Highway tunnel entrance sight guidance system evaluation method based on deep learning

Abstract

The invention relates to a deep learning-based expressway tunnel entrance sight guidance system evaluation method. The method comprises the following steps of S1, constructing a deep learning-based expressway tunnel outer side entrance section sight guidance system evaluation system; s2, collecting pictures of the sight line guidance system at the entrance of the highway tunnel, and processing the pictures to obtain a data set; s3, constructing a deep learning model for identifying the sight line induction facility at the entrance of the highway tunnel; s4, training and verifying different deep learning models by adopting the data set, and adjusting parameters of different models to obtain an optimal model; and S5, a dual-mode mechanism, a rating mechanism and a feedback mechanism are embedded into the evaluation system, and single-tunnel fine evaluation and multi-tunnel rapid examination are realized. The method is rapid and efficient, can inspect low safety levels in batches, can significantly improve the evaluation efficiency, saves human resources and cost, inspects tunnels needing safety improvement, and improves the tunnel driving safety in China.

Description

Technical Field

[0001] The present invention relates to the field of highway traffic safety, and more specifically, to an evaluation method for a sight guidance system at the entrance of a highway tunnel based on deep learning. Background Art

[0002] With the continuous increase in the number of highway tunnels, the traffic safety problems in tunnels have become increasingly prominent and have become important issues that need to be solved urgently. The occurrence of highway tunnel traffic accidents is not evenly distributed in all areas of the tunnel. The entrance and exit areas of the tunnel are often high-incidence areas of accidents. This is related to the difficulties of drivers in aspects such as visual adaptation and speed control when entering and leaving the tunnel, and is also closely related to the special environmental differences inside and outside the tunnel. At the entrance section of the tunnel, due to the significant differences in the roadbed sections inside and outside the tunnel, drivers often need to experience a visual adaptation process from bright to dim, which to a certain extent increases the driving difficulty and potential risks. Common accident types at the entrance section include vehicles hitting the portal. This is often because drivers fail to accurately judge the position of the tunnel entrance or drive too fast, resulting in the vehicle colliding with the tunnel portal.

[0003] Sight guidance facilities can improve the tunnel light environment at low cost and guide drivers to drive into the tunnel safely and comfortably. However, the existing industry specifications are difficult to meet the sight guidance needs of drivers, and the safety evaluation of the sight guidance system at the tunnel entrance is based on expert experience and on-site investigation. However, this method has low evaluation efficiency, limited scope, and high evaluation costs for hiring experts. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an evaluation method for a sight guidance system at the entrance of a highway tunnel based on deep learning, which alleviates the problems of low efficiency, limited scope, and high cost of the existing evaluation methods, and can significantly improve the evaluation efficiency, save human resources and costs.

[0005] The technical solution adopted by the present invention to solve its technical problems is: construct an evaluation method for a sight guidance system at the entrance of a highway tunnel based on deep learning, including the following steps:

[0006] S1. Construct an evaluation system for the sight guidance system at the outer entrance section of a highway tunnel based on deep learning;

[0007] S2. Collect pictures of the sight guidance system at the entrance of the highway tunnel, and process the pictures to obtain a data set;

[0008] S3. Construct a deep learning model for identifying sight guidance facilities at the entrance of a highway tunnel;

[0009] S4. Use the data set to train and verify different deep learning models, adjust the parameters of different models, and obtain an optimal model;

[0010] S5. Embed a dual-mode mechanism, a rating mechanism, and a feedback mechanism into the evaluation system to achieve fine evaluation of a single tunnel and rapid review of multiple tunnels.

[0011] According to the above solution, in the step S1, constructing an evaluation system for the sight guidance system at the outer entrance section of a highway tunnel based on deep learning includes the following steps:

[0012] S101. Segment the tunnel entrance area.

[0013] S102. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on redundancy.

[0014] S103. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on integrity.

[0015] S104. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on constancy.

[0016] S105. Integrate the evaluation systems for the sight guidance system at the highway tunnel entrance based on redundancy, integrity, and constancy to obtain an evaluation system for the sight guidance system at the highway tunnel entrance based on deep learning.

[0017] According to the above solution, in the step S101, the tunnel entrance area is segmented by the following method:

[0018] According to different types of sight guidance facilities, the highway tunnel entrance area includes a sensing and recognition section, a decision-making and execution section, and a relaxation and adaptation section:

[0019] The sensing and recognition section is located outside the tunnel, 150 - 300 meters away from the tunnel entrance. The sight guidance facilities in the sensing and recognition section play a role in sensing and recognition, warning the driver that there is a tunnel ahead.

[0020] The decision-making and execution section is located outside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in the decision-making and execution section mainly play a role in induction and protection, guiding the driver to safely drive into the tunnel.

[0021] The relaxation and adaptation section is located inside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in the relaxation and adaptation section mainly play a role in induction and relaxation, helping the driver gradually adapt to the light and space changes inside the tunnel to ensure driving safety and comfort.

[0022] According to the above solution, in the step S102,

[0023] The evaluation system of the sight guidance system at the highway tunnel entrance based on redundancy evaluates from the hierarchy and completeness of the sight guidance facilities. The sight guidance facilities include low-level, middle-level, and high-level ones according to the induction level.

[0024] In the sensing and recognition section, the low-level facilities are raised road markers and visual deceleration markings, the middle-level facilities are outline markers and elastic traffic columns, and the high-level facilities are warning linear induction markers and tunnel entrance advance signs.

[0025] In the decision-making and execution section, the low-level facilities are raised road markers and vibration-type deceleration markings, the middle-level facilities are outline markers, elastic traffic columns, crash barrels, crash cushions, and portal elevation markings, and the high-level facilities are tunnel information signs, warning linear induction markers, and portal elevation markings.

[0026] In the relaxation and adaptation section, the low-level facilities are raised road markers, inspection path outline markers, and road edge elevation markings, the middle-level facilities are elastic traffic columns, side wall outline markers, reflective strips, and waistlines, and the high-level facilities are reflective rings.

[0027] According to the low-level, middle-level, and high-level facility types in the sensing and recognition section, decision-making and execution section, and relaxation and adaptation section, the redundancy is evaluated. The calculation formula is as follows:

[0028]

[0029] Among them, i is different tunnel entrance area segments. i = 1 is the sensing and recognition section, i = 2 is the decision-making and execution section, and i = 3 is the relaxation and adaptation section; R i is the redundancy evaluation score for the i-th segment; j is different induction levels of the sight guidance facilities. j = 1 is low-level, j = 2 is middle-level, and j = 3 is high-level; α j is the weight coefficient of the j-level induction facilities, and α1 + α2 + α3 = 1. According to the importance of different-level induction facilities, the low-level facilities are set to 0.4, the middle-level facilities are set to 0.4, and the high-level facilities are set to 0.2; N ij is the total type number of the j-level induction facilities in the i-th segment; n ij is the actual existing type number of the j-level induction facilities recognized in the i-th segment; S1 is the total score of the highway tunnel entrance area based on the redundancy evaluation.

[0030] According to the above scheme, in step S103,

[0031] The evaluation based on integrity evaluates from the overall combined performance and effect of the sight guidance facilities. The induction effects of the sight guidance system in the highway tunnel entrance area include high-level induction, middle-level induction, basic induction, and unqualified induction.

[0032] In the sensing and identification section, the advanced guidance includes raised pavement markers, visual deceleration markings, elastic traffic columns, warning linear guide signs, outline markers, and tunnel entrance advance signs; the intermediate guidance includes raised pavement markers, elastic traffic columns, visual deceleration markings, outline markers, and tunnel entrance advance signs; the basic guidance includes raised pavement markers, outline markers, and tunnel entrance advance signs; the unqualified guidance is a combination of facilities that do not meet the basic guidance.

[0033] In the decision execution section: The advanced guidance consists of raised pavement markers, visual deceleration markings, crash barrels, crash cushions, elastic traffic columns, portal elevation markings, outline markers, warning linear guide signs, and tunnel information signs; the intermediate guidance includes raised pavement markers, visual deceleration markings, crash barrels, portal elevation markings, outline markers, warning linear guide signs, and tunnel information signs; the basic guidance consists of raised pavement markers, portal elevation markings, and outline markers; the unqualified guidance is a combination of facilities that do not meet the basic guidance.

[0034] In the decision execution section, the advanced guidance includes raised pavement markers, roadside elevation markings, elastic traffic columns, maintenance road outline markers, sidewall outline markers, waistlines, reflective strips, and reflective rings; the intermediate guidance includes raised pavement markers, maintenance road outline markers, sidewall outline markers, waistlines, and reflective strips; the basic guidance consists of raised pavement markers, maintenance road outline markers, and sidewall outline markers; the unqualified guidance is a combination of facilities that do not meet the basic guidance.

[0035] Based on the facility types of the advanced guidance, intermediate guidance, basic guidance, and unqualified guidance in the sensing and identification section, the decision execution section, and the decision execution section, evaluate the integrity. The calculation formula is as follows:

[0036]

[0037] Among them, S2 is the total score based on the integrity evaluation of the highway tunnel entrance area; i is different segments of the tunnel entrance area, i = 1 is the sensing and identification section, i = 2 is the decision execution section, i = 3 is the transition and adaptation section; Z i is the score based on the integrity evaluation of the i-th section. The advanced guidance is 90 points, the intermediate guidance is 80 points, the basic guidance is 60 points, and the unqualified guidance is 30 points.

[0038] According to the above scheme, in step S104,

[0039] The evaluation based on constancy evaluates from the stability and visibility of the sight guidance facilities, and realizes the quantitative evaluation of the brightness stability and facility visibility of some sight guidance facilities.

[0040] The brightness stability is evaluated by the color differences in the same picture caused by different distances from and lighting of the same type of facilities; the visibility of the facilities is evaluated by the contrast differences between the facilities and their surrounding environments.

[0041] In the sensing and recognition section, the objects of constancy evaluation are elastic traffic columns and warning linear guide signs; in the decision-making and execution section, the objects of constancy evaluation are elastic traffic columns, warning linear guide signs, and anti-collision barrels; in the mitigation and adaptation section, the objects of constancy evaluation are elastic traffic columns and reflective strips.

[0042] Based on the brightness stability and facility visibility of the evaluation objects in the sensing and recognition section, the sensing and recognition section, and the mitigation and adaptation section, the constancy is evaluated, and the calculation formula is as follows:

[0043]

[0044] Among them, S3 is the total score of the highway tunnel entrance area based on the constancy evaluation; i is different subsections of the tunnel entrance area, i = 1 is the sensing and recognition section, i = 2 is the decision-making and execution section, and i = 3 is the mitigation and adaptation section; W i is the evaluation score of the brightness stability of the i-th section; K i is the evaluation score of the facility visibility of the i-th section; β1 and β2 are the weight coefficients of brightness stability and facility visibility, and β1 + β2 = 1.

[0045] According to the above scheme, in the step S105,

[0046] By integrating the evaluation system of the highway tunnel entrance sight guidance system based on redundancy, integrity, and constancy, an evaluation system of the highway tunnel entrance sight guidance system based on deep learning is obtained, and the specific calculation formula is as follows:

[0047] S = a1S1 + a2S2 + a3S3

[0048] Among them, S is the total evaluation score of the highway tunnel entrance sight guidance system based on deep learning, with a value range of 0 - 100; S1, S2, and S3 are the evaluation scores based on redundancy, integrity, and constancy respectively; a1, a2, and a3 are the weight coefficients of the evaluations based on redundancy, integrity, and constancy respectively, and a1 + a2 + a3 = 1;

[0049] The total score S is divided into four induction effects according to the score size, from high to low are high-level induction (85 - 100), medium-level induction (70 - 85), basic induction (60 - 70), and unqualified induction (0 - 60).

[0050] According to the above solution, in step S2, the sources of the pictures of the highway tunnel entrance sight guidance system collected include real vehicle shooting and video capture from a driving recorder. Processing the pictures of the highway tunnel entrance sight guidance system includes the following steps:

[0051] Clean and screen the pictures, standardize the pictures, label the screened pictures, perform data augmentation on the labeled pictures, and divide the pictures into a data set;

[0052] The cleaning and screening include deleting pictures with unqualified shooting quality from the collected pictures and screening out front tunnel entrance pictures that meet the requirements;

[0053] The standardization step includes unifying the picture size and uniformly outputting the pictures in png format;

[0054] The data augmentation operations include flipping, rotating, scaling, adding noise, blurring, adjusting brightness or contrast, and affine transformation;

[0055] The specific operation of the division includes dividing the training set, validation set, and test set in a ratio of 6:2:2;

[0056] In step S3, construct different deep learning models for identifying highway tunnel entrance sight guidance facilities, and adjust the deep learning models using a multi-scale object recognition strategy;

[0057] The deep learning models include YOLOV7, EfficientDet, and Faster R-CNN;

[0058] The multi-scale object recognition strategy includes adding an attention mechanism to the model and introducing a feature pyramid network.

[0059] According to the above solution, in step S4, define the loss function, select the optimizer, set the hyperparameters, and judge the fitting situation of the deep learning model according to the training loss total_loss and validation loss val_loss during the training process;

[0060] Adjust the model structure or hyperparameters according to the performance of the deep learning model on the validation set, perform repeated training, and finally select the optimal model according to the comprehensive performance of the precision Precision, recall Recall, accuracy Accuracy, and mean average precision mAP of the deep learning model in the test set.

[0061] According to the above solution, in step S5, the dual-mode mechanism, rating mechanism, and feedback mechanism embedded in the evaluation system are specifically:

[0062] The dual - mode mechanism is single - tunnel evaluation and multi - tunnel evaluation; single - tunnel evaluation is used for fine evaluation of a single tunnel, and outputs the evaluation results of various indicators for each segment of the single tunnel; multi - tunnel evaluation is used for rapid review of multiple tunnels to identify tunnels with relatively large potential safety hazards in the input tunnels, that is, unqualified induced tunnels.

[0063] The rating mechanism is set according to the evaluation system of the highway tunnel entrance sight induction system based on deep learning, and controls the deep - learning model to output evaluation results and scores of various indicators, which are used for single - tunnel evaluation and multi - tunnel evaluation.

[0064] The feedback mechanism collects the list of tunnels with unqualified induced effects according to the output rating results, which is used for multi - tunnel evaluation.

[0065] Implementing the evaluation method of the highway tunnel entrance sight induction system based on deep learning of the present invention has the following beneficial effects:

[0066] 1. The present invention provides a unified, standard and objective evaluation basis for the evaluation of the highway tunnel entrance sight induction system based on deep learning, laying a foundation for future research on the evaluation of the highway tunnel entrance sight induction system based on deep learning; at the same time, it provides strong technical support for the safety management of highways, promotes the standardization and regularization development of related technologies, and is of great significance for improving the overall operation level and traffic safety of highways in China.

[0067] 2. Using the deep - learning model for evaluation in the present invention avoids the high costs of hiring experts for evaluation and on - site inspections, saves human resources and costs, and can objectively and comprehensively evaluate the effectiveness of the tunnel entrance sight induction system, providing a scientific reference basis for highway management departments on issues related to the improvement of tunnel entrance safety.

[0068] 3. The present invention uses means such as panoramic static images of Baidu Maps that can obtain pictures of highway tunnel entrances. Through the deep - learning model, it can quickly and efficiently conduct batch reviews of a large number of tunnel entrances, accurately identify tunnels with low safety levels, and timely discover potential safety hazards. Based on the batch review results of the present invention, the management department can formulate targeted safety improvement plans, prioritize the treatment of tunnels with serious safety hazards, thereby effectively improving the driving safety in tunnels, reducing the incidence of traffic accidents, and ensuring the travel safety of the people; this innovative application will have a profound impact on the safety management of highway tunnels in China and promote the further development of the driving safety level in tunnels in China. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0070] Figure 1Flow chart of the evaluation method for the line-of-sight induction system at the entrance of expressway tunnels based on deep learning of the present invention;

[0071] Figure 2 Schematic diagram of the sectionalization of the entrance area of expressway tunnels of the present invention;

[0072] Figure 3 Constitution diagram of the evaluation system of the present invention;

[0073] Figure 4 Schematic diagram of the line-of-sight induction facilities in the sensing and recognition section of the present invention;

[0074] Figure 5 Schematic diagram of the line-of-sight induction facilities in the decision-making and execution section of the present invention;

[0075] Figure 6 Schematic diagram of the line-of-sight induction facilities in the mitigation and adaptation section of the present invention;

[0076] Figure 7 Schematic diagram of the portal elevation markings of the present invention;

[0077] Figure 8 Schematic diagram of the steps for quantifying the brightness stability of the present invention;

[0078] Figure 9 Schematic diagram of the steps for quantifying the visibility of facilities of the present invention;

[0079] Figure 10 Schematic diagram of the steps for image acquisition and processing of the present invention;

[0080] Figure 11 Schematic diagram of the steps for model construction and selection of the present invention;

[0081] In the figure: 1, raised pavement markers; 2, visual deceleration markings; 3, outline markers; 4, elastic traffic posts; 5, warning line induction markers; 6, tunnel entrance warning signs; 7, vibration-type deceleration markings; 8, crash barrels; 9, crash cushions; 10, tunnel information signs; 11, inspection road outline markers; 12, curb elevation markings; 13, sidewall outline markers; 14, reflective strips; 15, waistlines; 16, reflective rings; 17, 2.5-meter portal elevation markings; 18, circular portal elevation markings; 19, circular arrowed portal elevation markings. Detailed implementation manners

[0082] For a clearer understanding of the technical features, purposes and effects of the present invention, the detailed implementation manners of the present invention will now be described in detail with reference to the accompanying drawings.

[0083] As Figures 1 - 11 shown, in the evaluation method for the line-of-sight induction system at the entrance of expressway tunnels based on deep learning of the present invention, the following steps are included:

[0084] S1. Construct an evaluation system for the sight guidance system at the outer entrance section of highway tunnels based on deep learning. Specifically:

[0085] S101. Segment the tunnel entrance area.

[0086] According to different types of sight guidance facilities, the entrance area of highway tunnels is divided into a perception and recognition section, a decision-making and execution section, and a mitigation and adaptation section.

[0087] The perception and recognition section is located outside the tunnel, 150 - 300 meters away from the tunnel entrance. The sight guidance facilities in this section mainly play the role of perception and recognition, warning the driver that there is a tunnel ahead.

[0088] The decision-making and execution section is located outside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in this section mainly play the role of guidance and protection, guiding the driver to safely drive into the tunnel.

[0089] The mitigation and adaptation section is located inside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in this section mainly play the role of guidance and mitigation, helping the driver gradually adapt to the light and space changes inside the tunnel to ensure driving safety and comfort.

[0090] S102. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on redundancy.

[0091] The evaluation system for the sight guidance system at the highway tunnel entrance based on redundancy evaluates from the hierarchy and completeness of the sight guidance facilities. The sight guidance facilities are divided into low-level, middle-level, and high-level according to their guidance levels.

[0092] In the perception and recognition section, the low-level facilities are raised road markings 1 and visual deceleration markings 2, the middle-level facilities are contour markers 3 and elastic traffic columns 4, and the contour marker 3 is the guardrail in this embodiment. The high-level facilities are warning linear induction markers 5 and tunnel entrance warning signs 6.

[0093] In the decision-making and execution section, the low-level facilities are raised road markings 1 and vibration-type deceleration markings 7, the middle-level facilities are contour markers 3, elastic traffic columns 4, crash barrels 8, crash pads 9, 2.5-meter portal elevation markings 17, and the contour marker 3 is the guardrail in this embodiment. The high-level facilities are tunnel information signs 10, warning linear induction markers 5, circular portal elevation markings 18 or circular arrowed portal elevation markings 19.

[0094] In the mitigation and adaptation section, the low-level facilities are raised road markings 1, inspection road contour markers 11, and roadside elevation markings 12, the middle-level facilities are elastic traffic columns 4, sidewall contour markers 13, reflective strips 14, and waistlines 15, and the high-level facilities are reflective rings 16.

[0095] Evaluate the redundancy according to the facility types at each segment and level, and the calculation formula is as follows:

[0096]

[0097] Wherein, i is different segments of the tunnel entrance area, i = 1 is the sensing and identification segment, i = 2 is the decision-making and execution segment, i = 3 is the mitigation and adaptation segment; R i is the redundancy evaluation score of segment i; j is different induction levels of the sight guidance facilities, j = 1 is the low level, j = 2 is the middle level, j = 3 is the high level; α j is the weight coefficient of the induction facilities at level j, and α1 + α2 + α3 = 1. According to the importance of the induction facilities at different levels, it is advisable to set the low-level facilities to 0.4, the middle-level facilities to 0.4, and the high-level facilities to 0.2; N ij is the total type number of the induction facilities at level j in segment i; n ij is the actual existing type number of the induction facilities at level j in segment i that are recognized; S1 is the total score of the redundancy evaluation of the highway tunnel entrance area.

[0098] S103. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system of the highway tunnel entrance based on integrity;

[0099] The evaluation based on integrity evaluates from the overall combined performance and effect of the sight guidance facilities. The induction effect of the sight guidance system in the highway tunnel entrance area is divided into high-level induction, middle-level induction, basic induction and unqualified induction;

[0100] In the sensing and identification segment, high-level induction includes raised pavement markers 1, visual deceleration markings 2, elastic traffic columns 4, warning line induction markers 5, profile markers 3 and tunnel entrance warning signs 6; middle-level induction includes raised pavement markers 1, elastic traffic columns 4, visual deceleration markings 2, profile markers 3 and tunnel entrance warning signs 6; basic induction includes raised pavement markers 1, profile markers 3 and tunnel entrance warning signs 6; unqualified induction is the facility combination that does not meet the basic induction. In this embodiment, the profile markers 3 are all guardrails.

[0101] In the decision-making and execution segment, high-level induction includes raised pavement markers 1, visual deceleration markings 2, crash barrels 8, crash pads 9, elastic traffic columns 4, ring-shaped arrow portal elevation markings 19, profile markers 3, warning line induction markers 5 and tunnel information signs 10; middle-level induction includes raised pavement markers 1, visual deceleration markings 2, crash barrels 8, ring-shaped portal elevation markings 18, profile markers 3, warning line induction markers 5 and tunnel information signs 10; basic induction includes raised pavement markers 1, 2.5-meter portal elevation markings 17 and profile markers 3. Unqualified induction is the facility combination that does not meet the basic induction. In this embodiment, the profile markers 3 are all guardrails.

[0102] In the mitigation and adaptation section, the advanced guidance includes raised pavement markers 1, curb face markings 12, flexible traffic posts 4, maintenance road profile markers 11, sidewall profile markers 13, waistlines 15, reflective strips 14, and reflective rings 16; the intermediate guidance includes raised pavement markers 1, maintenance road profile markers 11, sidewall profile markers 13, waistlines 15, and reflective strips 14; the basic guidance includes raised pavement markers 1, maintenance road profile markers 11, and sidewall profile markers 13; the non-compliant guidance is the facility combination that does not meet the basic guidance.

[0103] Evaluate the integrity based on the facility types with different induction effects in each section. The calculation formula is as follows:

[0104]

[0105] Among them, S2 is the total score based on the integrity evaluation of the highway tunnel entrance area; i is different sections of the tunnel entrance area, i = 1 is the sensing and recognition section, i = 2 is the decision-making and execution section, i = 3 is the mitigation and adaptation section; Z i is the score based on the integrity evaluation of section i. The advanced guidance is 90 points, the intermediate guidance is 80 points, the basic guidance is 60 points, and the non-compliant guidance is 30 points;

[0106] S104. In each section of the tunnel entrance area, construct an evaluation system for the sight guidance system of highway tunnel entrances based on constancy;

[0107] The evaluation based on constancy evaluates from the stability and visibility of the sight guidance facilities, and realizes the quantitative evaluation of the brightness stability and facility visibility of some sight guidance facilities;

[0108] The brightness stability is evaluated through the color differences caused by different distances and illuminations from the same type of facilities in the same picture; the facility visibility is evaluated through the contrast differences between the facility and its surrounding environment;

[0109] In the sensing and recognition section, the evaluation objects of constancy are flexible traffic posts 4 and warning line induction signs 5; in the decision-making and execution section, the evaluation objects of constancy are flexible traffic posts 4, warning line induction signs 5, and crash barrels 8; in the mitigation and adaptation section, the evaluation objects of constancy are flexible traffic posts 4 and reflective strips 14;

[0110] Evaluate constancy based on the brightness stability and facility visibility of the evaluation objects in each section. The calculation formula is as follows:

[0111]

[0112] Among them, S3 is the total score of the highway tunnel entrance area based on the constancy evaluation; i is different segments of the tunnel entrance area, where i = 1 is the perception and recognition segment, i = 2 is the decision execution segment, and i = 3 is the mitigation and adaptation segment; W i is the evaluation score of the brightness stability of the i-th segment; K i is the evaluation score of the facility visibility of the i-th segment; β1 and β2 are the weight coefficients of brightness stability and facility visibility, and β1 + β2 = 1.

[0113] S105. Integrate the evaluation system of the highway tunnel entrance sight guidance system based on redundancy, integrity, and constancy to obtain the evaluation system of the highway tunnel entrance sight guidance system based on deep learning;

[0114] Integrate the evaluation system of the highway tunnel entrance sight guidance system based on redundancy, integrity, and constancy to obtain the evaluation system of the highway tunnel entrance sight guidance system based on deep learning. The specific calculation formula is as follows:

[0115] S = a1S1 + a2S2 + a3S3

[0116] Among them, S is the total evaluation score of the highway tunnel entrance sight guidance system based on deep learning, with a value range of 0 - 100; S1, S2, and S3 are the evaluation scores based on redundancy, integrity, and constancy respectively; a1, a2, and a3 are the weight coefficients of the evaluation based on redundancy, integrity, and constancy respectively, and a1 + a2 + a3 = 1;

[0117] The total score S is divided into four induction effects according to the score size, from high to low are high - level induction (85 - 100), medium - level induction (70 - 85), basic induction (60 - 70), and unqualified induction (0 - 60).

[0118] S2. Collect pictures of the highway tunnel entrance sight guidance system and process the pictures to obtain a data set;

[0119] The sources of the collected pictures of the highway tunnel entrance sight guidance system include real - vehicle shooting and video capture from a driving recorder;

[0120] The picture processing steps include cleaning and screening the pictures, standardizing the pictures, annotating the screened pictures, performing data augmentation on the annotated pictures, and dividing the pictures into a data set;

[0121] Cleaning and screening include deleting pictures with unqualified shooting quality from the collected pictures and screening out positive tunnel entrance pictures that meet the requirements;

[0122] The standardization step includes unifying the picture size and uniformly outputting the pictures in png format;

[0123] The data augmentation operations include flipping, rotating, scaling, adding noise, blurring, adjusting brightness or contrast, and affine transformation;

[0124] The specific operation of partitioning includes partitioning the training set, validation set, and test set in a ratio of 6:2:2.

[0125] S3. Construct a deep learning model for identifying the sight guidance facilities at the entrance of highway tunnels;

[0126] Construct different deep learning models for identifying the sight guidance facilities at the entrance of highway tunnels, and adjust the deep learning model using a multi-scale object recognition strategy;

[0127] The deep learning models include YOLOV7, EfficientDet, and Faster R-CNN;

[0128] The multi-scale object recognition strategy includes adding an attention mechanism to the model and introducing a feature pyramid network.

[0129] S4. Use the dataset to train and validate different deep learning models, adjust the parameters of different models, and obtain an optimal model;

[0130] Define a loss function, select an optimizer, set hyperparameters, and judge the fitting situation of the deep learning model according to the training loss total_loss and validation loss val_loss during the training process;

[0131] Adjust the model structure or hyperparameters according to the performance of the deep learning model on the validation set, conduct repeated training, and finally select the optimal model according to the comprehensive performance of Precision, Recall, Accuracy, and mean average precision mAP of the model on the test set.

[0132] S5. Embed a dual-mode mechanism, a rating mechanism, and a feedback mechanism into the evaluation system to achieve fine evaluation of a single tunnel and rapid review of multiple tunnels.

[0133] Embed a dual-mode mechanism, a rating mechanism, and a feedback mechanism into the evaluation system;

[0134] The dual-mode mechanism includes single-tunnel evaluation and multi-tunnel evaluation; single-tunnel evaluation is used for fine evaluation of a single tunnel and outputs the evaluation results of various indicators for each segment of the single tunnel; multi-tunnel evaluation is used for rapid review of multiple tunnels to identify tunnels with relatively large potential safety hazards in the input tunnels, that is, unqualified induced tunnels.

[0135] The rating mechanism is set according to the evaluation system of the highway tunnel entrance sight guidance system based on deep learning, and controls the deep learning model to output the evaluation results and scores of various indicators for single-tunnel evaluation and multi-tunnel evaluation;

[0136] The feedback mechanism collects the list of tunnels with unqualified induction effects according to the output rating results for multi-tunnel evaluation.

[0137] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims. All of these are within the protection scope of the present invention.

Claims

1. An evaluation method for a sight guidance system at the entrance of a highway tunnel based on deep learning, characterized in that It includes the following steps: S1. Construct an evaluation system for the sight guidance system at the outer entrance section of highway tunnels based on deep learning; S2. Collect pictures of the sight guidance system at the highway tunnel entrance, and process the pictures to obtain a dataset; S3. Construct a deep learning model for identifying sight guidance facilities at the highway tunnel entrance; S4. Use the dataset to train and validate different deep learning models, adjust the parameters of different models, and obtain an optimal model; S5. Embed a dual-mode mechanism, a rating mechanism, and a feedback mechanism into the evaluation system to achieve fine evaluation of a single tunnel and rapid review of multiple tunnels.

2. The evaluation method of the sight guidance system at the entrance of a highway tunnel based on deep learning according to claim 1, wherein, In step S1, constructing an evaluation system for the sight guidance system at the outer entrance section of highway tunnels based on deep learning includes the following steps: S101. Segment the tunnel entrance area; S102. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on redundancy; S103. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on integrity; S104. In each segment of the tunnel entrance area, construct an evaluation system for the sight guidance system at the highway tunnel entrance based on constancy; S105. Integrate the evaluation systems for the sight guidance system at the highway tunnel entrance based on redundancy, integrity, and constancy to obtain an evaluation system for the sight guidance system at the highway tunnel entrance based on deep learning.

3. The evaluation method of the line-of-sight induction system at the entrance of a highway tunnel based on deep learning according to claim 2, wherein In step S101, the tunnel entrance area is segmented by the following method: According to different types of sight guidance facilities, the highway tunnel entrance area includes a sensing and recognition section, a decision-making and execution section, and a relaxation and adaptation section: The sensing and recognition section is located outside the tunnel, 150 - 300 meters away from the tunnel entrance. The sight guidance facilities in the sensing and recognition section play a role in sensing and recognition, warning the driver that there is a tunnel ahead; The decision-making and execution section is located outside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in the decision-making and execution section mainly play a role in induction and protection, guiding the driver to drive into the tunnel safely; The relaxation and adaptation section is located inside the tunnel, 0 - 150 meters away from the tunnel entrance. The sight guidance facilities in the relaxation and adaptation section mainly play a role in induction and relaxation, helping the driver gradually adapt to the changes in light and space inside the tunnel, and ensuring driving safety and comfort.

4. The evaluation method of the sight guidance system at the entrance of a highway tunnel based on deep learning according to claim 3, wherein In step S102, The evaluation system for the sight guidance system at the highway tunnel entrance based on redundancy evaluates from the hierarchy and completeness of the sight guidance facilities. The sight guidance facilities include low-level, middle-level, and high-level according to the induction level; In the sensing and recognition section, the low-level facilities are raised road markers and visual deceleration markings, the middle-level facilities are outline markers and elastic traffic columns, and the high-level facilities are warning linear induction markers and tunnel entrance warning signs; In the decision-making and execution section, the low-level facilities are raised road markers and vibration-type deceleration markings, the middle-level facilities are outline markers, elastic traffic columns, crash barrels, crash pads, and portal elevation markings, and the high-level facilities are tunnel information signs, warning linear induction markers, and portal elevation markings; In the mitigation and adaptation section, the low-level facilities are raised pavement markers, profile markers for maintenance roads, and roadside elevation markers; the medium-level facilities are flexible traffic posts, sidewall profile markers, reflective strips, and waistlines; the high-level facilities are reflective rings; Based on the types of low-level, medium-level, and high-level facilities in the perception and identification section, the decision execution section, and the mitigation and adaptation section, evaluate the redundancy. The calculation formula is as follows: Among them, i represents different segmented areas of the tunnel entrance. i = 1 is the sensing and recognition section, i = 2 is the decision-making and execution section, and i = 3 is the mitigation and adaptation section; R i is the redundancy evaluation score for section i; j represents different induction levels of the sight guidance facilities. j = 1 is the low level, j = 2 is the middle level, and j = 3 is the high level; α j is the weight coefficient of the induction facilities at level j, and α1 + α2 + α3 = 1. According to the importance of the induction facilities at different levels, the low-level facilities are set to 0.4, the middle-level facilities are set to 0.4, and the high-level facilities are set to 0.2; N ij is the total number of types of the induction facilities at level j in section i; n ij is the number of types of the induction facilities at level j in section i that are actually present and recognized; S1 is the total score of the highway tunnel entrance area based on the redundancy evaluation.

5. The evaluation method of the sight guidance system at the entrance of a highway tunnel based on deep learning according to claim 4, characterized in that, In step S103, The evaluation based on integrity evaluates from the overall combined performance and effect of the sight guidance facilities. The guidance effect of the sight guidance system in the entrance area of the highway tunnel includes high-level guidance, medium-level guidance, basic guidance, and unqualified guidance; In the perception and identification section, the high-level guidance includes raised pavement markers, visual deceleration markings, flexible traffic posts, warning linear guide signs, profile markers, and tunnel entrance warning signs; the medium-level guidance includes raised pavement markers, flexible traffic posts, visual deceleration markings, profile markers, and tunnel entrance warning signs; the basic guidance includes raised pavement markers, profile markers, and tunnel entrance warning signs; the unqualified guidance is a facility combination that does not meet the basic guidance; In the decision execution section: The high-level guidance is composed of raised pavement markers, visual deceleration markings, crash barrels, crash pads, flexible traffic posts, portal elevation markers, profile markers, warning linear guide signs, and tunnel information signs; the medium-level guidance includes raised pavement markers, visual deceleration markings, crash barrels, portal elevation markers, profile markers, warning linear guide signs, and tunnel information signs; the basic guidance consists of raised pavement markers, portal elevation markers, and profile markers; the unqualified guidance is a facility combination that does not meet the basic guidance; In the decision execution section, the high-level guidance includes raised pavement markers, roadside elevation markers, flexible traffic posts, profile markers for maintenance roads, sidewall profile markers, waistlines, reflective strips, and reflective rings; the medium-level guidance includes raised pavement markers, profile markers for maintenance roads, sidewall profile markers, waistlines, and reflective strips; the basic guidance consists of raised pavement markers, profile markers for maintenance roads, and sidewall profile markers; the unqualified guidance is a facility combination that does not meet the basic guidance; Based on the types of high-level guidance, medium-level guidance, basic guidance, and unqualified guidance facilities in the perception and identification section, the decision execution section, and the decision execution section, evaluate the integrity. The calculation formula is as follows: Among them, S2 is the total score of the highway tunnel entrance area based on the overall evaluation; i is different segmented tunnel entrance areas, where i = 1 is the perception and recognition section, i = 2 is the decision-making and execution section, and i = 3 is the mitigation and adaptation section; Z i is the score of the i-th section based on the overall evaluation. The score for high-level induction is 90 points, the score for medium-level induction is 80 points, the score for basic induction is 60 points, and the score for unqualified induction is 30 points.

6. The evaluation method of the sight guidance system at the entrance of a highway tunnel based on deep learning according to claim 5, characterized in that In step S104, The evaluation based on constancy evaluates from the stability and visibility of the sight guidance facilities, and realizes the quantitative evaluation of the brightness stability and facility visibility of some sight guidance facilities; The brightness stability is evaluated by the color differences caused by different distances from and illuminations of the same type of facilities in the same picture; The facility visibility is evaluated by the contrast difference between the facility and its surrounding environment; In the perception and identification section, the objects of constancy evaluation are flexible traffic posts and warning linear guide signs; in the decision execution section, the objects of constancy evaluation are flexible traffic posts, warning linear guide signs, and crash barrels; in the mitigation and adaptation section, the objects of constancy evaluation are flexible traffic posts and reflective strips; Evaluate the brightness stability and facility visibility of the object according to the sensing and recognition section, the sensing and recognition section, and the mitigation and adaptation section, and evaluate the constancy. The calculation formula is as follows: Among them, S3 is the total score based on the constancy evaluation of the highway tunnel entrance area; i is different sections of the tunnel entrance area, i = 1 is the perception and recognition section, i = 2 is the decision-making and execution section, and i = 3 is the mitigation and adaptation section; W i is the evaluation score of the brightness stability of section i; K i is the evaluation score of the visibility of the facilities in section i; β1 and β2 are the weight coefficients of brightness stability and facility visibility, and β1 + β2 = 1.

7. The evaluation method for the line-of-sight induction system at the entrance of a highway tunnel based on deep learning according to claim 6, wherein In the step S105, Based on the evaluation system of the highway tunnel entrance sight guidance system based on redundancy, integrity, and constancy, an evaluation system of the highway tunnel entrance sight guidance system based on deep learning is obtained. The specific calculation formula is as follows: S = a1S1 + a2S2 + a3S3 Where S is the total evaluation score of the highway tunnel entrance sight guidance system based on deep learning, and the value range is 0 - 100; S1, S2, and S3 are the evaluation scores based on redundancy, integrity, and constancy respectively; a1, a2, and a3 are the weight coefficients of the evaluation based on redundancy, integrity, and constancy respectively, and a1 + a2 + a3 = 1; According to the total score S, the induction effects are divided into four levels, from high to low: high-level induction (85 - 100), medium-level induction (70 - 85), basic induction (60 - 70), and unqualified induction (0 - 60).

8. The evaluation method of the line-of-sight induction system at the entrance of a highway tunnel based on deep learning according to claim 1, characterized in that In the step S2, the sources of the pictures of the highway tunnel entrance sight guidance system collected include real vehicle shooting and video capture from a driving recorder. The processing of the pictures of the highway tunnel entrance sight guidance system includes the following steps: Clean and screen the pictures, standardize the pictures, label the screened pictures, perform data augmentation on the labeled pictures, and divide the pictures into data sets; The cleaning and screening include deleting pictures with unqualified shooting quality from the collected pictures and screening out positive tunnel entrance pictures that meet the requirements; The standardization step includes unifying the picture size and uniformly outputting the pictures in png format; The data augmentation operations include flipping, rotating, scaling, adding noise, blurring, adjusting brightness or contrast, and affine transformation; The specific operation of the division includes dividing the training set, validation set, and test set in a ratio of 6:2:2; In the step S3, different deep learning models for identifying highway tunnel entrance sight guidance facilities are constructed, and the deep learning models are adjusted using a multi-scale object recognition strategy; The deep learning models include YOLOV7, EfficientDet, and Faster R-CNN; The multi-scale object recognition strategy includes adding an attention mechanism to the model and introducing a feature pyramid network.

9. The evaluation method of the line-of-sight induction system at the entrance of a highway tunnel based on deep learning according to claim 1, characterized in that, In the step S4, define the loss function, select the optimizer, set the hyperparameters, and judge the fitting situation of the deep learning model according to the training loss total_loss and the validation loss val_loss during the training process; Adjust the model structure or hyperparameters according to the performance of the deep learning model on the validation set, perform repeated training, and finally select the optimal model according to the comprehensive performance of the precision Precision, recall Recall, accuracy Accuracy, and mean average precision mAP of the deep learning model in the test set.

10. The evaluation method of the sight guidance system at the entrance of a highway tunnel based on deep learning according to claim 1, characterized in that In the step S5, the dual-mode mechanism, rating mechanism, and feedback mechanism embedded in the evaluation system are specifically as follows: The dual-mode mechanism is single-tunnel evaluation and multi-tunnel evaluation; single-tunnel evaluation is used for fine evaluation of a single tunnel, and outputs the evaluation results of various indicators for each segment of the single tunnel; multi-tunnel evaluation is used for rapid review of multiple tunnels to identify tunnels with relatively large potential safety hazards in the input tunnels, that is, unqualified induced tunnels; The rating mechanism is set according to the evaluation system of the highway tunnel entrance sight induction system based on deep learning, and controls the deep learning model to output evaluation results and scores of various indicators, which are used for single-tunnel evaluation and multi-tunnel evaluation; The feedback mechanism collects the list of tunnels with unqualified induction effects according to the output rating results, which is used for multi-tunnel evaluation.

Images