, called hidden layer parameters or feature parameters, is randomly sampled from the hidden layer distribution. This set of parameters is input into the decoder to increase the dimensionality of the data, outputting reconstructed data. The training objective of a VAE is to minimize the deviation between the input data and the reconstructed data, i.e., to minimize the reconstruction error; simultaneously, the hidden layer distribution must be constrained to conform to a Gaussian distribution.

[0018] This study investigates road geometry from the driver's perspective. Using perspective geometric images as input, feature parameters describing the geometric shape from the driver's perspective are extracted. A dataset containing various perspective geometric variations is established. First, various horizontal and vertical combination lines are designed according to specifications. Then, perspective transformation methods are used to generate various perspective geometric images as training data.

[0019] After training the driver's view road geometry, the variational autoencoder maps the geometric data to a low-dimensional space. Each driver's view road geometry uniquely corresponds to two visual parameters. Because the algorithm imposes regular constraints on the hidden layer, the low-dimensional space can effectively reflect the characteristics of the original data. The two feature parameters extracted in this paper are c1, which reflects the longitudinal geometric feature of the driver's view line shape, and c2, which reflects the lateral geometric feature of the driver's view line shape. These two visual parameters are quantitative description indicators of the driver's view geometric features, laying a theoretical foundation for subsequent driver's view line shape evaluation and design. In addition, as shown in the previous analysis, the decoder of the variational autoencoder has the function of a generator. Given a set of valid feature parameters, the decoder can generate the corresponding driver's view road geometry perspective shape. Even if the line shape has not appeared in the training set, it can still generate valid lane lines. Therefore, by modifying the feature parameters of a certain perspective line shape, the driver's view line shape can be modified to make the lane perspective effect more in line with the driver's needs.

[0020] The preferred process for generating the dataset is as follows:

[0021] First, design horizontal curves with various radii, and then combine these horizontal curves with various longitudinal profile designs to form different combined line shapes.

[0022] Then, a series of three-dimensional points are obtained by sampling at 10m intervals along the line. The simulated camera is placed on each three-dimensional point at a height from the ground that is the driver's viewpoint height. Perspective transformation is then used to generate a series of road geometric perspective images.

[0023] Preferably, step (4) highway alignment quality evaluation: is divided into two aspects: visual continuity of road geometry and coordination of running speed;

[0024] Visual continuity is assessed by analyzing the perspective lines of various unfavorable and favorable line combinations, calculating the visual parameters under each combination condition, comparing and discussing the visual parameters of good and poor continuity, transforming the qualitative indicators in the standard into quantitative indicators, focusing on the differences in visual parameters between unfavorable and good line combination combinations, establishing a visual continuity assessment model, and judging the dynamic continuity of visual lines.

[0025] To address the issue of operational speed coordination, a road segmentation method based on speed change characteristics is proposed. Using the V85 speed map as a basis, abrupt change point detection is employed to divide the road into different operational speed segments, thereby verifying the continuity of operational speed. Piecewise linear regression can divide the speed change map into multiple parts, each representing a different operational speed change trend. The location where the speed trend changes is the segmentation point. Operational speed assessment includes three aspects: coordination between operational speed and design speed, coordination between adjacent operational speed segments, and coordination of operational speed distribution. The operational speed prediction model is based on a long short-term memory neural network speed prediction model.

[0026] Beneficial effects:

[0027] Unsupervised machine learning algorithms were used to analyze the road geometry features from the driver's perspective. A variational autoencoder algorithm was employed to analyze the driver's perspective geometry model. This algorithm uses an unsupervised neural network to reduce the dimensionality of the image, ultimately extracting two feature values ​​as visual parameters. These visual parameters express the main features of the road geometry from the driver's perspective, such as curve turning direction, perspective curvature, and visual longitudinal slope. A set of visual parameters uniquely determines a set of visual lane alignments. These visual parameters allow for the design and adjustment of the driver's perspective geometry. The extraction of these visual parameters provides theoretical support for the quantitative analysis, evaluation, and optimization design of road geometry from the driver's perspective.

[0028] This study investigated road geometric safety and established a quantitative evaluation method. The significance of spatial geometric continuity was analyzed, along with the impact of combined alignments on the attenuation of spatial curve continuity. The perspective effects of different combined alignments were discussed, and alignments were categorized based on design principles. Using visual parameters as input, a visual continuity evaluation model was established based on a support vector machine algorithm. This model can quantitatively evaluate horizontal and vertical alignment combinations. Ultimately, road geometric design should ensure that design elements are compatible with desired speeds, guaranteeing speed coordination. Using visual parameters as input, a speed prediction model based on a long short-term memory neural network was established, and a quantitative evaluation method was proposed to address speed coordination requirements. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a highway alignment quality evaluation method based on visual continuity.

[0031] Figure 2 This is a schematic diagram of the perspective transformation of lane lines;

[0032] Figure 3 This is a schematic diagram for calculating plane coordinates;

[0033] Figure 4 It is a variational automatic encoder structure;

[0034] Figure 5 Generate a flowchart for perspective images;

[0035] Figure 6 Generate a schematic diagram for the perspective image. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] like Figure 1 A method for evaluating the alignment quality of highways based on visual continuity includes the following steps:

[0038] (1) Obtain preliminary two-dimensional design schemes for the road, including horizontal and vertical profile design documents;

[0039] (2) Perform perspective transformation on the preliminary two-dimensional design to obtain the driver's perspective line shape;

[0040] (3) Input the driver’s perspective line shape into the variational autoencoder to obtain perspective feature parameters;

[0041] (4) Based on the perspective feature parameters, conduct visual continuity and expected speed coordination assessment to evaluate the quality of highway alignment.

[0042] Specific steps:

[0043] (1) First, obtain the preliminary two-dimensional design documents for the road, including the horizontal and vertical profile design documents;

[0044] (2) Figure 2 Perspective Transformation: Then, a perspective transformation is performed on the preliminary two-dimensional design. The driver's viewpoint geometry is calculated from the two-dimensional geometry. First, the two-dimensional geometry needs to be converted to three-dimensional geometry before the perspective transformation is performed. The basic idea is to calculate the X and Y coordinates of each three-dimensional point using planar design parameters, such as... Figure 3As shown, the Z-coordinate of each pile point is calculated using the longitudinal profile design parameters. Combining the results of both calculations yields the three-dimensional coordinates (X, Y, Z) of each pile point. Perspective transformation is a type of linear transformation. A schematic diagram illustrating the transformation of the three-dimensional lane lines to the driver's perspective is shown below. Figure 2 As shown, converting points on the road route to the driver's perspective requires two coordinate system transformations. The first transformation converts the points from the three-dimensional world coordinate system to the three-dimensional camera coordinate system, which mainly involves translation and rotation. The second transformation uses the pinhole imaging principle to convert the points in the three-dimensional camera coordinate system to the perspective coordinate system, which is a transformation from three-dimensional to two-dimensional. After the perspective transformation, the line shape from the driver's perspective is obtained.

[0045] (3) Perspective feature parameter extraction: Unsupervised machine learning algorithm is used to analyze the geometric shape of the driver's perspective. Variational autoencoder is used to reduce the dimension of the driver's perspective line shape and extract feature parameters. There are two parameters: the first parameter is the longitudinal visual parameter, which describes the longitudinal slope of the lane line and establishes a quantitative description method of the driver's perspective geometry; the second parameter is the lateral visual parameter, which describes the curvature of the lane.

[0046] Variational autoencoders (VAEs) are an extension of autoencoders (AEs). A VAE is an unsupervised neural network model, and its model structure is as follows: Figure 4 As shown, a variational autoencoder consists of an encoder and a decoder, connected by a hidden layer. When data is input into the neural network, the encoder first compresses the data, outputting a lower-dimensional data distribution, usually called the hidden layer distribution. Then, a set of parameters, called hidden layer parameters or feature parameters, are randomly sampled from the hidden layer distribution. This set of parameters is input into the decoder to increase the dimensionality of the data and output the reconstructed data. The training objective of a variational autoencoder is to minimize the deviation between the input data and the reconstructed data, i.e., to minimize the reconstruction error; it also needs to constrain the hidden layer distribution to conform to a Gaussian distribution. Because the training data does not require labels but is trained using itself as a reference, a variational autoencoder is an unsupervised neural network.

[0047] This study investigates road geometry from a driver's perspective, using perspective geometric images as input to extract feature parameters describing the geometric shape from the driver's point of view. Therefore, a dataset containing various perspective geometric variations is required. Considering that currently operating roads generally conform to route design specifications and are designed with horizontal and vertical alignments separated, resulting in diverse perspective geometric shapes under different horizontal and vertical combinations, we first design various horizontal and vertical alignment combinations according to the specifications. Then, we use perspective transformation methods to generate various perspective geometric images as training data.

[0048] The flowchart and diagram for generating the dataset are as follows: Figure 5-6 As shown, firstly, horizontal curves with various radii are designed, and these curves are combined with various longitudinal profile designs to form different combined line shapes. Then, a series of three-dimensional points are obtained by sampling at equal intervals (10m) along the line shape. A simulated camera is placed at each three-dimensional point, at a height from the ground equal to the driver's viewpoint, and perspective transformation is used to generate a series of road geometric perspective images.

[0049] After training the variational autoencoder on the driver's perspective road geometry, it maps the geometric data to a low-dimensional space, where each driver's perspective road geometry uniquely corresponds to two visual parameters. Because the algorithm imposes regular constraints on the hidden layer, the low-dimensional space effectively reflects the characteristics of the original data. The two feature parameters extracted in this paper are c1, reflecting the longitudinal geometric feature of the driver's perspective line shape, and c2, reflecting the lateral geometric feature of the driver's perspective line shape. These two visual parameters are quantitative descriptions of the driver's perspective geometric features, laying a theoretical foundation for subsequent driver's perspective line shape evaluation and design. Furthermore, as the preceding analysis shows, the decoder of the variational autoencoder acts as a generator. Given a set of valid feature parameters, the decoder can generate the corresponding driver's perspective road geometry perspective shape, even if the line shape has not appeared in the training set, it can still generate valid lane lines. Therefore, by modifying the feature parameters of a certain perspective line shape, the driver's perspective line shape can be modified to make the lane perspective effect more in line with the driver's needs.

[0050] (4) Highway Alignment Quality Evaluation: This is divided into two aspects: visual continuity of road geometry and coordination of operating speed. Visual continuity involves analyzing the perspective alignment of various unfavorable and favorable alignment combinations, calculating visual parameters under each combination condition, comparing and discussing visual parameters with good and poor continuity, transforming qualitative indicators in the specifications into quantitative indicators, focusing on the differences in visual parameters between unfavorable and good alignment combinations, establishing a visual continuity evaluation model, and judging the dynamic continuity of visual alignment. Operating speed coordination proposes a segment division method based on speed change characteristics. Using the V85 speed map as a basis, through abrupt change point detection, the road is divided into different operating speed segments, and this serves as the basis for verifying the continuity of operating speed. Piecewise linear regression can divide the speed change map into multiple parts, each representing a different operating speed change trend. The location where the speed trend changes is the segmentation point. Operating speed evaluation includes three aspects: coordination between operating speed and design speed, coordination between adjacent operating speed segments, and coordination of operating speed distribution. The operating speed prediction model is based on a long short-term memory neural network speed prediction model.

[0051] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The descriptions in the foregoing invention and specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the alignment quality of highways based on visual continuity, characterized in that, Includes the following steps: (1) Obtain preliminary two-dimensional design schemes for the road, including horizontal and vertical profile design documents; (2) Perform perspective transformation on the preliminary two-dimensional design to obtain the driver's perspective line shape; (3) Input the driver’s perspective line shape into the variational autoencoder to obtain perspective feature parameters; Perspective feature parameter extraction: Unsupervised machine learning algorithm is used to analyze the geometric shape of the driver's view. Variational autoencoder is used to reduce the dimensionality of the driver's view line shape and extract feature parameters. There are two parameters: the first parameter is the longitudinal visual parameter, which describes the longitudinal slope of the lane line and establishes a quantitative description method of the driver's view geometry; the second parameter is the lateral visual parameter, which describes the curvature of the lane. (4) Based on the perspective feature parameters, conduct an assessment of visual continuity and speed coordination to evaluate the quality of the highway alignment; Highway alignment quality evaluation is divided into two aspects: visual continuity of road geometry and coordination of operating speed. Visual continuity is assessed by analyzing the perspective lines of various unfavorable and favorable line combinations, calculating the visual parameters under each combination condition, comparing and discussing the visual parameters of good and poor continuity, transforming the qualitative indicators in the standard into quantitative indicators, focusing on the differences in visual parameters between unfavorable and good line combination combinations, establishing a visual continuity assessment model, and judging the dynamic continuity of visual lines. To address the issue of speed coordination, a road segmentation method based on speed variation characteristics is proposed. Using the 85 speed map as a basis, the road is divided into different speed segments by detecting abrupt change points, and the continuity of speed is then verified based on this method. Piecewise linear regression can divide the speed change map into multiple parts, each part representing a different operating speed change trend. The location where the speed trend changes is the segment point. The operating speed assessment includes three aspects: the coordination between the operating speed and the design speed, the coordination between adjacent operating speed segments, and the coordination of the operating speed distribution. The operating speed prediction model adopts a long short-term memory neural network speed prediction model.

2. The method for evaluating highway alignment quality based on visual continuity as described in claim 1, characterized in that, Step (2): Perform perspective transformation on the preliminary two-dimensional design scheme. Calculate the driver's perspective geometry from two-dimensional geometry. First, the two-dimensional geometry needs to be transformed into three-dimensional geometry, and then the perspective transformation is performed. The basic idea is to calculate the coordinates of each three-dimensional point (x, y) using the planar design parameters, and calculate the coordinates of each point (x) using the longitudinal section design parameters. Combine the two calculation results to obtain the three-dimensional coordinates (x, y, x) of each point. Perspective transformation is a linear transformation. To convert points on the road route to the driver's perspective, two coordinate system transformations are required. The first coordinate transformation converts the points from the three-dimensional world coordinate system to the three-dimensional camera coordinate system. This process mainly includes translation and rotation. The second coordinate transformation uses the pinhole imaging principle to convert the points in the three-dimensional camera coordinate system to the perspective coordinate system, which is a transformation from three-dimensional to two-dimensional. After perspective transformation, the driver's perspective line shape is obtained.

3. The method for evaluating highway alignment quality based on visual continuity as described in claim 1, characterized in that, Step (3): Study the road geometry from the driver's perspective. Using perspective geometric images as input, extract the feature parameters that describe the geometric shape from the driver's perspective and establish a dataset containing various perspective geometric transformations. First, design various horizontal and vertical combination lines according to the specifications, and then use perspective transformation methods to generate various perspective geometric images as training data. After training the variational autoencoder on the road geometry from the driver's perspective, the geometric data is mapped to a low-dimensional space. Each road geometry from the driver's perspective uniquely corresponds to two visual parameters: one reflects the longitudinal geometric feature of the line shape from the driver's perspective, and the other reflects the lateral geometric feature of the line shape from the driver's perspective. The two visual parameters are quantitative description indicators of the geometric features from the driver's perspective.

4. The method for evaluating highway alignment quality based on visual continuity as described in claim 3, characterized in that, The process of generating the dataset: First, design horizontal curves with various radii, and then combine these horizontal curves with various longitudinal profile designs to form different combined line shapes. Then, a series of three-dimensional points are obtained by sampling at 10m intervals along the line. The simulated camera is placed on each three-dimensional point at a height from the ground that is the driver's viewpoint height. Perspective transformation is then used to generate a series of road geometric perspective images.

Images