Urban road riding friendliness evaluation method based on multi-modal data fusion

Abstract

The invention relates to an urban road riding friendliness evaluation method based on multi-modal data fusion, and the method comprises the following steps: carrying out the classification according to the use conditions of different roads, collecting subjective data through revealing preference investigation, building a subjective evaluation model through an entropy method and a gradient boosting decision tree importance analysis method, and carrying out the calculation of the subjective evaluation model. Determining the influence degree of each element on different types of road riding friendliness; a bicycle serves as a detection carrier, and vibration signals and image video data in the advancing process of the bicycle are collected; estimating the road surface flatness based on the vibration signal, identifying a non-motor vehicle lane separation condition, a non-motor vehicle lane occupation condition, a riding-friendly sign and a roadside landscape based on image video data, and determining various road riding-friendly evaluation indexes; and after normalization processing is performed on each evaluation index, weighted summation is performed on the influence degree of each element determined by the subjective evaluation model, and a road riding friendliness evaluation result is obtained. Compared with the prior art, the method has the advantages of objective evaluation result, high reliability and the like.

Description

Technical Field

[0001] The present invention relates to the field of performance detection and evaluation of slow traffic road infrastructure, and in particular to a method for evaluating the cycling friendliness of urban roads by multi-modal data fusion. Background Art

[0002] With the acceleration of urbanization, traffic congestion and environmental pollution are becoming increasingly serious. The promotion of slow traffic has become a key link in achieving low-carbon development. As the core of sustainable transportation, slow traffic has received attention from the government and society. Increasing the proportion of non-motorized vehicles and walking trips and building a safe and comfortable slow traffic system have become the future development trend. However, there are still many shortcomings in the existing slow traffic roads, which seriously affect the cycling experience and need to be improved and improved. At present, there is a lack of a set of scientific and effective evaluation methods, which makes it difficult to accurately find out the advantages and disadvantages of these roads and facilities, restricting the further development of slow traffic. Only by improving the level of slow traffic roads can we promote people to choose slow traffic as a travel mode. Therefore, it is the key to accurately locate problems and promote subsequent improvement work to understand residents' expectations for cycling-friendly roads and quickly and accurately evaluate slow traffic roads. It is also the core demand for promoting the further development of slow traffic and promoting the green and low-carbon transformation of transportation.

[0003] In order to evaluate the convenience of cycling, people proposed the concept of cycling friendliness based on the concept of walking convenience. Subsequently, many studies have developed an index system as a quantitative tool. However, the index collection of existing methods mainly relies on the feelings of experts, and the results are highly unstable. They cannot reflect the expectations of residents who actually use the roads for cycling-friendly roads. Moreover, most current evaluations use a unified standard and do not consider the impact of road types on evaluation indicators. In addition, the current evaluation has a low level of digitization and low efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a method for evaluating the cycling friendliness of urban roads based on multimodal data fusion, aiming to evaluate from both subjective and objective perspectives, combining questionnaire surveys with signal and computer vision analysis, and using artificial intelligence technology to conduct a multi-factor analysis of the friendliness of municipal road landscape facilities to obtain reliable evaluation results.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for evaluating the cycling friendliness of urban roads based on multimodal data fusion includes the following steps:

[0007] S1, Establish a subjective evaluation model using revealed preference survey: Classify by different road usage conditions, collect subjective data through revealed preference survey, and establish a subjective evaluation model using entropy method and gradient boosting decision tree importance analysis method to determine the influence of each factor on the cycling friendliness of different types of roads;

[0008] S2, objective data collection and processing: using bicycles as detection carriers, collecting vibration signals and image and video data during the bicycle's movement; estimating the road surface flatness based on vibration signals, identifying the separation of non-motorized vehicle lanes, non-motorized vehicle lane occupancy, cycling-friendly signs and roadside landscapes based on image and video data, and determining various road cycling-friendly evaluation indicators;

[0009] S3, automated evaluation: after normalizing the various evaluation indicators determined in step S2, a weighted sum is taken with the influence degree of each factor determined by the subjective evaluation model to obtain a road cycling friendliness evaluation result.

[0010] The classification according to different road usage conditions is specifically as follows: roads are divided into three categories: commuting roads, living roads and leisure roads.

[0011] The S1 comprises the following steps:

[0012] S11, pre-processing the collected subjective data to eliminate invalid and redundant data;

[0013] S12, using the gradient boosting decision tree importance analysis method, the factors are screened according to the importance of each factor, and the factors with important influence are selected as independent variables;

[0014] S13, using the entropy method to calculate the weight of each factor, establishing a subjective evaluation model for road cycling friendliness, and obtaining the degree of influence of each factor on the overall score of the road for different types of roads.

[0015] The road cycling-friendly evaluation index includes: road surface smoothness, separation form of the motor-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, cycling-friendly sign density, bicycle road side landscape, air quality and environmental noise, wherein the road surface smoothness is obtained by performing signal analysis on vibration signals, the separation form of the motor-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, and cycling-friendly sign density are obtained by processing image and video data based on the YOLO model, the bicycle road side landscape is obtained by processing image and video data based on the CV2 model, and the air quality and environmental noise are obtained by collecting sensors.

[0016] The method for obtaining the road surface flatness is:

[0017] Based on the Z-axis vertical vibration signal collected by the bicycle seat, wavelet decomposition is used to extract the key frequency band characteristics of the vibration. The outlier screening method is used in the key frequency band characteristics to determine the number of abnormal bumps. Based on the number of abnormal bumps and the road length, the road surface flatness is obtained.

[0018] The method for obtaining the separation form of the non-vehicle isolation belt is as follows: based on the collected image and video data, a deep learning convolutional neural network is used to perform target recognition, and the separation forms of the isolation belt are automatically identified and classified. The separation forms of the isolation belt include green belt isolation, guardrail isolation, line isolation and no isolation. Different values ​​are assigned to different separation methods as evaluation index values ​​of the separation form of the non-vehicle isolation belt.

[0019] The method for obtaining the occupancy status of the non-motorized vehicle lane is as follows: based on the collected image and video data, a deep learning convolutional neural network is used to perform target recognition, and the occupied area of ​​other objects in the bicycle lane is automatically counted, wherein the other objects include motor vehicles, motorcycles, pedestrians, and garbage, and the ratio of the occupied area to the total lane area is calculated as an evaluation indicator of the occupancy status of the non-motorized vehicle lane.

[0020] The method for obtaining the density of cycling-friendly signs is as follows: based on the collected image and video data, a deep learning convolutional neural network is used for target recognition, bicycle lane sign information is automatically recognized, and the number of friendly signs and the proportion of markings are calculated as evaluation indicators of the density of cycling-friendly signs.

[0021] The method for acquiring the bicycle road side landscape is as follows: based on the collected image video data, a CV2 algorithm is used to set an RGB threshold to segment the green belt, and the proportion of the green belt area to the total image area is output as an evaluation index of the bicycle road side landscape.

[0022] The method also includes: S4, output and visualization of evaluation results: integrating sensors to realize data collection and using a single-chip microcomputer to record data, combining global positioning system positioning with map public interface location information matching, and directly recording the evaluation results in the corresponding map layer for visualization, thereby realizing the integration of data collection, data processing and evaluation result output.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] (1) The evaluation results are objective and accurate:

[0025] The present invention integrates subjective and objective data, subjectively builds a model through scientific investigation and analysis, and objectively uses a variety of technologies to collect multi-dimensional indicators, thereby reducing subjective interference and achieving objective and accurate results.

[0026] (2) Consider road type factors:

[0027] The present invention classifies roads and designs surveys and models them separately. Objective indicators also reflect type differences and accurately evaluate the cycling friendliness of different roads.

[0028] (3) Improving digitalization and efficiency:

[0029] The present invention uses a variety of devices to automatically collect and process data, combines maps to achieve location matching, and integrates the process to improve the level of digitization and efficiency. The visualization of results facilitates evaluation and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the method of the present invention;

[0031] Figure 2 This is a schematic diagram of the influence of each factor obtained by the subjective evaluation model;

[0032] Figure 3 This is an algorithm architecture diagram for road information collection and evaluation based on machine learning technology;

[0033] Figure 4 Flowchart for automated evaluation and visual display of friendliness;

[0034] Figure 5 This is a schematic diagram of the YOLO model recognition results;

[0035] Figure 6 Schematic diagram of image semantic segmentation results. DETAILED DESCRIPTION

[0036] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0037] This embodiment provides a method for evaluating the cycling friendliness of urban roads by multimodal data fusion, which combines subjective survey data with objective digital measurements, providing a fast and accurate method for evaluating a large number of bicycle lanes while meeting the expectations of residents. Roads are classified according to different road usage conditions, and a tailored subjective evaluation model is established through revealing preference survey analysis. Advanced technologies such as vibration signal processing, computer vision and machine learning are used to quantitatively analyze the factors affecting cycling friendliness, such as road flatness, non-motor vehicle occupancy, bicycle-friendly facilities and road test environment, and the overall evaluation results are obtained through normalization processing. Finally, the sensor data and global positioning system location information are stored by the test vehicle on-site riding, and the three-dimensional data of position-vibration-image is generated by the background integration. The road attributes of the map layer are combined with the global positioning system location information to achieve road matching, and the evaluation results are stored and displayed in the geographic information system layer. The integrated process of data collection, data processing and evaluation result recording is realized, which provides feasible insights for determining the priority of infrastructure upgrades, formulating transportation policies and promoting sustainable urban transportation.

[0038] Specifically, Figure 1 As shown, the method comprises the following steps:

[0039] S1. Establish a subjective evaluation model using revealed preference surveys: Classify by different road usage conditions, collect subjective data through revealed preference surveys, and establish a subjective evaluation model using the entropy method and gradient boosting decision tree importance analysis method to determine the degree of influence of each factor on the cycling friendliness of different types of roads.

[0040] Considering the changes in road types and functions, the roads were classified and questionnaires were designed and surveyed. After the questionnaire results were screened, a subjective evaluation model was established using the entropy method and gradient boosting decision tree importance analysis method. The model reflects the degree of influence of each factor score on different types of roads.

[0041] In this embodiment, roads are divided into three categories: commuting roads, life roads, and leisure roads. Representative sections are selected to design questionnaires. The questionnaire asks respondents to first rate the road as a whole, and then select influencing factors. The 5-level Likert scale is used to measure the impact of factors, including no impact, small impact, medium impact, large impact, and very large impact. Finally, each factor is evaluated, with a logical progression.

[0042] The questionnaire was distributed in a combination of online and offline methods, and screening questions were set to ensure sample diversity so that the model results could reflect residents' expectations for bicycle-friendly roads.

[0043] In order to improve the credibility of the questionnaire, the questionnaire results with obvious logical errors were screened out based on the trap questions and the logical comparison of the questions (for example, the evaluation of each item must be consistent with the overall evaluation results, etc.). Cronbach's coefficient reliability test, KMO measurement and Bartlett's sphericity test were further used to ensure the validity of the questionnaire results. Therefore, the questionnaire was divided into three parts. The first part collected basic information, such as the frequency and purpose of cycling, to help determine a representative sample of cyclists. The second part gave an overall evaluation of the road based on the cyclists' scores, and the third part evaluated the degree of influence of various factors. Cyclists selected and evaluated the key factors of bicycle friendliness from multiple dimensions. The trap question at the end asked about the administrative area of ​​the evaluated road, thereby screening out cyclists who did not meet the survey criteria or did not answer.

[0044] Specifically, S1 includes the following steps:

[0045] S11, pre-process the collected subjective data to eliminate invalid data and redundant data.

[0046] The specific process is as follows:

[0047] (1) Regression analysis was performed to establish a linear equation and transform the data to reflect the difference between the questionnaire and the model-generated scores. Answers with an absolute deviation of more than 25 were deleted, and finally a valid questionnaire set was obtained.

[0048] (2) The reliability of the questionnaire was measured by Cronbach's coefficient.

[0049] (3) KMO and Bartlett's sphericity test were used to assess whether the data were suitable for factor analysis.

[0050] (4) Evaluate data distribution and verify data validity.

[0051] (5) Calculate the correlation of each factor and eliminate redundancy.

[0052] S12, using the gradient boosting decision tree importance analysis method, screen the factors according to the degree of importance of each factor, delete the factors with importance values ​​less than 0.2, and consider that they have little impact on the overall road score results, and select the factors with important influence as independent variables.

[0053] S13, using the entropy method to calculate the weight of each factor, establishing a subjective evaluation model for road cycling friendliness, and obtaining the degree of influence of each factor on the overall score of the road for different types of roads.

[0054] For the three types of roads, 15 typical sections were selected, and a cycling-friendliness questionnaire survey was conducted through a combination of online and offline methods. The gradient boosting decision tree importance screening and entropy method were used to establish an evaluation model to reflect the influence of each factor on the overall score. The modeling results are as follows: Figure 2As shown in the figure, it can be seen that for different types of roads, the influence of various elements is different, which proves the rationality and correctness of the classification modeling in this method.

[0055] S2, objective data collection and processing: using bicycles as detection carriers, collect vibration signals and image and video data of bicycles in motion; estimate the road surface flatness based on vibration signals, identify the separation of non-isolated belts, non-motorized vehicle lane occupancy, cycling-friendly signs and roadside landscapes based on image and video data, and determine various road cycling-friendly evaluation indicators. The process is as follows: Figure 3 shown.

[0056] In this embodiment, the evaluation indexes of road cycling friendliness include: road surface smoothness, separation form of vehicle-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, cycling-friendly sign density, bicycle road side landscape, air quality and environmental noise, etc., among which, the road surface smoothness is obtained by performing signal analysis on vibration signals, the separation form of vehicle-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, and cycling-friendly sign density are obtained by processing image and video data based on the YOLO model, the bicycle road side landscape is obtained by processing image and video data based on the CV2 model, and the air quality and environmental noise are obtained by collecting sensors.

[0057] (1) Road surface smoothness

[0058] Based on the Z-axis vertical vibration signal collected from the bicycle seat, the key frequency band features of the vibration are extracted using wavelet decomposition based on the Symlet wavelet basis. In this embodiment, the key frequency band features are collected by filtering the high-frequency signal, because the high-frequency signal mainly represents noise and bumps. According to the discomfort interval standard of the human body exposed to a vibration environment given by the international standard ISO-2631, the abnormal value screening method is used in the key frequency band features to determine the number of abnormal bumps. For inapplicable conditions, the p-value method is used for processing.

[0059] Based on the number of abnormal bumps and the length of the road, the road surface roughness is obtained:

[0060]

[0061] Among them, s1 is the road surface flatness, β is the coefficient, n1 is the number of abnormal bumps, and L is the length of the road section. The length of the road section is obtained through the open street map road network data.

[0062] For the vibration signal information, the number of abnormal bumps was obtained as 24 times through filtering processing, wavelet decomposition, etc., and the road surface smoothness was calculated to be 0.79.

[0063] (2) Separation form of machine and non-machine isolation zone

[0064] Different types of non-motorized vehicle separation can better reflect the independence and integrity of non-motorized vehicle lanes. Based on the collected image and video data, this embodiment uses a deep learning convolutional neural network (self-trained YOLO-OBB model) to perform target recognition, automatically identify and classify the separation forms of isolation belts. In this embodiment, the separation forms of isolation belts include green belt isolation, guardrail isolation, line isolation and no isolation. Different values ​​are assigned to different separation methods as evaluation index values ​​for the separation forms of non-motorized vehicle isolation belts. The specific assignments are shown in Table 1.

[0065] Table 1

[0066] evaluate high middle Low Difference Classification Green belt isolation Fence isolation Isolation by line No isolation Assignment 3 2 1 0

[0067] The calculation formula is:

[0068]

[0069]

[0070] Among them, s2 is the evaluation index value of the separation form of the non-separation zone, k is the corresponding value in Table 1, L k is the length of the corresponding road section, x max and x min They correspond to the upper and lower limits of the normalized operation respectively.

[0071] (3) Occupancy of non-motorized vehicle lanes

[0072] Illegal occupation of bicycle lanes by vehicles destroys the continuity of bicycle lanes and reduces the riding comfort of the road. This embodiment uses a deep learning convolutional neural network (pre-trained YOLO model) to perform target recognition based on the collected image and video data, automatically counts the occupied area of ​​other objects in the bicycle lane, including motor vehicles, motorcycles, pedestrians, garbage, etc., and calculates the ratio of the occupied area to the total lane area as an evaluation indicator for the occupation of non-motorized vehicle lanes.

[0073] Since the occupancy is a negative indicator, it needs to be converted to the maximum value of the above evaluation. Considering that the vehicle length is about 4.5m, the maximum value y is set max =1000 / 4.5=200 vehicles / km, that is, the road is completely occupied by illegally parked vehicles, y min To evaluate the minimum value, assume it is 0, that is, the road is not occupied by illegally parked vehicles. The calculation formula is:

[0074]

[0075]

[0076] Among them, s3 is the evaluation index value of the non-motorized vehicle lane occupancy, n2 is the number of other objects identified, and L is the length of the road section.

[0077] (4) Density of Cycling-Friendly Signs

[0078] When there are clear and appropriate bicycle-friendly signs on the bicycle lane, it helps to enhance the sense of security of cyclists and helps to improve the overall experience. This embodiment uses a deep learning convolutional neural network (pre-trained YOLO model) to perform target recognition based on the collected image and video data, automatically recognize bicycle lane sign information, and calculate the number of friendly signs and the proportion of markings as evaluation indicators of cycling friendly sign density. The calculation formula is:

[0079] s4=n3 / L

[0080] Among them, s4 is the evaluation index value of the density of cycling-friendly signs, n3 is the number of recognized friendly signs, and L is the length of the road section.

[0081] (5) Bicycle path side landscape

[0082] Based on the collected image and video data, the CV2 algorithm is used to set the RGB threshold for green belt segmentation, and the output green belt area accounts for the proportion of the total image area as the evaluation index of the bicycle road side landscape. The calculation formula is:

[0083] s5=S green / S total

[0084] Among them, s5 is the evaluation index value of the bicycle road side landscape, S green is the area of ​​the green belt separated, S total is the total area of ​​the image.

[0085] The computer vision library is used to sample the video at a certain frequency (0.5s / frame), extract the image and output it, and use the YOLO model to identify the machine-non-motorized vehicle separation mode, non-motorized vehicle lane occupation, and cycling-friendly signs, such as Figure 5 As shown, the bicycle road side landscape is identified as Figure 6 Substitute the above corresponding formula to calculate the original score of each factor.

[0086] S3, automated evaluation: after normalizing the various evaluation indicators determined in step S2, a weighted sum is taken with the influence degree of each factor determined by the subjective evaluation model to obtain a road cycling friendliness evaluation result.

[0087] Based on the preliminary survey data of nearly 100 roads, the scores of each factor were obtained, and the upper and lower limits were obtained by normal distribution fitting. The results are shown in Table 2 below.

[0088] Table 2

[0089]

[0090] The data was normalized to obtain the final score of each factor, and the overall cycling friendliness evaluation index was obtained based on the evaluation model. The results are shown in Table 3 below.

[0091] Table 3

[0092]

[0093]

[0094] S4, output and visualization of evaluation results: Integrate sensors to realize data collection and use single-chip microcomputer to record data, combine global positioning system positioning with map public interface location information matching, and directly record the evaluation results in the corresponding map layer for visualization, so as to realize the integration of data collection, data processing and evaluation result output.

[0095] Specifically, Figure 4 As shown, the following steps are included:

[0096] S41, install a dual-axis vibration sensor, a video camera probe, Bluetooth and a single-chip microcomputer on the test bicycle, ride the bicycle on the road to be tested, the sensor collects road vibration signals at a fixed sampling rate of 1kHz, converts the data into digital signals through an analog-to-digital converter, and stores them in the single-chip microcomputer memory. The video acquisition device communicates with the single-chip microcomputer, and performs real-time video recording while collecting vibration data. The data is stored in an external storage card or uploaded to the background.

[0097] S42, install a Bluetooth module (such as HC-05), call the map application programming interface data to achieve automatic matching of the position (latitude and longitude, road name), and locate the road where the current test vehicle is located through the road number (ID) attribute in the vector map.

[0098] S43, the background automatically processes the data stored in the single-chip microcomputer, processes it according to the above methods S1-S3, obtains the final score, matches the score to the corresponding road, and stores the evaluation result data according to the image data index and displays it visually, realizing the integrated process of data collection, data processing, and evaluation result recording and display.

[0099] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for evaluating the cycling friendliness of urban roads based on multimodal data fusion, characterized in that: The following steps are involved: S1, Establish a subjective evaluation model using revealed preference survey: Classify by different road usage conditions, collect subjective data through revealed preference survey, and establish a subjective evaluation model using entropy method and gradient boosting decision tree importance analysis method to determine the influence of each factor on the cycling friendliness of different types of roads; S2, objective data collection and processing: using bicycles as detection carriers, collecting vibration signals and image and video data during the bicycle's movement; estimating the road surface flatness based on vibration signals, identifying the separation of non-motorized vehicle lanes, non-motorized vehicle lane occupancy, cycling-friendly signs and roadside landscapes based on image and video data, and determining various road cycling-friendly evaluation indicators; S3, automated evaluation: after normalizing the various evaluation indicators determined in step S2, a weighted sum is taken with the influence degree of each factor determined by the subjective evaluation model to obtain a road cycling friendliness evaluation result.

2. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 1 is characterized in that: The classification according to different road usage conditions is specifically as follows: roads are divided into three categories: commuting roads, living roads and leisure roads.

3. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 1, characterized in that: The S1 comprises the following steps: S11, pre-processing the collected subjective data to eliminate invalid and redundant data; S12, using the gradient boosting decision tree importance analysis method, the factors are screened according to the importance of each factor, and the factors with important influence are selected as independent variables; S13, using the entropy method to calculate the weight of each factor, establishing a subjective evaluation model for road cycling friendliness, and obtaining the degree of influence of each factor on the overall score of the road for different types of roads.

4. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 1, characterized in that: The road cycling-friendly evaluation index includes: road surface smoothness, separation form of the motor-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, cycling-friendly sign density, bicycle road side landscape, air quality and environmental noise, wherein the road surface smoothness is obtained by performing signal analysis on vibration signals, the separation form of the motor-non-motorized vehicle isolation zone, bicycle lane width, non-motorized vehicle lane occupancy, and cycling-friendly sign density are obtained by processing image and video data based on the YOLO model, the bicycle road side landscape is obtained by processing image and video data based on the CV2 model, and the air quality and environmental noise are obtained by collecting sensors.

5. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 4 is characterized in that: The method for obtaining the road surface flatness is: Based on the Z-axis vertical vibration signal collected by the bicycle seat, wavelet decomposition is used to extract the key frequency band characteristics of the vibration. The outlier screening method is used in the key frequency band characteristics to determine the number of abnormal bumps. Based on the number of abnormal bumps and the road length, the road surface flatness is obtained.

6. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 4 is characterized in that: The method for obtaining the separation form of the non-vehicle isolation belt is as follows: based on the collected image and video data, a deep learning convolutional neural network is used to perform target recognition, and the separation forms of the isolation belt are automatically identified and classified. The separation forms of the isolation belt include green belt isolation, guardrail isolation, line isolation and no isolation. Different values ​​are assigned to different separation methods as evaluation index values ​​of the separation form of the non-vehicle isolation belt.

7. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 4 is characterized in that: The method for obtaining the occupancy status of the non-motorized vehicle lane is as follows: based on the collected image and video data, a deep learning convolutional neural network is used to perform target recognition, and the occupied area of ​​other objects in the bicycle lane is automatically counted, wherein the other objects include motor vehicles, motorcycles, pedestrians, and garbage, and the ratio of the occupied area to the total lane area is calculated as an evaluation indicator of the occupancy status of the non-motorized vehicle lane.

8. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 4 is characterized in that: The method for obtaining the density of cycling-friendly signs is as follows: based on the collected image and video data, a deep learning convolutional neural network is used for target recognition, bicycle lane sign information is automatically recognized, and the number of friendly signs and the proportion of markings are calculated as evaluation indicators of the density of cycling-friendly signs.

9. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 4, characterized in that: The method for acquiring the bicycle road side landscape is as follows: based on the collected image video data, a CV2 algorithm is used to set an RGB threshold to segment the green belt, and the proportion of the green belt area to the total image area is output as an evaluation index of the bicycle road side landscape.

10. The method for evaluating the cycling friendliness of urban roads by multimodal data fusion according to claim 1, characterized in that: The method also includes: S4, output and visualization of evaluation results: integrating sensors to realize data collection and using a single-chip microcomputer to record data, combining global positioning system positioning with map public interface location information matching, and directly recording the evaluation results in the corresponding map layer for visualization, thereby realizing the integration of data collection, data processing and evaluation result output.

Images