TWI828495B - Traffic road intelligent detection method and cloud road surface identification module - Google Patents

Abstract

The invention relates to a traffic road surface intelligent detection method and a cloud road surface identification module, which mainly uses an edge-end mobile device to record and analyze the MP4 of the PCI pavement characteristics of the selected road and its corresponding F-IMU value, B-IMU value and The GPS data is converted into several metadata, and then the several metadata are uploaded to a cloud server respectively, and then a cloud road surface recognition module built by the cloud server is based on each metadata. , respectively identify the characteristic value of a PCI pavement condition indicator and store it. Finally, the identification result information will be displayed on a geographical information system based on the GPS location information to facilitate the inquiry of traffic road management personnel and maintenance personnel.

Description

Traffic road intelligent detection method and cloud road surface identification module

The present invention relates to the technical field of traffic road management, and in particular, to the category of a road intelligent detection method and a cloud road surface recognition module for traffic roads.

According to reports, "flat roads, bright lights, and water connections" have always been public needs that all counties and municipal governments are striving to achieve, and among them, smooth roads are at the top of the list. However, the current general road level maintenance is carried out by two methods: manual visual inspections and public complaints. The existing road inspection equipment is either "very expensive" or "very complicated". For example, the country's new patent No. TWM386314U "Road Inspection Vehicle pavement flatness measurement unit".

Road condition represents the safety of many people or vehicles. By the time people complain, it is often very serious or traffic accidents have occurred. Therefore, the inventor hopes to find a way to allow the road maintenance authority to accurately identify the road sections that need to be repaired, so that the people can be more aware of road maintenance.

Therefore, in order to cooperate with the development of smart life and smart cities, the inventor has put great effort into developing a smart traffic road detection method and a cloud road surface recognition module. Therefore, the main purpose of the present invention is to provide a low-cost construction A traffic road surface intelligent detection method and a cloud road surface identification module; and the secondary purpose of the present invention is to provide an intelligent traffic road surface detection method and a cloud road surface identification module that improve the accuracy of traffic road pavement identification.

In order to achieve the above object, the technical means used by the present invention are as follows: a cloud road surface recognition module, which builds a cloud road road recognition module in a cloud server, and the cloud road road recognition module The group will identify and store the characteristic values of the PCI pavement condition indicators of bad road surfaces based on the video streams of the selected PCI pavement characteristics and the corresponding inertial posture data. Among them, the aforementioned inertial posture data is Refers to the previous inertial attitude of the three-axis attitude angle, the three-axis acceleration and the absolute pointing of the three-axis earth's magnetic field measured by a front inertial measurement unit and a rear inertial measurement unit installed at the front and rear ends of a mobile vehicle at the edge. , data of rear inertial attitude.

Regarding a traffic road surface intelligent detection method of the present invention, it includes: an edge end data collection step, which collects several streaming videos and a corresponding forward inertial posture of bad road surfaces of each traffic road through at least one edge end mobile device. , the data information of a rear inertial attitude and a GPS information; a step of edge-end bad road detection and frame selection, which is to conduct preliminary identification of the several streaming videos to frame a PCI pavement feature; a step of uploading data information to the cloud , the system integrates several streaming videos of bad road surfaces and their corresponding front inertial attitude, rear inertial attitude, and GPS information into several metadata and uploads them to the cloud server for storage; a cloud bad road surface identification step is performed in the The cloud server also includes a cloud road surface recognition module, and the cloud road road recognition module will identify the data based on the several streaming videos on the bad road surface and their corresponding front inertial posture and rear inertial posture data. The characteristic value of the store condition indicator of the streaming video is stored, in which the data of the front inertial attitude and the rear inertial attitude refer to a front inertial measurement unit and a rear end of a mobile vehicle located at the edge. The data of the three-axis attitude angle, the three-axis acceleration and the absolute pointing of the three-axis earth's magnetic field measured by the rear inertial measurement unit are the front and rear inertial attitude data; and a road surface identification result output step is based on the identification result information of the previous step. Location information based on GPS is displayed on a geographic information system.

The above-mentioned cloud road surface recognition module uses the cloud road road recognition module developed by Tensorflow development software and uses scikit-learn as the machine learning function library for data pre-processing. and establish a model, and the cloud road surface recognition module will separately identify and store the characteristic values of the pavement condition indicators of the several streaming videos based on the metadata of the bad road surface of the traffic road.

The cloud road surface recognition module adopts a supervised deep learning model framework and uses the YOLO algorithm to identify the characteristic values of pavement condition indicators of several bad road streaming videos and their corresponding front inertia posture and rear inertia posture. The data is used as a mark file of a training sample for supervised learning, and then the characteristic value of the pavement condition indicator is correspondingly associated with the front inertial attitude data and the rear inertial attitude data to improve the identification effect of bad road surfaces.

The training and testing of the cloud road surface recognition module sequentially go through a data pre-processing step, a machine learning model construction and training step, an evaluation model step and a prediction step, and in the data pre-processing step, the training samples It is divided into a 75% training set and a 25% test set.

The mark file of the training sample also contains GPS speed data.

Among them, the characteristic value of the pavement condition index is based on the damage items defined by ASTM D6433-11. There are a total of 19 damages, including alligator cracks, oil leakage, block cracks, convexities and depressions, wavy pavement, depressions, edge cracks, Reflective cracks, shoulder height differences, longitudinal and transverse cracks, patches, granular smoothing, potholes, railroad crossings, ruts, jostling, sliding cracks, bulges, and weathering and spalling.

Accordingly, the present invention can achieve the following effects through the above technical means:

1. The present invention only uses a mobile vehicle, a cloud server, an edge mobile device, an AI image cloud road recognition module, etc., and adopts an edge computing architecture to share the computing burden of the cloud server. Therefore, this invention is used The invention has low construction cost and can assist in providing road surface inspection and quick judgment of road conditions.

2. The cloud road surface recognition module of the present invention adopts a supervised deep learning model framework, and the characteristic values of the previous inertial attitude, the rear inertial attitude and the PCI pavement condition indicator are used as different Using the labeled files of training samples for good road surface image recognition, the cloud road surface recognition module trained has highly accurate recognition results.

A: Traffic road intelligent detection method

a, b, c, d, e: steps

B: Traffic road intelligent detection system

C:Mobile vehicle

C1: mobile phone holder

1: Edge mobile devices

11: Lens unit

12:Display unit

13:GPS positioning unit

14: Front inertial measurement unit

15:Memory unit

16: Communication unit

17:Arithmetic processor

2: Inertial measurement device

21:Rear inertial measurement unit

22:Transmission unit

3: Integrate application units

4:Cloud server

41: Cloud road surface recognition module

5:Geographic information system

[Figure 1] The step flow chart of the "intelligent traffic road detection method" of the present invention.

[Figure 2] Schematic diagram of the architecture of the "intelligent traffic road detection system" of the present invention.

[Figure 3] A schematic diagram of the video recording angle of a motorcycle traveling on the road for the "mobile vehicle" of the present invention.

[Figure 4] A schematic diagram of the video recording angle of a car driving on the road for the "mobile vehicle" of the present invention.

[Figure 5] Schematic diagram of the operation of the "edge mobile device" of the present invention to select bad road pavement features and record their corresponding GPS, F-IMU values, B-IMU values, and MP4 data.

[Figure 6] The operation flow chart of the "intelligent traffic road detection system" of the present invention.

[Figure 7] The flow chart of the calculation and construction of the "image recognition module of the cloud server" of the present invention.

[Figure 8] Schematic diagram of the architecture of the "edge mobile device" of the present invention.

[Figure 9] Schematic diagram of the structure of the "inertial measurement device" of the present invention.

[Fig. 10] A schematic diagram of the pavement of poor road surface according to the present invention.

[Fig. 11] A schematic diagram of the "front inertia measurement data" and "rear inertia measurement data" of the present invention.

The present invention relates to an intelligent detection method of traffic road surface and a cloud road surface identification module. It mainly provides a low-cost construction system to accurately identify the condition of bad road surface on traffic roads to assist road acceptance and maintenance personnel to quickly judge the road surface conditions. It is helpful to notify the maintenance unit to carry out road maintenance; the traffic road intelligent detection method A, as shown in Figure 1, includes: an edge end data collection step a, an edge end bad road surface detection frame selection step b , One step of uploading data information to the cloud Step c. Step d of identification of bad road surface in the cloud. Each of the above steps is described below with reference to Figures 1 and 2.

The edge end data collection step a, as shown in Figure 3 and Figure 4, is to collect several streaming videos (which can be AVI, WMV, MOV, MP4, RM and other formats (hereinafter collectively referred to as MP4) and the corresponding data information of one front inertial attitude, one rear inertial attitude, and one GPS information; among them, the optimal recording length of several streaming videos on bad road surfaces is approximately It is a 150 cm long patch of bad road surface.

That is, this step a mainly consists of installing an edge-end mobile device 1 and an inertial measurement device 2 on the front and rear ends of a mobile vehicle C. The best choice for mobile vehicle C is a motorcycle, the second best is a car, and the edge-end The mobile device 1 is preferably a smart phone (or a tablet computer, a device that is separated from the lens, such as a laptop with a photographic lens), in which the so-called inertial attitude data is obtained by an inertial measurement unit ( Data on the three-axis attitude angle, three-axis acceleration and the absolute direction of the three-axis earth's magnetic field sensed by the Inertial Measurement Unit (IMU); therefore, the front inertial attitude refers to the inertial attitude of the front end of the mobile vehicle C. The invention calls this the F-IMU value; the rear inertial attitude refers to the inertial attitude of the rear end of the mobile vehicle C, and the invention calls it the B-IMU value.

When the mobile vehicle C is driving on the road to take pictures of bad road surfaces and record the current F-IMU value, B-IMU value, and GPS data, the F-IMU value and the GPS data are obtained from the edge mobile device. 1 is responsible for sensing and collecting, and the inertial measurement device 2 is responsible for sensing and collecting the B-IMU value data. The collected B-IMU value data is transmitted to the edge mobile device 1 through wireless or wired transmission.

The edge edge bad road surface detection frame selection step b, as shown in Figure 5, is based on the object detection (Object detection) of the Tensorflow development software for the several streaming videos (MP4) on the traffic road. detector) to conduct preliminary identification and Bounding Box to identify a PCI pavement feature and simultaneously record its GPS, F-IMU value, and B-IMU value data respectively to combine them into several metadata; in particular, The object detector is an object recognition function of the Tensorflow development software.

The data information is uploaded to the cloud in step c, as shown in Figure 6. The system integrates the previous steps and uploads several metadata collected for the bad road surface of the traffic road to the cloud server 4 for storage.

The cloud bad road surface identification step d, as shown in Figure 7, is that the cloud server 4 further includes a cloud road surface identification module 41 developed by Tensorflow development software and uses scikit-learn as a machine learning function. A library is used to perform data pre-processing and build models, and the cloud road surface identification module 41 will respectively identify the pavement condition indicators of the several streaming videos of the bad road surface based on several meta-data of the bad road surface of the traffic road ( The characteristic values of the Pavement Condition Index (PCI) are stored. The characteristic values of the PCI index of bad pavement are based on the damage items defined by ASTM D6433-11. There are four major categories and 19 damage items in total, which are NO.01 crocodile. Alligator Craking, NO.02 Bleeding, NO.03 Block Cracking, NO.04 Bumps and Sags, NO.05 Corrugation, NO. .06 Depression, NO.07 Edge Cracking, NO.08 Reflection Cracking, NO.09 Lane/Shoulder Drop Off, NO.10 Longitudinal and transverse cracks (Long & Trans Cracking), NO.11 Patching & Util Cut Patching, NO.12 Polished Aggregate, NO.13 Potholes (Potholes), NO.14 Railroad crossing, NO.15 Rutting, NO.16 Shoving, NO.17 Slippage Cracking, NO.18 Swell, NO.19 Weathering/Raveling, as shown below Those shown in Table 1.

Further, in this step d, the cloud road surface recognition module 41 adopts a supervised deep learning model framework and uses the pavement condition indicators of streaming videos of several bad road surfaces identified by the YOLO algorithm. The feature value and its corresponding F-IMU value, B-IMU value and other data are used as the label of the training sample for supervised learning, thereby allowing the cloud road surface recognition module 41 The characteristic values of the pavement condition index can be identified and correlated with the data of F-IMU value, B-IMU value, etc. to improve the identification effect of bad road pavement on traffic roads; further, the training sample is composed of 75% training data (training data ) and 25% of the test set, and the marker file of the training sample can be additionally added with parameters of GPS speed data to improve the accuracy of identification.

In particular, the road surface recognition model is adapted on the training set. For supervised learning, a training set is a collection of examples used to tune parameters (such as the weights of connections between neurons in an artificial neural network). In implementation, the training set is usually a pair of input vectors (scalars) and output vectors (scalars). where the output vector (scalar) is called the target or label. During training, the current road surface recognition model makes predictions for each example in the training set and compares the predictions with the target. Based on the comparison results, the learning algorithm updates the parameters of the road surface recognition model. The process of model tuning may include both feature selection and parameter estimation.

The test set is used to provide an unbiased evaluation of the final road surface identification model.

Further, the training steps of the cloud road surface recognition module 41, as shown in Figure 7, are as follows:

Step 1: Data pre-processing

1. Use scikit-learn's train test split() function to randomly sort (shuffle) the sample data and divide it into two parts: 75% of the samples are the training set, and 25% of the samples are the test set (the proportion can be set by yourself, but based on experience Generally speaking, this ratio is good); 2. stratify=y: Stratify according to the classification ratio of the original target; 3. random_state=0: Set an integer fixed seed value (other integers are also acceptable), so that each experiment can be run The data are the same; 4. The train_test_split() function returns 4 NumPy arrays; 5. #X_train, The test answers are all one-dimensional arrays, and each element is a category data;

Step 2: Construct a machine learning model and train it

Establish a machine learning model, given training data and answers, and use the fit() fitting method to train the model;

Step 3: Evaluating the model

Use the score() method to calculate training performance: given test data and answers, calculate the proportion of correct predictions; and

Step 4: prediction

After training is completed and the performance meets the standard, prediction can be made: given new data, use the predict() method to predict the results.

Therefore, in this step d, the cloud road surface recognition module 41 will obtain an optimal recognition model through a series of training, testing and adjustment using training samples; therefore, the cloud road road surface recognition module 41 actually recognizes During the process, the MP4 files transmitted from the edge mobile device 1 will be converted into several images in advance, and then the images will be recognized by the cloud road surface recognition module 41 after noise reduction.

The road surface recognition result output step e is based on the recognition result information of the previous step being displayed on a geographical information system 5 based on the GPS location information.

Please refer to Figure 2, Figure 3 and Figure 8. Regarding a traffic road intelligent detection system B, which is constructed and implemented by the aforementioned traffic road surface intelligent detection method A of the present invention, thereby achieving the purpose of the present invention, it includes: : An edge mobile device, an inertial measurement device 2, an integrated application unit 3 and a cloud server 4 and other devices; the above-mentioned devices are described separately with reference to the drawings as follows.

The edge mobile device 1, as shown in Figure 8, is a device that can perform edge computing and is installed at the front end of a mobile vehicle C, such as the handlebar of a motorcycle. The edge mobile device 1 further includes a lens unit 11, a display unit 12, a GPS positioning unit 13, a front inertial measurement unit 14, a memory unit 15 and a communication unit 16 that are each electrically connected to a computing unit. Processor 17; wherein the lens unit 11 records the bad road surface of the road, and the GPS positioning unit 13 senses the GPS information of the mobile vehicle C (which can also be called the GPS information of the bad road surface); the front inertial measurement The unit 14 senses the F-IMU value at the front end of the mobile vehicle C, and the memory unit 15 provides the integrated application unit 3 for storing the mobile device 1, as well as MP4, GPS, F-IMU value, B- Several metadata of the IMU value; the communication unit 16 communicates with the cloud through the mobile network, the Internet After the server 4 is coupled, the plurality of metadata can be uploaded, and the computing processor 17 executes the integrated application unit 3 to provide the video MP4 of the bad road surface and frame selection, and at the same time obtain the sensed and the The GPS, F-IMU value, and B-IMU value corresponding to MP4 are used to complete the collection of metadata several times; further, the mobile vehicle C is preferably a motorcycle as shown in Figure 3, and the second best is a car as shown in Figure 4. And the grip of the front end of the motorcycle is equipped with a mobile phone holder C1 with automatic or manual angle adjustment function. The height of the mobile phone holder C1 from the ground is between 0.5 and 1.6 meters, and the speed of the mobile vehicle is set to less than Below 120km/hr to obtain a clear image; if using a car, the smartphone should be installed on the bumper, rear view mirror or windshield with a ground height between 0.5 and 1.6 meters; and the edge end The mobile device 1 is set as a smartphone (or uses a tablet computer, a device separated from the lens, such as a laptop with a camera lens, etc.) and the smartphone can be installed on the mobile phone holder C1 of the motorcycle. , and the angle θ of the lens unit 11 of the smartphone to capture the road is set to between 22 and 50 degrees; in particular, motorcycles and smartphones are very simple tools. During the process of riding the road, when people When there is a bad road surface visually, you can ride directly to the bad road surface and record it to obtain several metadata.

The inertial measurement device 2, as shown in Figures 3 and 9, is installed at the rear end of the mobile vehicle C and further includes a rear inertial measurement unit 21 and a transmission unit 22; wherein the rear inertial measurement unit 21 senses The B-IMU value at the rear end of the mobile vehicle C is measured, and the transmission unit 22 can be wireless or wired, and the data of the rear inertial attitude is pre-transmitted to the edge mobile device 1 for integration, and the obtained rear Assuming data, as shown in Table 2 and Table 3 below together with Figure 10, is one of the metadata:

In particular, the transmission unit 22 preferably adopts Bluetooth transmission.

The integrated application unit 3, as shown in Figure 8, is pre-stored and installed in the memory unit 15 of the edge mobile device 1, and is called and executed by the computing processor 17 to record bad road surfaces and collect the data. The metadata is stored, and the communication unit 16 is uploaded to the cloud server 4 at the same time.

The cloud server 4 receives several metadata including MP4, GPS, F-IMU values, and B-IMU values from the edge mobile device 1, and the cloud server 4 further includes a cloud road surface recognition Module 41, and the cloud road surface recognition module 41 will respectively identify the characteristic values of the PCI pavement condition index of the bad road surface based on the several metadata and store them, and the final recognition result will provide location information through the GPS The output is displayed on a geographic information system.

Therefore, the present invention provides a better solution for a smart traffic road detection method and a cloud road surface recognition module. After the edge mobile device 1 is installed on the mobile vehicle C and started, the integrated application unit 3 is executed. , the recognition function of the edge-end mobile device 1 will be turned on, and after correcting the shooting angle θ of the lens unit 11, the GPS, F-IMU, lens unit 11 and the inertial measurement device 2 in the edge-end mobile device 1 will be called. B-IMU; Then, continue to load the Tensorflow function library and call the road marking file to the cloud server 4, and then use Tensorflow to initially identify the PCI pavement features. When cracks, potholes, etc. are identified, start recording GPS, F-IMU, B- IMU and MP4 post-configuration data; finally start background transmission, if there is a network, the recorded post-configuration data will be transmitted to the cloud server 4 for further detailed identification of PCI pavement characteristics, that is, the cloud road surface recognition module 41 will be based on F- IMU and B-IMU data, as shown in Figure 11, find the characteristic waveform and various nine-axis parameters. Since different PCI pavement characteristics will be generated, the cloud road surface identification module 41 will finally use different nine-axis parameters to determine Identify the characteristics of each PCI interface. If there is no network, the recorded metadata will be transferred to the local end and stored in advance, and then uploaded to the cloud when there is a network.

Provide an edge collection device for the aforementioned intelligent detection method of traffic road surface, which only includes: the aforementioned edge-side mobile device 1, the aforementioned inertial measurement device 2, the aforementioned integrated application unit 3, etc., and the edge collection device It is specifically designed to collect several metadata such as MP4, GPS, F-IMU value, B-IMU value, etc. of defective pavements, and upload them to the cloud server 4 for subsequent processing. In particular, as shown in Figure 13, when the present invention collects information, since relevant data such as GPS, F-IMU, B-IMU, and images of smartphones enter at any time, this is a big data source. That is, the present invention can use big data technology for data processing. In other words, when a situation occurs (problem characteristics) from a certain device (data source), it is defined as a certain situation event, and relevant personnel are notified for processing (scripted response) , to improve the efficiency of solving equipment problems.

A: Traffic road intelligent detection method

a, b, c, d, e: steps

Claims (11)

A cloud road surface recognition module, which builds a cloud road surface recognition module in a cloud server, and the cloud road road recognition module will base on the audio and video streams of selected PCI pavement features and their corresponding inertia The attitude data are used to identify and store the characteristic values of the PCI pavement condition indicators of bad roads respectively; among them, the aforementioned inertial attitude data refers to a front inertial measurement unit located at the front end and rear end of a mobile vehicle located at the edge. 1. The data of the three-axis attitude angle, the three-axis acceleration and the absolute pointing of the three-axis earth's magnetic field measured by the inertial measurement unit before and after the inertial attitude. The cloud road surface recognition module described in request item 1 uses the cloud road road recognition module developed by Tensorflow development software and uses scikit-learn as a machine learning function library for data preprocessing and model building, and the The cloud road surface recognition module will identify and store the characteristic values of the pavement condition indicators of the several streaming videos based on the meta-data of the bad road surface of the traffic road. The cloud road surface identification module as described in claim 2, wherein the cloud road surface identification module adopts a supervised deep learning model framework and uses the YOLO algorithm to identify the characteristic values of pavement condition indicators of several bad road surface streaming videos. And its corresponding front inertial attitude, rear inertial attitude and other data are used as a mark file of a training sample for supervised learning, so that the characteristic value of the pavement condition indicator is correspondingly associated with the front inertial attitude data and the rear inertial attitude data. To improve the identification effect of bad road surfaces. The cloud road surface recognition module described in claim 3, wherein the training and testing of the cloud road road recognition module sequentially go through a data pre-processing step, a machine learning model construction and training step, an evaluation model step and a prediction step. , and in the data pre-processing step, the training samples are divided into a 75% training set and a 25% test set, and the mark file of the training sample further includes GPS speed data. For the cloud road surface identification module described in request item 1, the characteristic value of the pavement condition indicator is based on the damage items defined by ASTM D6433-11. There are a total of 19 types of damage, including crocodile cracks, oil leakage, massive cracks, and bulges. Related to depressions, wavy pavement, depressions, edge cracks, reflective cracks, shoulder height differences, longitudinal and transverse cracks, patches, granular smoothing, potholes, railroad crossings, ruts, jostling, sliding cracks, bulges and weathering and peeling Take off. A traffic road surface intelligent detection method includes: an edge end data collection step, which collects several streaming videos of bad road surfaces on each traffic road and the corresponding front inertial posture, and a rear end through at least one edge end mobile device. Inertial posture, data information of GPS information; an edge-end bad road surface detection frame selection step, which is to perform preliminary identification on the several streaming videos to frame a PCI pavement feature; a data information uploading step to the cloud, which is integrated Several streaming videos of bad road surfaces and their corresponding front inertial attitude, rear inertial attitude, and GPS information are uploaded to a cloud server as metadata for storage; a cloud bad road surface identification step is performed on the cloud server It also includes a cloud road surface recognition module, and the cloud road road recognition module will identify the several streams based on the data of the stream videos on the bad road surface and their corresponding front inertial posture and rear inertial posture. The characteristic values of the audio and video pavement condition indicators are stored, in which the data of the front inertial attitude and the rear inertial attitude refer to a front inertial measurement unit and a rear inertial measurement unit located at the front end and rear end of a mobile vehicle at the edge. The data of the three-axis attitude angle, the three-axis acceleration and the absolute pointing of the three-axis earth's magnetic field measured by the unit before and after the inertial attitude; and a road identification result output step, which is based on the identification result information of the previous step and will be based on the GPS Location information is displayed on a geographic information system. The intelligent detection method of traffic road as described in claim 6, wherein in the edge-end bad road detection frame selection step, the object detection of the Tensorflow development software is used to initially identify and identify the several streaming videos on the traffic road respectively. Select a PCI pavement feature and simultaneously record its GPS, front inertial attitude, and rear inertial attitude data to combine into several metadata. The traffic road surface intelligent detection method described in claim 7, wherein the cloud road surface identification module in the cloud bad road surface identification step uses the cloud road surface identification module developed by Tensorflow development software and uses scikit-learn as the machine learning A function library is used to perform data preprocessing and build models, and the cloud road surface recognition module will identify the characteristics of the pavement condition indicators of the several streaming videos of bad roads based on the meta-data of bad road surfaces on traffic roads. value and store it. The intelligent traffic road detection method described in claim 8, wherein the cloud road surface recognition module adopts a supervised deep learning model framework and uses the YOLO algorithm to identify the characteristic values of pavement condition indicators of several bad road streaming videos And its corresponding front inertial attitude, rear inertial attitude and other data are used as a mark file of a training sample for supervised learning, so that the characteristic value of the pavement condition indicator is correspondingly associated with the front inertial attitude data and the rear inertial attitude data. To improve the identification effect of bad road surfaces. The intelligent traffic road detection method described in claim 9, wherein the training and testing of the cloud road surface recognition module sequentially go through a data pre-processing step, a machine learning model construction and training step, an evaluation model step and a prediction step. , and in the data pre-processing step, the training samples are divided into a 75% training set and a 25% test set, and the mark file of the training sample further includes GPS speed data. The traffic road surface intelligent detection method described in claim 6, wherein in the cloud bad road surface identification step, the characteristic value of the pavement condition indicator is based on the damage items defined by ASTM D6433-11 There are a total of 19 damages in the project, including crocodile cracks, oil leakage, block cracks, convex and concave, wavy pavement, depression, edge cracks, reflective cracks, shoulder height difference, longitudinal and transverse cracks, patches, and granular materials. Slicks, potholes, track crossings, ruts, jostling, slip cracks, heaving and weathering and spalling.