Comprehensive evaluation and optimization method of radar and visual integrated machine erection scheme based on variance maximization combination weighting

Abstract

This invention relates to a comprehensive evaluation and optimization method for radar-visual integrated machine installation schemes based on variance maximization combined weighting, belonging to the field of traffic engineering technology. It includes: analyzing the main factors affecting the installation scheme and designing different working conditions as single variables; selecting index parameters and constructing an evaluation system; using the analytic hierarchy process (AHP) as the subjective weighting method, and the information weighting method, entropy weighting method, and CRITIC method as objective weighting methods, and combining them based on variance maximization to comprehensively evaluate and optimize various installation schemes for radar-visual integrated machines. This invention proposes a comprehensive evaluation method based on variance maximization combined subjective and objective weighting, eliminating the empirical defects of the subjective weighting method and absorbing the objective condition advantages of the objective weighting method. It calculates the combined weights based on variance maximization, and then calculates the comprehensive score of the installation scheme, improving the accuracy of radar-visual integrated machine installation scheme evaluation and optimization.

Description

Technical Field

[0001] This invention relates to a comprehensive evaluation and optimization method for the installation scheme of integrated radar and vision systems based on variance maximization combined weighting, belonging to the field of traffic engineering technology. Background Technology

[0002] Roadside intelligent traffic sensing devices can acquire a wealth of road traffic information, such as vehicle identification and capture, vehicle trajectory tracking, traffic violation monitoring and early warning, and the construction of intelligent transportation systems to achieve intelligent road traffic management. Common roadside sensing sensors include cameras, millimeter-wave radar, magnetic sensors, infrared sensors, and lidar. By installing and setting up these sensors and processing the collected data, road vehicles can be perceived and identified, and relevant traffic information such as location and speed can be obtained.

[0003] When setting up a single sensor, only its own parameter characteristics need to be considered, making installation relatively easy. However, for multi-sensor fusion devices, more factors need to be considered, making installation more complex. Currently, research on multi-sensor fusion mainly focuses on the fusion of radar and camera sensors. Visual perception based on high-definition cameras can intuitively acquire vehicle image data and process the images to achieve target detection, but its spatial positioning capability is poor, unable to determine the vehicle's position in the real world, and is easily affected by environmental factors such as lighting. Road perception based on millimeter-wave radar has high accuracy in acquiring distance and speed data and strong anti-interference capabilities, but it suffers from poor target classification and insufficient recognition capabilities.

[0004] Scholars have developed radar-video integrated units based on millimeter-wave radar and cameras. Road traffic scenarios are highly variable and subject to environmental constraints; different installation methods can significantly impact the monitoring and identification capabilities of these radar-video integrated units. Furthermore, radar-video integrated unit sensing devices have only been around for a short time, and current research on the optimal installation schemes for these devices is limited. Therefore, research on comprehensive evaluation and optimization methods for radar-video integrated unit installation schemes is extremely necessary and urgent. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a comprehensive evaluation and optimization method for the installation scheme of the integrated radar-visual machine based on variance maximization combined weighting. By analyzing and collecting data, a comprehensive evaluation index system is established; by using a combination of subjective and objective weighting methods, the defects of subjective and objective single weighting methods are overcome; based on variance maximization, the combined weights are calculated, and the evaluation score is obtained, thereby optimizing the installation scheme of the integrated radar-visual machine.

[0006] The technical solution of the present invention is as follows:

[0007] A comprehensive evaluation and optimization method for the installation scheme of integrated radar-visual equipment based on variance maximization combined weighting includes:

[0008] Step 1: Analyze the main factors affecting the erection scheme and design different working conditions as single variables to conduct a control experiment;

[0009] Step 2: Analyze the experimental data collected by the Ravec all-in-one machine, select indicator parameters, and construct an evaluation system;

[0010] Step 3: The Analytic Hierarchy Process (AHP) is selected as the subjective weighting method, while the information weighting method, entropy weighting method, and CRITIC method are selected as objective weighting methods. Based on the maximization of variance, the weighting methods are combined to comprehensively evaluate and optimize the various installation schemes of the Raid-view all-in-one machine.

[0011] Preferably, the method for analyzing the main factors affecting the erection scheme in step 1 is as follows:

[0012] S1-1: Query the specifications of millimeter-wave radar and camera sensing devices;

[0013] S1-2: Considering the road environment: Due to installation limitations, consider the installation location and height. The installation location includes the center of the road and one side of the road.

[0014] S1-3: Considering the rotation of the equipment itself, in the road environment, the horizontal direction of the road is the X-axis, the vertical direction of the road is the Y-axis, and the vertical line is the Z-axis. The radar-view integrated machine is established with the ground projection point as the origin. The rotation angle of the radar-view integrated machine around the X, Y, and Z coordinate axes is a factor affecting the installation scheme.

[0015] S1-4: Design working conditions based on five variables, including installation location, height, and three rotation angles, and conduct a control experiment.

[0016] Preferably, step 2 specifically includes:

[0017] S2-1: Analyze the data collected by the Rayvision all-in-one machine;

[0018] S2-2: Constructing an evaluation index system

[0019] The monitoring capabilities of the integrated radar-visual machine are taken as the overall layer, while the ability to identify ordinary traffic incidents, the ability to identify special traffic incidents, and the monitoring range are taken as the primary evaluation indicators. The primary indicators are further divided into secondary evaluation indicators.

[0020] The traffic data generated by the radar-video integrated machine includes records of ordinary traffic incidents and records of special traffic incidents. 450 records of ordinary traffic incidents were selected, and the average confidence level (C1), standard deviation (C2), and frequency of no license plate occurrence (C3) of the 450 vehicles were used as the secondary indicator layer for ordinary traffic incident recognition capability. 100 records of special traffic incidents were selected, and the average confidence level (C4), standard deviation (C5), frequency of no license plate occurrence (C6), and average confidence level (C7) and standard deviation (C8) of each recognized character of the license plate were used as the secondary indicator layer for special traffic incident recognition capability. The maximum longitudinal monitoring range (C9) along the road direction was used as the secondary evaluation indicator for the radar-video integrated machine's monitoring range. C9 is defined as the y-range, which is the difference between the farthest and closest distances monitored by the radar-video integrated machine in the longitudinal direction of the road.

[0021] The evaluation index system is shown in Table 1. The index attributes represent the positive or negative correlation between each secondary index and the overall level. A positive correlation indicates that the evaluation index is positively correlated with the previous level index, and a negative correlation indicates that the evaluation index is negatively correlated with the previous level index.

[0022] Table 1: Evaluation Index System

[0023]

[0024]

[0025] The formula for calculating the average confidence level C1 for vehicle identification in ordinary traffic incidents is as follows:

[0026]

[0027] Among them, EntireBelieve1 i This represents the vehicle identification confidence level for the i-th ordinary traffic incident;

[0028] The formula for calculating the standard deviation C2 of the confidence score for vehicle identification in ordinary traffic incidents is as follows:

[0029]

[0030] The probability of a vehicle without license plates appearing in a typical traffic incident is C3:

[0031]

[0032] n′ represents the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file for ordinary traffic events; C4 represents the average confidence score of vehicle recognition in special traffic events.

[0033]

[0034] Among them, EntireBelieve2i This represents the confidence level for vehicle identification in the i-th special traffic incident;

[0035] Standard deviation of confidence level C5 for vehicle identification in special traffic incidents:

[0036]

[0037] The probability of vehicles without license plates appearing in special traffic incidents (C6):

[0038]

[0039] n″ represents the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file for special traffic events;

[0040] Average confidence level C7 for license plate character recognition in special traffic incidents:

[0041]

[0042] D j This represents the confidence level of the j-th license plate recognition character;

[0043] Standard deviation of confidence level for license plate character recognition in special traffic incidents (C8)

[0044]

[0045] The y-value range C9 is the difference between the farthest and closest distances monitored by the integrated radar-vision system in the longitudinal direction of the road. Its calculation formula is as follows:

[0046] C9 = y max -y min (9)

[0047] y max The maximum distance that the radar-vision integrated machine can monitor in the longitudinal direction of a road is y min This is the closest distance that the radar-vision integrated machine can monitor in the longitudinal direction of the road.

[0048] Preferably, step 3 includes:

[0049] S3-1: Calculate the weights of each indicator for the subjective weighting method using the analytic hierarchy process;

[0050] S3-2: Calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method based on SPSSAU;

[0051] S3-3: Weights obtained by combining the analytic hierarchy process (AHP), information weighting method, entropy weighting method, and CRITIC method with variance maximization as the objective function;

[0052] S3-4: Using the weights of each indicator obtained from S3-3, a comprehensive score is given to the erection scheme for each working condition, and the optimal erection scheme is selected.

[0053] The preferred calculation method for the analytic hierarchy process is as follows:

[0054] A. For indicators at the same level, their relative importance is quantified using a 1-9 scale. The larger the number, the higher the importance. The judgment matrix A1 is defined as follows:

[0055]

[0056] a kp This represents the ratio of the importance of factor k to that of factor p.

[0057] B. Solve for the largest eigenvalue λ based on the judgment matrix. max And the eigenvector W, and perform normalization on it;

[0058]

[0059]

[0060] Where A1 is the judgment matrix, W is the eigenvector, and λ max For the largest eigenvalue, w i It is the weight of each indicator after normalizing the feature vector W, w l This refers to each vector value in the feature vector W;

[0061] C. Based on the maximum eigenvalue of each judgment matrix, calculate the consistency index CI, query the average random consistency index table, find the RI according to the order of the judgment matrix, and calculate the consistency ratio CR:

[0062]

[0063]

[0064] n represents the order of the judgment matrix;

[0065] If CR≤0.1, which indicates consistency, the judgment matrix can be used. The weights of each indicator factor are reasonably allocated, and the vector corresponding to its largest eigenvalue is the weight vector of each indicator in that layer.

[0066] If CR≤0.1 is not satisfied, repeat steps A to C to compare the importance of each indicator again and construct a judgment matrix.

[0067] Preferably, the weights obtained from the information weight method, entropy weight method, and CRITIC method in the objective weighting method are calculated as follows:

[0068] For the collected raw data r lj Take the reciprocal of the negative indicator to make it positive, and then normalize all the positive indicators to obtain r. lj The matrix is ​​as shown in formula (15);

[0069]

[0070] Where, r lj Let be the value of the l-th erection scheme under the j-th index parameter, m represent the total number of erection schemes in the working condition, and r lj正向 A new indicator obtained by taking the reciprocals of the positive and negative indicators in the evaluation system;

[0071] The normalized matrix is ​​input into the SPSSAU analysis software to calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method.

[0072] Preferably, the weights obtained by the above four methods are combined with the objective function of maximizing variance, as follows:

[0073] ① Regarding the weight value of the j-th indicator parameter, assuming the weight range obtained by the four methods is as follows: Combined weight w j It should meet the following conditions:

[0074]

[0075] ② Four subjective and objective weighting methods were used to obtain the index weight matrix, as shown in formula (17);

[0076]

[0077] ③ Take the largest variance of the score of the setup scheme under the combined weight as the objective function, make the sum of the reasonable range of the attribute and the attribute weight equal to 1, establish an optimization model, and obtain the combined weight;

[0078] The optimized model is as follows:

[0079]

[0080]

[0081] Where maxσ(θ) represents the objective function for maximizing variance, w j ' represents a combination matrix, Indicates r lj The average value of ′ and These represent the four methods for finding the maximum and minimum weights, respectively.

[0082] ④ Based on the combined weights obtained in step ③, the comprehensive score is calculated as follows:

[0083]

[0084] The option with the highest overall score is selected as the preferred option.

[0085] For any details not covered in this invention, please refer to the prior art.

[0086] The beneficial effects of this invention are as follows:

[0087] (1) This invention can study the installation scheme of newly developed radar-visual integrated machine, analyze the influencing factors and use them as a single variable to design different working conditions. This method can be used for the installation and application of multiple types of radar-visual integrated machines.

[0088] (2) This invention constructs a three-level evaluation index system based on the data collected by the RaidVision all-in-one machine, which is highly systematic;

[0089] (3) This invention proposes a comprehensive evaluation method based on the combination of subjective and objective weights with the maximization of variance. This method eliminates the empirical defects of the subjective weighting method and draws on the objective condition advantages of the objective weighting method. It calculates the combination weights based on the maximization of variance and then calculates the comprehensive score of the installation scheme, thereby improving the accuracy of the evaluation and selection of the installation scheme of the integrated radar screen. Attached Figure Description

[0090] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0091] Figure 2 This is a schematic diagram of the radar-visual integrated machine data acquisition platform;

[0092] Figure 3 This is a schematic diagram of the spatial coordinate system of the Raveo all-in-one machine;

[0093] Figure 4 This is an experimental schematic diagram of the RaidVision all-in-one machine with the horizontal installation position as the variable, where (a) is the roadside and (b) is above the center of the road;

[0094] Figure 5 The diagram shows an experiment of the Rayvision integrated machine with the lateral rotation angle around the road as the variable, where (a) represents no rotation and (b) represents rotation angle.

[0095] Figure 6 This is a schematic diagram of an experiment using the longitudinal rotation angle of the radar-visual integrated machine around the road as a variable.

[0096] Figure 7 This is a schematic diagram of an experiment using the rotation angle around a vertical line as a variable for the Raveo all-in-one machine;

[0097] Figure 8This is a schematic diagram of an experiment using height as a variable for the Raveo all-in-one machine. Detailed implementation method:

[0098] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments. However, this description is not limited thereto. All aspects not described in detail in the present invention are based on conventional techniques in the field.

[0099] Example:

[0100] A radar-based integrated machine data acquisition platform was built, and experiments were conducted, such as... Figure 2 As shown;

[0101] This invention provides a comprehensive evaluation and optimization method for the installation scheme of integrated radar-visual systems based on variance maximization combined weighting, such as... Figure 1 As shown, it includes:

[0102] Step 1: Analyze the main factors affecting the erection scheme and design different working conditions as single variables to conduct a control experiment, specifically:

[0103] S1-1: Query the specifications of the millimeter-wave radar and camera sensors. In this embodiment, the millimeter-wave radar has a maximum detection range of 100m, a maximum horizontal angle of 47°, and a maximum elevation angle of 18°. According to the Pythagorean theorem, the horizontal distance at which the millimeter-wave radar detects a target is affected by the device's installation height. The camera lens has a focal length of 12mm and a maximum elevation angle of 26.2°. This means that the upper and lower edges of the camera's field of view will also change with the device's height, and the image clarity captured by the camera gradually decreases as the distance increases.

[0104] S1-2: Considering the road environment: Due to installation limitations, such as tree obstruction, the horizontal installation position needs to be adapted to local conditions. Common installation locations include the center of the road and one side of the road, such as... Figure 4 As shown, in addition to the above, the erection height is also one of the factors affecting the erection plan, such as... Figure 8 ;

[0105] S1-3: Considering the equipment's own rotation, within the road environment, a spatial rectangular coordinate system is established with the horizontal plane (ground) of the road as the X-axis, the vertical plane as the Y-axis, and the vertical line as the Z-axis. The projection point of the radar-view all-in-one machine on the ground is used as the origin. Figure 3 As shown, the rotation angle of the integrated radar-visual unit around the X, Y, and Z coordinate axes is a factor affecting the installation scheme;

[0106] S1-4: Design working conditions based on five variables: erection location, height, and three rotation angles. Conduct a control experiment, where the three rotation angles are... Figure 5 , Figure 6 , Figure 7 θ, β,

[0107] Step 2: Analyze the experimental data collected by the Ravec all-in-one machine, select indicator parameters, and construct an evaluation system, specifically as follows:

[0108] S2-1: Analyze the data collected by the Rayvision all-in-one machine;

[0109] S2-2: Constructing an evaluation index system

[0110] The monitoring capabilities of the integrated radar-visual machine are taken as the overall layer, while the ability to identify ordinary traffic incidents, the ability to identify special traffic incidents, and the monitoring range are taken as the primary evaluation indicators. The primary indicators are further divided into secondary evaluation indicators.

[0111] The traffic data generated by the radar-video integrated machine includes records of ordinary traffic incidents and records of special traffic incidents. 450 records of ordinary traffic incidents were selected, and the average confidence level (C1), standard deviation (C2), and frequency of no license plate occurrence (C3) of the 450 vehicles were used as the secondary indicator layer for ordinary traffic incident recognition capability. 100 records of special traffic incidents were selected, and the average confidence level (C4), standard deviation (C5), frequency of no license plate occurrence (C6), and average confidence level (C7) and standard deviation (C8) of each recognized character of the license plate were used as the secondary indicator layer for special traffic incident recognition capability. The maximum longitudinal monitoring range (C9) along the road direction was used as the secondary evaluation indicator for the radar-video integrated machine's monitoring range. C9 is defined as the y-range, which is the difference between the farthest and closest distances monitored by the radar-video integrated machine in the longitudinal direction of the road.

[0112] The evaluation index system is shown in Table 1. The index attributes represent the positive or negative correlation between each secondary index and the overall level, which can be qualitatively determined by torque. A positive correlation indicates that the evaluation index is positively correlated with the previous level index, and a negative correlation indicates that the evaluation index is negatively correlated with the previous level index.

[0113] Table 1: Evaluation Index System

[0114]

[0115]

[0116] Taking working condition one as an example, with the erection location as the single variable, the design of the erection scheme working condition is as follows:

[0117] Table 2: Erection Scheme for Working Condition 1

[0118]

[0119] The raw data collected by the Rave Integrated Machine was in multiple txt files. Python was used for data preprocessing to extract the valid data, as shown in Tables 3, 4, and 5.

[0120] Table 3: Data Files from the Host Computer System for Common Traffic Incidents

[0121]

[0122] Table 4: Data Files from the Host Computer System for Special Traffic Incidents

[0123]

[0124]

[0125] Table 5: Web Vehicle Identification Data Files

[0126] Serial Number serial number Location (x, y) Lane number speed Model 1 101 2，55.700001 3 59.8 small cars 2 103 -0.6，69.199997 3 50 small cars 3 101 2，53.2999999 3 59.4 small cars

[0127] Note: In Table 4, D1-D7 represent the confidence level of each identification character of the license plate. In the license plate number Lan Yu A00001 in Table 4, "Lan" indicates that the license plate is a blue license plate.

[0128] This embodiment selects 450 records of ordinary traffic incidents and 100 records of special traffic incidents. The average confidence score C1 for vehicle identification in ordinary traffic incidents is calculated using the following formula:

[0129]

[0130] Among them, EntireBelieve1 i This represents the vehicle identification confidence level for the i-th ordinary traffic incident;

[0131] The formula for calculating the standard deviation C2 of the confidence score for vehicle identification in ordinary traffic incidents is as follows:

[0132]

[0133] The probability of a vehicle without license plates appearing in a typical traffic incident is C3:

[0134]

[0135] n′ represents the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file of ordinary traffic events;

[0136] Average confidence level C4 for vehicle identification in special traffic incidents:

[0137]

[0138] Among them, EntireBelieve2 i This represents the confidence level for vehicle identification in the i-th special traffic incident;

[0139] Standard deviation of confidence level C5 for vehicle identification in special traffic incidents:

[0140]

[0141] The probability of vehicles without license plates appearing in special traffic incidents (C6):

[0142]

[0143] n″ represents the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file for special traffic events;

[0144] Average confidence level C7 for license plate character recognition in special traffic incidents:

[0145]

[0146] D j This represents the confidence level of the j-th license plate recognition character;

[0147] Standard deviation of confidence level for license plate character recognition in special traffic incidents (C8)

[0148]

[0149] The y-value range C9 is the difference between the farthest and closest distances monitored by the integrated radar-vision system in the longitudinal direction of the road. Its calculation formula is as follows:

[0150] C9 = y max -y min (9)

[0151] y max The maximum distance that the radar-vision integrated machine can monitor in the longitudinal direction of a road is y min This is the closest distance that the radar-vision integrated machine can monitor in the longitudinal direction of the road.

[0152] Based on the above work, the actual raw data collected under working condition one is shown in Table 6:

[0153] Table 6. Actual raw data collected under operating condition 1

[0154]

[0155] Step 3: The Analytic Hierarchy Process (AHP) is selected as the subjective weighting method, while the information weighting method, entropy weighting method, and CRITIC method are used as objective weighting methods. A combined weighting approach based on variance maximization is employed to comprehensively evaluate and optimize the various installation schemes for the integrated radar-visual system. Specifically:

[0156] S3-1: Calculate the weights of each indicator for the subjective weighting method using the analytic hierarchy process;

[0157] S3-2: Calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method based on SPSSAU (Statistical Product and Service Software Automatically, which is an existing software);

[0158] S3-3: Weights obtained by combining the analytic hierarchy process (AHP), information weighting method, entropy weighting method, and CRITIC method with variance maximization as the objective function;

[0159] S3-4: Using the weights of each indicator obtained from S3-3, a comprehensive score is given to the erection scheme for each working condition, and the optimal erection scheme is selected.

[0160] The calculation method of the analytic hierarchy process is as follows:

[0161] A. For indicators at the same level, their relative importance is quantified using a 1-9 scale, as shown in Table 7:

[0162] Table 7: Definition of Matrix Scale

[0163]

[0164] The larger the number, the higher the importance. The judgment matrix A1 of def is defined as follows:

[0165]

[0166] a kp This represents the ratio of the importance of factor k to that of factor p.

[0167] The judgment matrix for the subjectively defined evaluation indicators in this embodiment is as follows:

[0168] The primary evaluation indicators, namely the ability to identify ordinary traffic incidents (B1), the ability to identify special traffic incidents (B2), and the monitoring scope (B3), form a judgment matrix as follows:

[0169]

[0170] The subjective construction of the judgment matrix for the secondary evaluation indicators under the primary evaluation indicators is as follows. Since the third primary evaluation indicator contains only one parameter, only the first two judgment matrices need to be constructed:

[0171]

[0172]

[0173] B. Based on the judgment matrix, use MATLAB software to write code to solve for the largest eigenvalue λ. max And the eigenvector W, and perform normalization on it;

[0174]

[0175]

[0176] Where A1 is the judgment matrix, W is the eigenvector, and λ max For the largest eigenvalue, w i It is the weight of each indicator after normalizing the feature vector W, w l It refers to each vector value in the feature vector W.

[0177] The maximum eigenvalue λ of the judgment matrix of the primary and secondary evaluation indicators was calculated. max And the eigenvectors, after normalization of the eigenvectors, yield the weights W of each evaluation index, where the first-level evaluation index is:

[0178] λ max1 =3

[0179] W1 = [30% 60% 10%]

[0180] Secondary evaluation indicators:

[0181] λ max2 =3

[0182] W′2 = [60% 20% 20%]

[0183] λ max3 =5.3

[0184] W′3=[35.38% 12.47% 20.51% 21.11% 10.53%]

[0185] C. Calculate the consistency index CI based on the maximum eigenvalue of each judgment matrix obtained from the calculation. For example, look up the average random consistency index table in formula (13), find RI based on the order of the judgment matrix, as shown in Table 8, and calculate the consistency ratio CR:

[0186]

[0187]

[0188] n represents the order of the judgment matrix;

[0189] Table 8: Average Random Consistency Index

[0190]

[0191] In this embodiment, the consistency ratio CR of the judgment matrices of the primary evaluation index and the secondary evaluation index is calculated, wherein the primary evaluation index is:

[0192] CR=0

[0193] Secondary evaluation indicators:

[0194]

[0195] If CR≤0.1, which indicates consistency, the judgment matrix can be used. The weights of each indicator factor are reasonably allocated, and the vector corresponding to its largest eigenvalue is the weight vector of each indicator in that layer.

[0196] If CR≤0.1 is not satisfied, repeat steps A to C to compare the importance of each indicator again and construct a judgment matrix;

[0197] In this embodiment, based on the calculation results, CR ≤ 0.1, indicating consistency. The vector corresponding to its largest eigenvalue is the weight vector of each indicator in this layer. By multiplying the weight of the secondary indicator with the corresponding weight of the primary indicator and normalizing, the indicator weight of the secondary indicator relative to the overall layer can be obtained, as follows:

[0198] W

[0199] = [18% 6% 6% 21.23% 7.48% 12.31% 12.66% 6.32% 10%]

[0200] Among them, 18% = 30% * 60%, 6% = 30% * 20%, 6% = 30% * 20%, 21.23% = 60% * 35.38%, 7.48% = 60% * 12.47%, 12.31% = 60% * 20.51%, 12.66% = 60% * 21.11%, 6.32% = 60% * 10.53%, 10% = 10% * 1.

[0201] The weights obtained from the information weight method, entropy weight method, and CRITIC method in the objective weighting method are calculated based on SPSSAU (Statistical Product and Service Software Automatically). Specifically:

[0202] For the collected raw data r lj Take the reciprocal of the negative indicator to make it positive, and then normalize all the positive indicators to obtain r. lj The matrix is ​​as shown in formula (15);

[0203]

[0204] Where, r ljLet be the value of the l-th erection scheme under the j-th index parameter, m represent the total number of erection schemes in the working condition, and r lj正向 A new indicator obtained by taking the reciprocals of the positive and negative indicators in the evaluation system;

[0205] After normalization, the original data table 6 is transformed into a 2×9 data matrix.

[0206]

[0207] like

[0208] The normalized matrix is ​​input into the SPSSAU analysis software to calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method.

[0209] Using variance maximization as the objective function, the weights obtained by combining the above four methods are as follows:

[0210] ① Regarding the weight value of the j-th indicator parameter, assuming the weight range obtained by the four methods is as follows: Combined weight w j It should meet the following conditions:

[0211]

[0212] ② Four subjective and objective weighting methods were used to obtain the index weight matrix, as shown in formula (17);

[0213]

[0214] In this matrix, w 32 This represents the weight value assigned to the second evaluation indicator by the third weighting method, and the same applies to others.

[0215] The weights of the nine secondary evaluation indicators are obtained based on the above-mentioned analytic hierarchy process and are used as the first weighting method, i.e., the first row of the indicator weight matrix. The weights obtained by the objective weighting methods, information weighting method, entropy weighting method and CRITIC method are calculated and used as the second to fourth rows of the indicator weight matrix. The weight vectors obtained by the four subjective and objective weighting methods form the indicator weight matrix.

[0216]

[0217] ③ Take the largest variance of the score of the setup scheme under the combined weight as the objective function, make the sum of the reasonable range of the attribute and the attribute weight equal to 1, establish an optimization model, and obtain the combined weight;

[0218] The optimized model is as follows:

[0219]

[0220]

[0221] Where maxσ(θ) represents the objective function for maximizing variance, w j ' represents a combination matrix, Indicates r lj The average value of ′ and These represent the four methods for finding the maximum and minimum weights, respectively.

[0222] Based on the above formula, for ease of calculation, the combined weights can be obtained by writing code in Python:

[0223] W′

[0224] = [0.11099 0.11099 0.11099 0.11099 0.11099 0.1231 0.11099 0.11099 0.1]

[0225] The Python code for maximizing variance with subjective and objective weights is as follows:

[0226]

[0227]

[0228]

[0229] ④ Using the weights of each indicator obtained in step ③, a comprehensive score is given to the erection scheme for each working condition, and the optimal erection scheme is selected. Based on the combined weights obtained in step ③, the comprehensive score is as follows:

[0230]

[0231] Table 9. Comprehensive Evaluation Score for Working Condition 1 (×10) -3 )

[0232]

[0233] For example, the data in Table 9 includes 5.741 * 10. -3 =0.11099*0.493.

[0234] Based on the comprehensive score obtained from the above evaluation techniques, Experiments 1-2 scored higher, meaning that the installation scheme with the integrated radar-visual unit installed above the center of the road at a height of 8.9m and rotating 0 degrees around the three coordinate axes of the spatial coordinate system is the preferred scheme.

[0235] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A comprehensive evaluation and optimization method for the installation scheme of integrated radar-visual equipment based on variance maximization combined weighting, characterized in that, include: Step 1: Analyze the main factors affecting the erection scheme and design different working conditions as single variables to conduct a control experiment; Step 2: Analyze the experimental data collected by the Ravec all-in-one machine, select indicator parameters, and construct an evaluation system; Step 3: Select the analytic hierarchy process (AHP) as the subjective weighting method, and the information weighting method, entropy weighting method, and CRITIC method as the objective weighting methods. Combine them for weighting based on variance maximization, and comprehensively evaluate and optimize the various installation schemes of the integrated radar screen. The method for analyzing the main factors affecting the erection scheme in step 1 is as follows: S1-1: Query the specifications of millimeter-wave radar and camera sensing devices; S1-2: Considering the road environment: Due to installation limitations, consider the installation location and height. The installation location includes the center of the road and one side of the road. S1-3: Considering the rotation of the equipment itself, in the road environment, the horizontal direction of the road is the X-axis, the vertical direction of the road is the Y-axis, and the vertical line is the Z-axis. The radar-view integrated machine is established with the ground projection point as the origin. The rotation angle of the radar-view integrated machine around the X, Y, and Z coordinate axes is a factor affecting the installation scheme. S1-4: Design working conditions based on five variables, including installation location, height, and three rotation angles, and conduct a control experiment; Step 2 is as follows: S2-1: Analyze the data collected by the Rayvision all-in-one machine; S2-2: Constructing an evaluation index system The monitoring capabilities of the integrated radar-visual machine are taken as the overall layer, while the ability to identify ordinary traffic incidents, the ability to identify special traffic incidents, and the monitoring range are taken as the primary evaluation indicators. The primary indicators are further divided into secondary evaluation indicators. The traffic data generated by the radar-visual integrated machine includes records of ordinary traffic incidents and records of special traffic incidents. 450 ordinary traffic incident records were selected, and the average confidence level C1 and standard deviation of the recognition confidence levels for these 450 vehicles were used as the basis for the analysis. and the frequency of vehicles without license plates As a secondary indicator layer for the ability to identify ordinary traffic incidents; 100 special traffic incident information records were selected, and the average confidence score of the 100 vehicles was used as the basis for the evaluation. Standard deviation Frequency of vehicles without license plates The average confidence score of each recognition character in the license plate and standard deviation As a secondary indicator layer for the ability to identify special traffic incidents; based on the maximum monitoring range along the road direction, i.e., longitudinally. As a secondary evaluation indicator of the monitoring range of the radar video integrated machine, among which, Defined as the y-value range, which is the difference between the farthest and closest distances monitored by the integrated radar-vision system in the longitudinal direction of the road; Average confidence level of vehicle identification in common traffic incidents The calculation formula is as follows: （1） This represents the vehicle identification confidence level for the i-th ordinary traffic incident; Standard deviation of confidence level for vehicle identification in ordinary traffic incidents The calculation formula is as follows: （2） Probability of vehicles without license plates appearing in ordinary traffic incidents : （3） This indicates the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file for ordinary traffic incidents. Average confidence level of vehicle identification in special traffic incidents : （4） This represents the confidence level for vehicle identification in the i-th special traffic incident; Standard deviation of confidence level for vehicle identification in special traffic incidents : （5） Probability of vehicles without license plates appearing in special traffic incidents : （6） This indicates the number of null values ​​in the overall confidence score of license plate recognition in the host computer data file for special traffic events. Average confidence level of license plate character recognition in special traffic incidents : （7） This represents the confidence level of the j-th license plate recognition character; Standard deviation of confidence level for license plate character recognition in special traffic incidents : （8） Very bad y value The difference between the farthest and closest distances monitored by the radar-vision integrated camera in the longitudinal direction of the road is calculated using the following formula: （9） This represents the furthest distance that the radar-vision integrated machine can monitor in the longitudinal direction of a road. This is the closest distance that the radar-vision integrated machine can monitor in the longitudinal direction of the road.

2. The comprehensive evaluation and optimization method for the installation scheme of the integrated radar-visual machine based on variance maximization combined weighting as described in claim 1, characterized in that, Step 3 includes: S3-1: Calculate the weights of each indicator for the subjective weighting method using the analytic hierarchy process; S3-2: Calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method based on SPSSAU; S3-3: Weights obtained by combining the analytic hierarchy process (AHP), information weighting method, entropy weighting method, and CRITIC method with variance maximization as the objective function; S3-4: Using the weights of each indicator obtained from S3-3, a comprehensive score is given to the erection scheme for each working condition, and the optimal erection scheme is selected.

3. The comprehensive evaluation and optimization method for the installation scheme of the integrated radar-visual machine based on variance maximization combined weighting as described in claim 2, characterized in that, The calculation method of the analytic hierarchy process is as follows: A. For indicators at the same level, their relative importance is quantified using a 1-9 scale; the larger the number, the higher the importance. A judgment matrix is ​​defined. as follows: （10） This represents the ratio of the importance of factor k to that of factor p. B. Find the largest eigenvalue based on the judgment matrix. and eigenvectors And normalize it; （11） （12） in, To determine the matrix, For feature vectors, The largest eigenvalue, It is the feature vector The weights of each indicator after normalization. This refers to each vector value in the feature vector W; C. Calculate the consistency index based on the maximum eigenvalue of each judgment matrix obtained from the calculation. Query the average random consistency index table and find the value based on the order of the judgment matrix. Calculate the consistency ratio : （13） （14） n represents the order of the judgment matrix; If satisfied If the value is ≤0.1, indicating consistency, the judgment matrix is ​​used. The weights of each indicator factor are reasonably allocated, and the vector corresponding to its largest eigenvalue is the weight vector of each indicator in that layer. If not satisfied If the value is ≤0.1, repeat steps A to C to compare the importance of each indicator again and construct a judgment matrix.

4. The comprehensive evaluation and optimization method for the installation scheme of the integrated radar-visual machine based on variance maximization combined weighting as described in claim 3, characterized in that, The weights obtained from the information weight method, entropy weight method, and CRITIC method in the objective weighting method are calculated as follows: The raw data collected Take the reciprocal of the negative indicators to make them positive, and then normalize all the positive indicators. Matrix, as shown in formula (15); （15） in, For the first The value of each erection scheme under the j-th index parameter, where m represents the total number of erection schemes in the working condition. A new indicator obtained by taking the reciprocals of the positive and negative indicators in the evaluation system; The normalized matrix is ​​input into the SPSSAU analysis software to calculate the weights obtained by the objective weighting method, information weighting method, entropy weighting method, and CRITIC method.

5. The comprehensive evaluation and optimization method for the installation scheme of the integrated radar-visual machine based on variance maximization combined weighting as described in claim 4, characterized in that, Using variance maximization as the objective function, the weights obtained by combining the above four methods are as follows: For the weight value of the j-th indicator parameter, assuming the weight range formed by the weights obtained from the four methods is as follows: Combined weights The following conditions must be met: （16） Four subjective and objective weighting methods were used to obtain the index weight matrix, as shown in formula (17). （17） The objective function is to take the largest variance of the score of the setup scheme under the combined weights, so that the sum of the reasonable ranges of the attributes and the attribute weights is 1. An optimization model is established to obtain the combined weights. The optimized model is as follows: （18） （19） in, This represents the objective function that maximizes variance. Represents a combination matrix. express The average value, and These represent the four methods for finding the maximum and minimum weights, respectively. Based on steps The combined weights were used to obtain the overall score as follows: （20） The option with the highest overall score is selected as the preferred option.

Images