Available sight distance detection method considering incapable sun glare influence

Abstract

The invention discloses an available sight distance detection method considering the influence of disable sun glare. According to the method, the sight distance satisfaction rates of different positions in a road can be detected, a road model is constructed by moving laser radar scanning equipment to collect point clouds, observation points are extracted through radar scanning vehicle track points to simulate driver viewpoints, then sight lines are established to judge whether the sight lines are blocked or not, and the available sight distances of the different positions of the road are comprehensively detected. According to the method, accurate positioning of the glare position in the road and glare time prediction can be realized, specifically, the sun position is determined through the simulated light, the driver viewpoint is determined through the track point, and after the light incident driver visual recognition angle is calculated, light shielding is judged through the road model. According to the method, the sight distance attenuation rate model is constructed, compared with a subjective glare detection method, accurate analysis of the available sight distance of the driver under the disable glare condition is successfully achieved, the available sight distance is recalculated based on the available sight distance of the road and the glare position, the process is simple, and the result can be quantified.

Description

Technical Field

[0001] The invention relates to a method for testing available sight distance taking into account the influence of solar glare, and belongs to the field of transportation engineering. Background Art

[0002] Sun glare harms the driver's vision and traffic safety. Strong glare reduces the driver's available sight distance (ASD) and easily causes accidents. Compared with sections with sufficient sight distance, drivers are more likely to have accidents on sections with insufficient available sight distance. Drivers exposed to glare will be visually disturbed or even have temporary sight distance reduction, forming more risky positions with insufficient sight distance. Strong glare can lead to a reduction in the driver's sight distance. Affected drivers cannot fully identify road infrastructure, and limited sight distance is bound to lead to high driving risks. Therefore, drivers affected by glare are more likely to have accidents. Therefore, detecting the insufficient ASD position on the road, especially exploring the impact of strong glare on sight distance, and then solving the problem of quantitative analysis of available sight distance in glare areas, has important theoretical and application value for detecting driving risks. The research results can provide support for intelligent road warning and the formulation of road improvement measures.

[0003] Using laser point cloud to inspect ASD requires setting observation points (driver's eye position) and target points (driver's visual focus position) to simulate the driver's line of sight, and searching for obstacles by determining whether the obstacle point is in line with the line of sight. Based on the ASD model of a single line of sight, since the target point is set in the middle of the lane and the line of sight is mostly located in the lane as a whole, the detection results may miss some roadside obstacles and ignore the visibility of the lane boundary position, especially on rural roads where tree invasion is prone to occur. In order to more comprehensively evaluate the visibility of different positions on the road, three lines of sight are set for two-lane highways to detect the visibility of the boundary lines on both sides of the road and the middle position of the lane respectively.

[0004] The current glare analysis methods have certain deficiencies and are difficult to quantify the hazards of disabling glare. Due to the overexposure of the camera, the grayscale value of the image exceeds the normal range. The detection method based on the image brightness index cannot assess the hazard of disabling glare. Eye tracking technology cannot evaluate the degree of glare at different times. The detection method based on the three-dimensional road model ignores the key variable of sunlight intensity and is difficult to accurately define the severity of glare.

[0005] Therefore, in view of the impact of solar glare on the driver's visual range, the impact of disabling glare on the available visual range is difficult to evaluate. The present invention proposes a quantitative model of available visual range coupled with a glare factor, which can be used to analyze the change of the driver's available visual range under disabling glare conditions.

[0006] The present invention uses the measured LiDAR point cloud data on the road to introduce in detail a digital analysis method for available sight distance under disabling glare. This method first constructs an ASD model that adapts to the driving characteristics of two-lane highways. This model resets the driver's sight line and is used to calculate the available sight distance on two-lane highways; a solar glare detection method based on a three-dimensional road model is designed to solve the problem of detecting solar glare at different times and locations on the road; thirdly, a sight distance attenuation rate model is proposed to quantitatively analyze the impact of disabling glare on the driver's sight distance on the road, which is used to detect the position of disabling glare and recalculate the available sight distance, solving the quantification problem of available sight distance under disabling glare. Summary of the invention

[0007] The technical solution adopted by the present invention is to propose a method for testing the available viewing distance under disabling glare, comprising the following steps:

[0008] Step 1: Design an ASD calculation model for two-lane highways and calculate the available sight distance at different locations on the road

[0009] Build observer and driver vectors from trajectory points

[0010] Construct an ASD model suitable for two-lane roads and calculate the road ASD. It includes two steps: first, establish a driver observation point set and a corresponding driver vector set, then set the observation point / target point position, establish the line of sight and calculate the ASD according to the characteristics of two-lane roads with pedestrians and non-motor vehicles crossing the road.

[0011] Before calculating the available sight distance, establish the driver observation point set and the corresponding driver vector set for the two-lane highway. Considering the driver's eye height during vehicle driving, set the elevation value of all observation points to a certain value above the ground. Regarding the plane coordinates of the observation points, calculate the distances between adjacent trajectory points in turn, and accumulate the distances between adjacent trajectory points. When the accumulated value is greater than a certain threshold (km), use the plane coordinates of the trajectory point at this time as the observation point (O i , where i = 1, 2, ..., w, w is the total number of observation points in the data set). Construct the driver vector from the current observation point to the next observation point, and use the next observation point coordinate (O i+1 (x,y,z) minus the current observation point coordinates (O i (x,y,z)) Get the driver vector of the current position

[0012] Establishing sight lines and calculating ASD

[0013] The commonly used ASD detection model generally traverses all the point clouds ahead in the order of observation points / target points. Due to the large amount of point cloud data and low efficiency, in order to improve the calculation efficiency, three core sight lines are designed to replace the point cloud traversal calculation with reference to the driving characteristics of drivers on two-lane roads. The calculation process mainly includes three steps: constructing a point cloud kd tree, defining a core sight function, and ASD calculation. The first step is to establish a point cloud kd tree structure and search for nearby points. Input the original point cloud and perform a distance downsampling operation. Then, a kd tree structure is constructed in the downsampled point cloud (denoted as M = {P(x, y, z)}). The second step is to define three core sight functions in order to quickly detect whether there is a sight obstacle. Considering the characteristics of two-lane roads, three sight lines that can cover the entire road are set. The third step is to search for obstacles around the sight and calculate ASD. The logic of detecting obstacles around the sight is divided into finding the nearest point and determining whether the sight is blocked. When looking for the nearest point, take detection points L at equal intervals on each sight line. o ,When there is a point cloud within a certain range of the detection point, the line of sight is determined to be blocked, and the distance from the nearest point to the observation point is calculated as ASD. If the line of sight is unobstructed, the ASD is calculated as the distance from the observation point to the

[0014] Step 2: Build a point cloud-based two-lane highway glare detection method to detect the glare location on the road

[0015] The road glare detection method is to determine whether glare occurs based on the relative position of the driver and the sun, and determine the glare area and time. The glare detection process of the present invention: firstly, the sun position at different times is calculated to simulate the sun's rays, then each observation point is determined to determine whether the light enters the visual recognition angle, and whether the sun's rays intersect with the roadside ground point cloud, and the sun's rays that do not cause glare are filtered out to simulate the specific glare time of each observation point in clear weather. The visual recognition angle refers to the critical angle value between the interfering glare and the acceptable glare, that is, the maximum relative angle range between the sun and the driver's position when the driver feels glare.

[0016] First, in order to analyze the solar glare in the study area, the solar azimuth and solar altitude angles are used to simulate the sun's position and generate sunlight. There are two main steps: the first step is to generate the solar declination and solar hour angle based on the annual accumulated day and clock time, and then generate the solar altitude angle based on this; the second step is to generate the solar azimuth angle with the solar declination angle and the longitude and latitude of the study area as input. The final output is the solar azimuth and solar altitude angle with timestamps to represent the sunlight.

[0017] Then, after the sunlight is generated, the glare detection is performed on each observation point. Only the sunlight that is incident on the driver's visual recognition angle and is not blocked by roadside objects will cause glare. The observation point glare detection process is divided into two steps: first, the angle difference between the sunlight and the driver's central line of sight is calculated at each point on the road, and the sunlight within the visual recognition angle is retained; second, the intersection of the sunlight and the point cloud is calculated, and the sunlight that intersects with the point cloud is deleted. In order to reduce the amount of calculation, before performing occlusion judgment, the observation point (O i ) Points within a certain range are sent into calculation. After two steps, the observation points where light still exists are the glare observation points. The glare time, solar altitude angle, and solar azimuth angle of each point are output.

[0018] In the 12 months of the year, one day is selected each month to analyze road glare. In order to coordinate with the dates of the spring equinox, summer solstice, autumnal equinox and winter solstice, the analysis date is selected on the 22nd or 21st of each month. Based on the above judgment process, this section outputs the glare area observation point, the heading angle and pitch angle of the driver vector corresponding to the observation point, and also outputs the glare time, solar altitude angle and solar azimuth.

[0019] Step 3: Establish a viewing distance attenuation rate model and recalculate the available viewing distance at the disabling glare position

[0020] The model of the viewing distance attenuation rate of coupled glare factor

[0021] In order to explore the available visual distance of drivers under solar glare, the following design was made based on the available visual distance of drivers to road signs in glare and non-glare scenes recorded in the sign recognition experiment: the visual distance attenuation rate (η) was constructed, and the visual distance attenuation rate model under disabling glare conditions was fitted using five regression methods. Then, the disabling glare data was selected from the glare detection results, and the ASD was recalculated based on the visual distance attenuation rate at the location where the disabling glare occurred, based on the ASD obtained in the visual distance test.

[0022] First, the line of sight attenuation rate model was fitted based on the experimental data. In the first step, based on the available line of sight recorded during the experiment, the line of sight attenuation rate η that describes the degree of glare impact was constructed. In the second step, in order to restore the relative positions of the driver's viewpoint, signboards, and the sun during driving, the recorded solar altitude angle and solar azimuth angle were corrected. The original angle values ​​were measured by the solar altitude angle meter and the solar azimuth angle meter in the sign recognition experiment, respectively. In order to avoid errors caused by different dimensions and data ranges, the azimuth angle β was corrected before training. S , altitude angle α S , and the illumination value I is normalized and mapped to the interval [0, 1]. Finally, the azimuth angle β S , altitude angle α SWith light intensity I as input and viewing distance attenuation rate η as output, five regression methods including linear regression, Bayesian, support vector machine, decision tree, and linear perceptron are used to fit the model relationship between input and output. 2 The scores and mean square error (MSE) were used to select the best fitting random forest method and obtain the line-of-sight attenuation rate model f.

[0023] Fitting and analysis of available viewing distance at disabling glare position

[0024] In order to calculate the available sight distance value of the driver at the disabling glare position, according to the observation point of the glare road section output by the glare detection step, the observation point corresponds to the heading angle and pitch angle of the driver vector, the solar altitude angle, and the solar azimuth angle. The calculation process is as follows: the first step is to select the disabling glare observation point; the second step is to correct the solar azimuth angle and solar altitude angle, and then combine the sunlight illumination value I to input into the attenuation rate model, and output the sight distance attenuation rate at different times; the third step is based on the sight distance attenuation rate, at each observation point O i Recalculate the available sight distance at each moment, compare it with the standard value of the parking sight distance, and detect the observation point with insufficient sight distance O i , conduct security analysis.

[0025] Based on the available viewing distance and glare detection results, the viewing distance attenuation rate model f is used to couple the available viewing distance under disability glare conditions and output the available viewing distance of the disability glare observation point at each glare moment.

[0026] Compared with the prior art, the present invention has the following obvious advantages and beneficial effects:

[0027] First, the present invention constructs a sight distance inspection model suitable for two-lane highways, and uses point cloud and trajectory point data to calculate the three-dimensional available sight distance of the road. In addition to considering the visibility at the center of the lane in front of the vehicle, the visibility of the boundary lines on both sides of the road is detected, which is more suitable for two-lane highway scenes with pedestrians or non-motor vehicles crossing the road. Secondly, the present invention constructs a glare detection model divided into light simulation and occlusion discrimination, and realizes glare detection at different positions and times on two-lane highways. Since a program is designed to divide the point cloud into blocks according to the neighborhood relationship, there is no need to perform glare detection after dividing the point cloud, which simplifies the calculation steps. Finally, the present invention proposes a sight distance attenuation rate model to solve the problem of sight distance quantification under disabling glare conditions. It enables road management departments to accurately calculate the position of insufficient sight distance and glare time, facilitate the formulation of improvement measures, and improve road safety. The overall flow chart of the present invention is attached. Figure 1 . BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Technical flow chart of the present invention. DETAILED DESCRIPTION

[0029] Computer configuration and environment of the present invention.

[0030] All algorithms mentioned in the experiment were developed in Python language in Win10 system, using PyCharm development platform, Python version 3.7.16, and the data processing computer was configured with Intel core i7-12700 and 64GB memory.

[0031] Sign recognition experiment

[0032] The present invention designs a subjective visual recognition method to directly measure the driver's subjective feelings and replace the identification object with road signs. In the actual road, a part of the east-west road section is selected as the research scene. Four consecutive groups of turning signs (eight signs in both directions) are selected in this road section to study the driver's available viewing distance for traffic signs under disabling glare. The experimental data is used for subsequent model fitting.

[0033] The experimental process is as follows. In the first stage, in clear weather, the driver drove under glare-free conditions at a speed of about 40 km / h. When the driver could see the sign clearly (the sign had no reflection), the experimenter recorded the position of the vehicle and the sign. The distance between adjacent street lamp poles in the experimental section was 30 meters. Based on this, the distance between the sign and the driver was estimated, which was called the cognitive sight distance. In the second stage, the driver returned along the original route. Experimenter 1 recorded the position and time when the driver saw the sign clearly to estimate the driver's cognitive sight distance. Experimenter 2 used an illuminance meter to record the sunlight illuminance at the driver's eye position. Experimenter 3 recorded the solar altitude angle and azimuth angle according to the time. The solar altitude angle and azimuth angle were measured using a solar altitude meter and a solar azimuth meter, respectively. Since the error in evaluating the driver's available sight distance for moving objects is large, the cognitive sight distance is used as the available sight distance. The available sight distance, sunlight illuminance, solar altitude angle and azimuth angle recorded in this experiment are used for subsequent fitting of the sight distance attenuation rate model.

[0034] Sight distance test

[0035] Build observer and driver vectors from trajectory points

[0036] Construct an ASD model suitable for two-lane roads and calculate the road ASD. It includes two steps: first, establish a driver observation point set and a corresponding driver vector set, then set the observation point / target point position, establish the line of sight and calculate the ASD according to the characteristics of two-lane roads with pedestrians and non-motor vehicles crossing the road.

[0037] Before calculating the available sight distance, establish the driver observation point set and the corresponding driver vector set for the two-lane highway. Considering the driver's eye height during vehicle driving, set the elevation value of all observation points to a certain value above the ground. Regarding the plane coordinates of the observation points, calculate the distances between adjacent trajectory points in turn, and accumulate the distances between adjacent trajectory points. When the accumulated value is greater than a certain threshold (km), use the plane coordinates of the trajectory point at this time as the observation point (O i , where i = 1, 2, ..., w, w is the total number of observation points in the data set). Construct the driver vector from the current observation point to the next observation point, and use the next observation point coordinate (O i+1 (x,y,z) minus the current observation point coordinates (O i (x,y,z)) Get the driver vector of the current position Since the experimental vehicle type is a small car, the elevation of the observation point is set to 1.2m. In order to facilitate the setting of observation points and consider the driving sight distance requirements, the distance accumulation threshold k of adjacent trajectory points is set to 10m.

[0038] The commonly used ASD detection model generally traverses all point clouds in front according to the order of observation points / target points. Due to the large amount of point cloud data, the efficiency is low. In order to improve the calculation efficiency, three core sight lines are designed to replace the point cloud traversal calculation based on the driving characteristics of drivers on two-lane roads. The calculation process mainly includes three steps: constructing the point cloud kd tree, defining the core sight function, and ASD calculation. Before detection, the distance threshold of the point cloud downsampling operation is 0.2m.

[0039] Establishing sight lines and calculating ASD

[0040] The first step is to establish a point cloud kd tree structure and search for nearby points. Input the original point cloud and perform distance downsampling. Then, construct a kd tree structure in the downsampled point cloud (denoted as M = {P(x, y, z)}). When constructing a k-dtree in a point cloud, the data set is divided into subtrees level by level, with two subtrees at each level. Each level of division is performed cyclically in different coordinate dimensions. This operation establishes the positional connection between adjacent points, which facilitates the rapid search of neighborhood points. When searching for the nearest point of a query point in a kd tree, it includes two steps of recursive descent and backtracking. The pruning operation in the recursive descent can reduce the amount of calculation and thus increase the speed, while backtracking ensures the integrity of the search for adjacent points. When searching, first input a query point. Before recursive descent, initialize the nearest point to none, set the initial minimum distance to infinity, and continuously approach the nearest point of the query point during the recursive descent and backtracking process. When the recursive descent and backtracking steps are completed, the nearest point can be searched.

[0041] The second step is to define three core sight lines functions in order to quickly detect whether there is a sight line obstacle. Set three sight lines that can cover the entire road surface. For the i-th observation point O i (x, y, z), along the direction of the observation point, define the target point n meters in front of it, and establish the core line of sight L from the observation point to the target point Mi With the core sight line as the center and referring to the double-lane boundary line, the target point corresponding to the observation point is offset left and right respectively along the direction perpendicular to the road, forming another two core sight lines (L Ri and L Li ). Finally, three core sight lines are formed at each observation point, covering the opposite lane, the center of the own lane, and the right boundary line of the road.

[0042] Compared with the voxel ray method that only detects a single ray, the present invention sets two additional lines of sight at each observation point. Due to the existence of pedestrians crossing the road and oncoming vehicles using the same lane on two-lane roads, the additional two lines of sight are used to detect the visibility of the boundary lines on both sides of the road. Due to the phenomenon of using the same lane on two-lane roads, it is necessary to detect the passing sight distance and the parking sight distance. Taking into account the random changes in the driver's driving speed during actual driving, the distance (n) between the observation point and the corresponding target point is taken as 80m and 150m (refer to the "Highway Route Design Specification" (JTG D20-2017), hereinafter referred to as the "Specification"). When inspecting ASD at a distance of 80m, because the lane width is 3.5 meters, the three core lines of sight (L Mi , L Ri , L Li )'s landing point (target point T i ) are three positions: the eighth observation point O in front of the observation point i+8 , perpendicular to the road direction i+8 Shift 4.0m to the left and change O i+8 When checking the sight distance at a distance of 150m, the core sight line (L Mi ) falls at the fifteenth observation point O in front of the observation point i+15 , the left and right offset distances of the other two core sight points remain unchanged.

[0043] The third step is to search for obstacles around the line of sight and calculate ASD. The logic of detecting obstacles around the line of sight is divided into finding the nearest point and determining whether the line of sight is blocked. When finding the nearest point, take points at equal intervals on each line of sight (L o ), forming a point set For each L o , find L from the point cloud (M = {P (x, y, z)}) through kd tree o The closest point P c , the nearest point of each line of sight forms a set Calculate from the sight point (Lo ) to its closest point (P c )’s Euclidean distance d LP . L for each line of sight o The distance to the nearest point constitutes a set When determining whether the line of sight is obstructed, i Start traversing the closest point set for each line of sight and the corresponding distance set When the distance d LP When it is less than the average point spacing of the roadside obstacle point cloud, it is determined that the line of sight is blocked, and the corresponding P c is considered as an obstacle point. If there are multiple obstacle points in the line of sight, only the first P is taken. c When the line of sight is obstructed, the distance from the driver's point to the nearest point P c The distance from the driver point O is taken as ASD. Otherwise, i To the target point T i The distance is taken as ASD. After downsampling the point cloud (M = {P(x, y, z)}) with a distance threshold of 0.2m, the average distance from each point to its three nearest points in the roadside feature point cloud is about 0.25m, so 0.25 is used as the threshold for judging line of sight obstruction.

[0044] According to the above calculation process, the original road point cloud and the observation point set are used as input, and the output of each observation point O in the road can be i The available sight distance ASD corresponding to the three core sight lines can be compared with the parking sight distance to detect the observation points where the ASD does not meet the requirements.

[0045] Glare detection

[0046] Generate Sun Rays

[0047] The road glare detection method is to determine whether glare occurs based on the relative position of the driver and the sun, and determine the glare area and time. Glare detection process: First, the sun's position at different times is calculated to simulate the sun's rays, then each observation point is determined to determine whether the light enters the visual recognition angle, and check whether the sun's rays intersect with the roadside ground point cloud, filter out the sun's rays that do not cause glare, and simulate the specific glare time of each observation point in clear weather. The visual recognition angle refers to the critical angle value between interference glare and acceptable glare, that is, the maximum relative angle range between the sun and the driver's position when the driver feels glare.

[0048] In order to analyze the solar glare in the study area, the solar azimuth and solar altitude angles are used to simulate the sun's position and generate sunlight. There are two main steps: the first step is to generate the solar declination and solar hour angle based on the annual accumulated day and clock time, and then generate the solar altitude angle based on this; the second step is to generate the solar azimuth angle with the solar declination angle and the longitude and latitude of the study area as input. The final output is the solar azimuth and solar altitude angle with a timestamp to represent the sunlight.

[0049] Define N as the cumulative number of days, which is the number of days starting from January 1st of each year. For example, N on January 18th is 18. Fractional years (γ) can be calculated from the cumulative number of days:

[0050]

[0051] After generating the fractional year γ, the solar declination angle (δ) is calculated using Formula 5:

[0052] δ=0.006918–0.399912cos(γ)+0.070257sin(γ)–0.006758cos(2γ)+0.000907sin(2γ)–0.002697cos(3γ)+0.00148sin(3γ) (2)

[0053] The solar declination angle is equal to the latitude where the sun shines directly on the earth. Because there is a certain error between clock time and true solar time, the solar hour angle can be calculated from true solar time. True solar time t s It can be calculated according to formula 3:

[0054] t s =h our *60+m in +sc / 60+t offset (3)

[0055] Among them, h our is the clock time in hours, m in and sc refer to the minute and second readings of the clock time respectively, t offset is the error between true solar time and clock time. Formula 4 gives t offset The calculation method of .

[0056] t offset =e qtime +4*lon–60*t zone (4)

[0057] Where lon is the longitude of the study area, t zone is the time zone number of the study area, e qtime is the correction factor. e qtime Calculated by formula 5:

[0058] eqtime =229.18*(0.000075+0.001868cos(γ)–0.032077sin(γ)–0.014615cos(2γ)–0.040849sin(2γ)) (5)

[0059] Wherein, ω is the hour angle of true solar time (referred to as solar hour angle), which can be calculated according to Formula 6.

[0060] ω=(t s / 4)–180 (6)

[0061] After obtaining the solar declination angle (δ) and solar hour angle (ω), calculate the angle between the sun's rays and the ground normal vector, that is, the zenith angle (φ):

[0062] φ=acos(sin(lat)sin(δ)+costan(lat)cos(δ)cos(ω))(7)

[0063] Where lat is the latitude of the study area, and the solar altitude angle (α) is the angle between the sun's rays and the horizontal plane, which can be calculated using Equation 8:

[0064] α=90-φ(8)

[0065] At the same time, the solar azimuth (β) refers to the angle from the north clockwise to the projection of sunlight on the ground, which can be calculated according to formula 9:

[0066]

[0067] Formulas 1 to 2 explain how to calculate the solar declination angle by the annual accumulation of days, formulas 3 to 6 explain how to generate the solar hour angle by the clock hours, formulas 7 to 8 generate the solar altitude angle (α) based on the solar declination angle and the solar hour angle, and formula 9 generates the solar azimuth angle (β). To generate sunlight, the date input is the annual accumulation of days, the clock input is a number between 5 and 20 with one decimal place, and the longitude and latitude coordinates of the center of the study area are selected for input. The sunrise and sunset times vary slightly due to seasonal changes. Selecting a clock time of 5 to 20 can cover all daytime hours. Since the position of the sun changes very little within 6 minutes, light data is generated every 6 minutes, and the detection accuracy meets actual needs. Since the experimental roads are located close to each other, the longitude and latitude of the center of the study area are selected for input into the program.

[0068] Glare detection at each observation point

[0069] After the sunlight is generated, glare detection is performed on each observation point. Only sunlight that enters the driver's visual recognition angle and is not blocked by roadside objects will cause glare. The observation point glare detection process is divided into two steps: first, the angle difference between the sunlight and the driver's central line of sight is calculated at each point on the road, and the sunlight within the visual recognition angle is retained; second, the intersection of the sunlight and the point cloud is calculated, and the sunlight that intersects with the point cloud is deleted. The observation point where there is still light after the two steps is the glare observation point, and the glare time, solar altitude angle, and solar azimuth angle of each point are output.

[0070] First, at each observation point (O i ), compare the driver vector angle, the sun angle, and the visual recognition angle range to determine whether the sunlight is incident on the driver's visual recognition angle. Define the north direction clockwise to the driver vector The angle between Define the driver vector The angle with the horizontal plane is the pitch angle Heading angle of the driver vector and pitch angle Calculated by formula 10:

[0071]

[0072] in is the driver vector The three-dimensional coordinate value of .

[0073] Input the sunlight simulated in Section 4.2.1, calculate the angle difference between the driver vector and the sunlight, and determine whether the sunlight is incident on the driver's visual recognition angle, as shown in discriminant formula 11:

[0074]

[0075] If the azimuth angle (β) of the sun's rays is equal to the heading angle of the driver's vector The difference is within ±β1°, and the altitude angle (α) of the sun’s rays is equal to the pitch angle of the driver’s vector. The difference is greater than 0 and less than α1, the sunlight is judged to be incident within the driver's glare recognition angle, at the corresponding observation point (O i )Record the time, azimuth and altitude data of sunlight.

[0076] According to the generated sunlight data, glare identification angle incidence detection and light shielding judgment are performed point by point at the observation points of road1, road2, and road3 to detect the time and observation point of solar glare. In the glare identification angle incidence detection, the angle threshold β1 is 57 and the angle threshold α1 is 29.

[0077] The second step is to detect whether the sunlight incident on the glare recognition angle is blocked by roadside objects and delete the rays intersecting with the point cloud of the objects. In previous studies on occlusion discrimination, the spatial relationship between a single sunlight ray passing through the driver's position and the object was determined to determine whether the light was blocked. Therefore, this method determines whether the light is blocked by the distance from the point cloud to the light. i ) as an example, in order to reduce the amount of calculation, before performing occlusion judgment, the observation point (O) is searched from the point cloud (M = {P (x, y, z)}) according to k-dtree. i ) Radius r s Points within the range of m, r s The value indicates that the occlusion judgment is performed within a certain range. Since the sight line below the driver's eye level will not block the sun's rays, the elevation filter is used to filter out the elevation below the observation point (z Oi ) of the point cloud, and only the remaining point cloud (M Or ) is sent to the calculation.

[0078] Formula 12 defines the point cloud (M Or ) to the distance (d SP ), if d SP If it is smaller than the average point spacing of the roadside feature point cloud, the corresponding sunlight is judged to intersect with the point cloud.

[0079] d SP =|PS t |(12)

[0080] Among them, S t is the foot of the perpendicular of point P on the sun's rays and can be calculated using formula 13.

[0081]

[0082] The t in the above formula is a position parameter and can be calculated by formula 14:

[0083]

[0084] where the vector From the observation point (O i ) points to the point cloud P, calculated by subtracting the coordinates of the two points. is the direction vector of the sun's rays, from the observation point (O i ) points to point S. By subtracting the coordinates of the two points, the position of point S can be calculated using formula 15.

[0085] S=O i +(l*sin(β),l*cos(β),l*sin(α))(15)

[0086] Where l is the distance parameter set for calculating point S; α is the solar altitude angle, which can be obtained by a solar altitude angle meter; β is the solar azimuth angle, which can be obtained by a solar azimuth angle meter.

[0087] In the light occlusion judgment step, the neighborhood radius is r s =100, representing the search observation point (O i ) The point cloud within 100m before and after is used for occlusion judgment, and the neighborhood radius is r s =100, a good balance can be achieved between calculation speed and accuracy. The distance parameter l is 100, which means that point S is taken on the sun's rays at a distance of 100m from the observation point. The average distance from each point to its three nearest points is about 0.25m. Or ) within 0.25m, the ray is deleted.

[0088] If the sunlight intersects with the point cloud of the roadside objects, it means that the roadside objects block the sunlight. i ) delete this ray; if they do not intersect, the sunlight does enter the driver's visual recognition angle, causing glare. The above calculation process is iterated at each observation point, and all observation points where there is still light are determined to be glare areas, and the time when the sunlight is retained is determined to be glare time.

[0089] In the 12 months of the year, one day is selected each month to analyze road glare. In order to coordinate with the dates of the spring equinox, summer solstice, autumnal equinox and winter solstice, the analysis date is selected on the 22nd or 21st of each month. Based on the above judgment process, this section outputs the glare area observation point, the heading angle and pitch angle of the driver vector corresponding to the observation point, and also outputs the glare time, solar altitude angle and solar azimuth.

[0090] Calculation of viewing distance under disabling glare

[0091] The model of the line-of-sight attenuation rate of coupled glare factor

[0092] In order to explore the available viewing distance of drivers under solar glare, according to the available viewing distance of drivers to road signs in glare and non-glare scenes recorded in the sign recognition experiment, the following design was made: the viewing distance attenuation rate (η) was constructed, and the viewing distance attenuation rate model under disabling glare conditions was fitted using five regression methods.

[0093] Firstly, according to the available viewing distance recorded during the experiment, the viewing distance attenuation rate η describing the degree of glare impact was constructed, and the calculation formula is as follows.

[0094]

[0095] Among them, ASD (Available Sight Distance) and ASD gAvailable Sight Distance Under Glare refers to the available sight distance of traffic signs for drivers in non-glare and glare conditions, respectively, and is obtained from sign recognition experiments.

[0096] In the second step, in order to restore the relative positions of the driver's viewpoint, the sign and the sun during driving, the recorded solar altitude angle and solar azimuth angle are corrected. The original angle values ​​are measured by the solar altitude angle measuring device and the solar azimuth measuring device in the sign recognition experiment. Usually, it is assumed that the height of the traffic sign from the ground is h s When calculating the altitude angle between the driver's line of sight and the sun, the elevation angle when the driver observes the traffic sign in front is added. The altitude angle between the driver's line of sight and the sun position (α S ) is calculated as follows:

[0097]

[0098] The azimuth angle between the driver's line of sight and the sun's position (β S ) is calculated as in formula 18

[0099]

[0100] In the formula is the heading angle of the driver vector (calculated based on the driver vector at the location of the driver during the experiment), which is the angle from the north direction to the direction of the car's forward movement in a clockwise direction. To avoid errors caused by different dimensions and data ranges, the azimuth angle β is set before training. S , altitude angle α S , and the illumination value I is normalized and mapped to the interval [0, 1].

[0101] Usually the point cloud elevation difference between the sign and the ground is 6.3m. When correcting the solar altitude angle, set h s The value is 6.3. When correcting the solar azimuth, It is calculated based on the direction of the road centerline of the road section to which the sign belongs. Specifically, the coordinates of two points in the road centerline marking point cloud are taken and calculated according to Formula 13. After preprocessing, a total of 26 valid data of drivers are obtained, with 4 data for each driver, totaling 104 data, of which 84 data are used as training set fitting and 20 data are used as test set.

[0102] Finally, with the azimuth angle β S , altitude angle α S With light intensity I as input and viewing distance attenuation rate η as output, five regression methods including linear regression, Bayesian, support vector machine, decision tree, and linear perceptron are used to fit the model relationship between input and output. The random forest model R 2The score is the largest and the mean square error (MSE) is the smallest. The random forest model is selected to fit the line-of-sight attenuation rate model (η) (as shown in Formula 19).

[0103] η=f(β S ,α S ,I;θ)(19)

[0104] Where f represents the regression model used, and θ represents the number of trees in the model, which is set to 40. To ensure the repeatability of the experiment, the random seed is set to 42.

[0105] Fitting and analysis of available viewing distance at disabling glare position

[0106] In order to calculate the available sight distance value of the driver in the glare area, the observation points of the glare road section outputted in the glare detection step are used, and the observation points correspond to the heading angle and pitch angle of the driver vector, the solar altitude angle, and the solar azimuth angle. The calculation process is as follows: the first step is to select the disabling glare observation points; the second step is to correct the solar azimuth angle and solar altitude angle, and then combine the solar illuminance value I to input into the attenuation rate model, and output the sight distance attenuation rate at different times; the third step is based on the sight distance attenuation rate, at each observation point O i Re-fit the available sight distance at each moment, compare it with the parking sight distance value in the "Specification", and detect the observation point O with insufficient sight distance i , conduct security analysis.

[0107] In the first step, the visual recognition angle range set during glare detection represents the critical value between acceptable glare and interference glare. Most of the experimental data obtained in the sign recognition experiment are data under disabling glare, in which the solar azimuth and solar altitude angles are often smaller than the visual recognition angle. Therefore, the angle range of disabling glare in the experimental data is queried, and the observation points for sight distance fitting are selected based on this range. The solar azimuth and solar altitude angle ranges recorded in Experiment 3.3 are horizontal [-42, 42] and vertical [5, 17]. Based on this range, the disabling glare observation points are retained in the data output from the glare detection step.

[0108] In the second step, the following operations are performed at each observation point: based on the line-of-sight attenuation rate model f (Formula 19), the solar altitude angle, solar azimuth angle and solar illuminance value at each moment are input, and the line-of-sight attenuation rate (η) is calculated at each observation point. Since there is no accurate reference formula to calculate the solar illuminance value I, a certain value is taken to represent the solar illuminance value on a sunny day.

[0109] The third step is to obtain the viewing distance attenuation rate of each observation point, and then calculate the available viewing distance (ASD) according to formula 20 at each glare moment. g ):

[0110] ASD g =ASD*(1-η)(20)

[0111] In the formula, ASD is the ASD value calculated in the sight distance test (step 1), assuming that the driver has normal vision and can identify traffic signs 150m away. Then the available sight distance (ASD) of each observation point is calculated. g ), ASD g Although it is actually the available visual distance of the driver to the sign under glare, the visual distance error of the driver to identify other objects under disabling glare is large, and the visual distance to the road sign can reflect the degree of visual distance attenuation caused by disabling glare. Therefore, based on the available ASD g The results were compared with the driving sight distance standards to carry out safety analysis.

[0112] Finally, the present invention will output the original available sight distance of the non-glare observation point on the road and the available sight distance of the disabled glare observation point. This information can be used for road warning and road maintenance.

Claims

1. A method for testing the available viewing distance considering the influence of disabling solar glare, characterized in that: The following steps are involved: Constructing an ASD model and calculating the available sight distance (ASD) of the road includes two steps: first, establishing a driver observation point set and a corresponding driver vector set, and then setting the observation point / target point position according to the characteristics of pedestrians and non-motor vehicles crossing the road on a two-lane road, establishing the sight line and calculating the ASD; Before calculating the available sight distance, establish the driver observation point set and the corresponding driver vector set on the two-lane highway; consider the driver's eye height during vehicle driving, set the elevation value of all observation points to a certain value above the ground, and take the fixed value as 1.2 meters; with respect to the plane coordinates of the observation points, calculate the distances between adjacent trajectory points in turn, and accumulate the distances between adjacent trajectory points. When the accumulated value is greater than a certain threshold, the threshold k is 10 meters, and the plane coordinates of the trajectory point at this time are used as the observation point O i Plane coordinates; where i = 1, 2, ..., w, w is the total number of observation points in the set of data; construct the driver vector from the current observation point to the next observation point, and the next observation point coordinate (O i+1 (x,y,z) minus the current observation point coordinates (O i (x,y,z)) Get the driver vector of the current position Three core sight lines are designed to replace the point cloud traversal calculation; the calculation process includes three steps: building a point cloud kd tree, defining a core sight function, and ASD calculation; the first step is to establish a point cloud kd tree structure and search for nearby points; input the original point cloud and perform distance downsampling operation; then, build a kd tree structure in the downsampled point cloud (denoted as M = {P(x, y, z)}); the second step is to define three core sight functions to detect whether there is a sight obstacle; considering the characteristics of a two-lane highway, set three sight lines that can cover the entire road surface; the third step is to search for obstacles around the sight line and calculate ASD; the logic of detecting obstacles around the sight line is divided into finding the nearest point and determining whether the sight line is blocked; when finding the nearest point, take detection points L at equal intervals on each sight line o ,When there is a point cloud within 0.25 meters of the detection point, the line of sight is determined to be blocked, and the distance from the nearest point to the observation point is calculated as ASD. If the line of sight is unobstructed, ASD is calculated as the distance from the observation point to the target point; Step 2: Build a point cloud-based two-lane highway glare detection method to detect the glare location on the road Glare detection process: First, the sun position at different times is calculated to simulate sunlight. Then, each observation point is judged to see whether the light enters the visual recognition angle, and whether the sunlight intersects with the point cloud of roadside objects. The sunlight that does not cause glare is filtered out to simulate the specific glare time of each observation point in clear weather. The visual recognition angle refers to the critical angle value between interfering glare and acceptable glare, that is, the maximum relative angle range between the sun and the driver's position when the driver feels glare. First, in order to analyze the solar glare in the study area, the solar azimuth and solar altitude angles are used to simulate the sun's position, and the sun's rays are generated accordingly. This is divided into two steps: the first step is to generate the solar declination angle and solar hour angle based on the annual accumulated day and clock time, and then generate the solar altitude angle based on this; the second step is to generate the solar azimuth angle using the solar declination angle and the longitude and latitude of the study area as input; the final output is the solar azimuth angle and solar altitude angle with a timestamp to represent the sun's rays; Subsequently, after the sunlight is generated, glare detection is performed on each observation point. Only sunlight that is incident on the driver's visual recognition angle and is not blocked by roadside objects will cause glare. The observation point glare detection process is divided into two steps: first, the angle difference between the sunlight and the driver's central line of sight is calculated at each point on the road, and the sunlight within the visual recognition angle is retained; the second step is to calculate the intersection of the sunlight and the point cloud, and delete the sunlight that intersects with the point cloud. In order to reduce the amount of calculation, before performing occlusion judgment, according to k-dtre e Search for observation point O from the global point cloud i Points within a 100-meter range are sent for calculation. After two steps, the observation points where light still exists are glare observation points. The glare time, solar altitude angle, and solar azimuth angle of each point are output. In the 12 months of the year, one day is selected each month to analyze road glare; in order to coordinate with the dates of the spring equinox, summer solstice, autumnal equinox and winter solstice, the analysis date is selected on the 22nd or 21st of each month; according to the above judgment process, this section outputs the glare area observation point, the heading angle and pitch angle of the driver vector corresponding to the observation point, and also outputs the glare time, solar altitude angle and solar azimuth; Step 3: Establish a viewing distance attenuation rate model and recalculate the available viewing distance at the disabling glare position The model of the viewing distance attenuation rate of coupled glare factor In order to explore the available visual distance of drivers under solar glare, the following design was made based on the available visual distance of drivers to road signs in glare and non-glare scenes recorded in the sign recognition experiment: the visual distance attenuation rate (η) was constructed, and the visual distance attenuation rate model under disabling glare conditions was fitted using the regression method; then, the disabling glare data was selected from the glare detection results, and the ASD was recalculated based on the visual distance attenuation rate at the location where the disabling glare occurred based on the ASD obtained in the visual distance test; First, the viewing distance attenuation rate model was fitted according to the experimental data. In the first step, the viewing distance attenuation rate η, which describes the degree of glare influence, was constructed according to the recorded available viewing distance. In the second step, the recorded solar altitude angle and solar azimuth angle were corrected in order to restore the relative positions of the driver's viewpoint, the signboard and the sun during driving. The original angle values ​​were measured by the solar altitude angle meter and the solar azimuth angle meter in the sign recognition experiment respectively. Before training, the azimuth angle β S , altitude angle α S , and the illumination value I is normalized and mapped to the interval [0, 1]; finally, the azimuth angle β S , altitude angle α S With light intensity I as input and viewing distance attenuation rate η as output, five regression methods including linear regression, Bayesian, support vector machine, decision tree, and linear perceptron are used to fit the model relationship between input and output. 2 The score and mean square error MSE were used to select the best fitting random forest method and obtain the line-of-sight attenuation rate model f; In order to calculate the available sight distance value of the driver at the disabling glare position, according to the observation point of the glare road section output by the glare detection step, the observation point corresponds to the heading angle and pitch angle of the driver vector, the solar altitude angle, and the solar azimuth angle; the calculation process is as follows: the first step is to select the disabling glare observation point; the second step is to correct the solar azimuth angle and solar altitude angle, and then combine the solar illuminance value, input it into the attenuation rate model, and output the sight distance at different times at the same time. The third step is to calculate the attenuation rate at each observation point O based on the line of sight. i Recalculate the available sight distance at each moment, compare it with the standard value of the parking sight distance, and detect the observation point with insufficient sight distance O i , Based on the available viewing distance and glare detection results, the viewing distance attenuation rate model f is used to couple the available viewing distance under disability glare conditions and output the available viewing distance of the disability glare observation point at each glare moment.

Images