Road surface defect detection model establishment method, road surface defect detection method and system
Through the road defect detection model based on machine learning, using the up and down jitter parameters and camera video data of the test vehicle, a cascading Lasso and SVM model is established, which solves the problem of low road defect detection accuracy in the existing technology, and achieves high accuracy and Fast defect detection.
Patent Information
- Application Number
- CN202110196541.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-02-22
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-02-22
AI Technical Summary
The existing road defect detection methods have low detection accuracy in complex traffic conditions and are easily affected by features such as shading, water stains, stains and rut marks, resulting in misjudgment.
A road defect detection model based on machine learning is adopted, and the up and down jitter parameters of the test vehicle are used as the training set, combining camera video and amplitude sensor data, a cascading Lasso model and linear SVM model are established to reduce the calculation amount and improve the detection accuracy.
It improves the accuracy of road surface defect detection, reduces the impact on water stains and stain noise, reduces the amount of model establishment and detection calculation, and improves the detection speed.
Smart Images

Figure CN112906575B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of road surface detection, and in particular to a road surface defect detection model establishment method, a road surface defect detection method and a road surface defect detection system. Background Art
[0002] Common road surface defects include asphalt cracks, cracks, concrete panel cracks, rutting, potholes, subsidence, etc. These defects can easily cause traffic accidents. In order to facilitate timely treatment of road surface defects, road surface defects need to be detected in a timely manner. Currently, commonly used road surface defect detection methods include computer vision-based detection methods and physical property detection-based methods.
[0003] Detection methods based on computer vision include: image algorithms for pixel feature analysis, such as extracting gradient edge features, analyzing histogram features extracted using the HOG operator, etc.; algorithms based on machine learning and feature extraction, extracting important features through feature extraction algorithms, and then performing defect detection through machine learning classification algorithms, such as sparse coding, dictionary learning, etc.; defect detection algorithms based on deep learning, mainly based on various detection and segmentation networks, such as deeplab, Mobile net, etc., extracting and analyzing image features through convolutional layers, and obtaining the bounding box or pixel set of the defect corresponding area.
[0004] Methods based on physical property detection include: ultrasonic defect detection, infrared thermal imaging flaw detection, etc.
[0005] Due to the complex road traffic conditions and the large number of pedestrians and vehicles on the road, the lens is easily blocked when collecting road images, which will seriously affect the detection accuracy of the analysis method based on computer vision. In addition, both the analysis method based on computer vision and the detection method based on physical properties are easily affected by features such as water stains, stains and wheel marks on the road surface, resulting in misjudgment. Summary of the invention
[0006] The problem solved by the present invention is how to improve the accuracy of road surface defect detection.
[0007] In order to solve the above problems, the present invention provides a road surface defect detection model establishment method, a road surface defect detection method and a system.
[0008] In a first aspect, the present invention provides a method for establishing a road surface defect detection model, comprising:
[0009] Obtaining road surface defects at various locations on the calibration road section, and up and down jitter parameters at various locations when the test vehicle travels on the calibration road section;
[0010] A model to be trained based on machine learning is established, and the model to be trained is trained with the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to obtain a road surface defect detection model, wherein the road surface defect detection model includes a cascaded Lasso model and a linear SVM model.
[0011] Optionally, the test vehicle is equipped with a camera and an amplitude sensor for detecting the up and down vibration amplitude of the test vehicle, and the obtaining of up and down vibration parameters of the test vehicle at various positions when the test vehicle is traveling on the calibration section includes:
[0012] Acquire a road surface video captured by the camera when the test vehicle travels on the calibrated road section, and a first up and down vibration amplitude of the test vehicle at each position of the calibrated road section detected by the amplitude sensor;
[0013] Processing the road surface video to obtain a second up and down shaking amplitude of the test vehicle at each position of the calibrated road section;
[0014] A weighted average is performed on the first up-and-down jitter amplitude and the second up-and-down jitter amplitude to obtain up-and-down jitter parameters of the test vehicle at various positions on the calibrated road section.
[0015] Optionally, the processing of the road surface video to obtain a second up and down shaking amplitude of the test vehicle at each position of the calibrated road section includes:
[0016] Based on the optical flow method, at least one corner point in the first frame of the road video is extracted, each corner point is tracked in the road video, and the second up and down jitter amplitude of the test vehicle at each position of the calibrated road section is determined according to the position change value of each corner point.
[0017] Optionally, a positioning device and an inertial measurement device are installed on the test vehicle, and the up and down jitter parameters of the test vehicle at various positions when the test vehicle travels on the calibration section include:
[0018] Acquire the detection position of the test vehicle on the calibration section collected by the positioning device, the motion data of the test vehicle collected by the inertial measurement device, and the up and down jitter parameters corresponding to the detection position;
[0019] The detection position is corrected according to the motion data to obtain the position of the test vehicle, and the position of the test vehicle corresponds to the up and down jitter parameters one by one.
[0020] Optionally, before training the to-be-trained model using the up-and-down jitter parameters as a training set and the corresponding road surface defects as a test set, the method includes:
[0021] The data in the training set and the test set are preprocessed using the RobustScaler algorithm to obtain a preprocessed training set and a preprocessed test set.
[0022] Optionally, the establishing of a model to be trained based on machine learning, using the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to train the model to be trained, and obtaining a road surface defect detection model comprises:
[0023] Establish Lasso model and linear SVM model respectively;
[0024] Based on a cross-validation method, the Lasso model and the linear SVM model are trained respectively using the training set and the test set to obtain a trained Lasso model and a trained linear SVM model;
[0025] The trained Lasso model and the trained linear SVM model are integrated by using a blending algorithm to obtain the road surface defect detection model.
[0026] Optionally, the adopting the training set and the test set to respectively train the Lasso model and the linear SVM model comprises:
[0027] Divide the training set into two groups of data, one group is a sub-training set, and the other group is a validation set;
[0028] Using the sub-training set to train the Lasso model and the linear SVM model respectively, to obtain the trained Lasso model and the trained linear SVM model;
[0029] Using the trained Lasso model and the trained linear SVM model to predict the validation set and the test set respectively, to obtain a first prediction set and a second prediction set;
[0030] Establishing a classifier, and using the first prediction set to train the classifier to obtain a trained classifier;
[0031] The trained classifier is used to predict the second prediction set to obtain a first weight of the trained Lasso model and a second weight of the trained linear SVM model.
[0032] Optionally, the adopting a blending algorithm to integrate the trained Lasso model and the trained linear SVM model to obtain the road surface defect detection model comprises:
[0033] The road surface defect detection model is obtained by fusing the trained Lasso model and the trained linear SVM model according to the first weight and the second weight.
[0034] In a second aspect, the present invention provides a road surface defect detection method, comprising:
[0035] Obtain the up and down vibration parameters of the test vehicle at various positions when it is driving on the road;
[0036] Input the up and down jitter parameters into a road surface defect detection model, and output road surface defects at various locations;
[0037] Wherein, the pavement defect detection model is established using the pavement defect detection model establishment method as described above.
[0038] In a third aspect, the present invention provides a road surface defect detection model establishment system, comprising:
[0039] Sensors, used to obtain up and down vibration parameters at various positions when the test vehicle is traveling on the calibration road section;
[0040] The controller is used to obtain road surface defects at various positions on the calibrated road section, establish a model to be trained based on machine learning, train the model to be trained with the upper and lower jitter parameters as a training set and the corresponding road surface defects as a test set, and obtain a road surface defect detection model, wherein the road surface defect detection model includes a cascaded Lasso model and a linear SVM model.
[0041] In a fourth aspect, the present invention provides a road surface defect detection system, comprising:
[0042] Sensors, used to obtain up and down vibration parameters at various positions when the test vehicle is driving on the road;
[0043] A controller, used for inputting the up and down shaking parameters into a road surface defect detection model and outputting road surface defects at various positions;
[0044] Wherein, the pavement defect detection model is established using the pavement defect detection model establishment method as described above.
[0045] In a fourth aspect, the present invention provides an electronic device, comprising a memory and a processor;
[0046] The memory is used to store computer programs;
[0047] The processor is used to implement the above-mentioned road surface defect detection model establishment method or the above-mentioned road surface defect detection method when executing the computer program.
[0048] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the pavement defect detection model establishment method or the pavement defect detection method as described above is implemented.
[0049] The beneficial effects of the road surface defect detection model establishment method, road surface defect detection method and system of the present invention are: obtaining road surface defects at various positions on the calibration section, which can be detected by methods such as manual detection, and the up and down jitter parameters at various positions when the test vehicle is traveling on the calibration section. The up and down jitter parameters can be the up and down jitter amplitude of the body of the test vehicle. The road surface defect at each position corresponds to the up and down jitter parameters at the position. The up and down jitter parameters are used as the training set and the corresponding road surface defects are used as the test set to train the model to be trained based on machine learning, and the road surface defect detection model can be obtained. The road surface defect detection model reflects the connection between the up and down jitter parameters and the road surface defects. Subsequently, it is only necessary to detect the up and down jitter parameters of the test vehicle and input them into the road surface defect model to quickly determine the road surface defects at the corresponding position.
[0050] The technical solution of the present invention uses the up and down jitter parameters of the test vehicle and the road surface defects to establish a road surface defect detection model. Compared with the prior art, the present invention will not be disturbed by noise such as water stains and stains, and can greatly improve the accuracy of defect detection. In addition, the present application establishes a road surface defect detection model based on the up and down jitter parameters of the test vehicle. The up and down jitter parameters are one-dimensional signals. Compared with the analysis method based on computer vision, the present invention can greatly reduce the amount of calculation for model establishment and defect detection, and improve the speed of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 A schematic diagram of a road surface defect detection scenario according to an embodiment of the present invention;
[0052] Figure 2 A schematic diagram of a process for establishing a road surface defect detection model according to an embodiment of the present invention;
[0053] Figure 3 A schematic flow chart of a method for establishing a road surface defect detection model according to another embodiment of the present invention;
[0054] Figure 4 A schematic diagram of a process of a road surface defect detection method according to an embodiment of the present invention;
[0055] Figure 5 A schematic diagram of the structure of a road surface defect detection model establishment system and a road surface defect detection system according to an embodiment of the present invention.
[0056] Description of reference numerals:
[0057] 1- camera; 2- amplitude sensor; 3- positioning device; 4- controller; 5- road network map; 6. Segmented observation data; 7- road surface defect detection model; 8- Internet; 9- server; 10- computer. DETAILED DESCRIPTION
[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0059] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein.
[0060] like Figure 1 As shown, in the present invention, when the test vehicle is traveling on a certain road section, the position of the test vehicle is detected in real time by a positioning device 3 installed on the test vehicle. The positioning device 3 may be a GPS (Global Positioning System) positioning device 3 or a Beidou satellite navigation system, etc., and the first up and down jitter amplitude of the test vehicle when traveling at various positions of the road section is detected by an amplitude sensor 2 installed on the test vehicle. The amplitude sensor 2 may be a gyroscope, etc., and a road surface video is captured by a camera 1 installed on the test vehicle. The first up and down jitter amplitude and road surface video at various positions of the road section are input into a controller 4, and the controller 4 processes the road surface video to determine the second up and down jitter amplitude of the test vehicle at various positions, and performs weighted averaging of the first up and down jitter amplitude and the second up and down jitter amplitude to obtain the up and down jitter parameters of the test vehicle at various positions. The road network of the road section obtained in advance is input into the controller 4, and the controller 4 processes the road surface video to determine the second up and down jitter amplitude of the test vehicle at various positions. The controller 4 performs weighted averaging of the first up and down jitter amplitude and the second up and down jitter amplitude to obtain the up and down jitter parameters of the test vehicle at various positions. Figure 5 The input controller 4 segments the up and down jitter parameters at predetermined time intervals to obtain a plurality of segmented observation data 6, each segmented observation data 6 includes all the up and down jitter parameters corresponding to the time period, and the segmented observation data 6 is input into the road surface defect detection model 7 to output the detection result, which includes the road surface defect in the road network. Figure 5 The position in the road surface, the type of road surface defect, etc. The detection result can be input to the server 9 through the Internet 8 for storage, and can be read by a computer 10 or other device when used.
[0061] like Figure 2 and Figure 3 As shown, a method for establishing a road surface defect detection model provided by an embodiment of the present invention includes:
[0062] Step S110, obtaining road surface defects at various positions on the calibrated road section, and up and down jitter parameters at various positions when the test vehicle travels on the calibrated road section.
[0063] Specifically, all road surface defects on the calibration section can be determined in advance by manual exploration methods, and the types of road surface defects and their locations on the calibration section can be recorded. The road surface may include cracks, foreign objects, settlements, ruts and potholes, etc. The test vehicle can be a bus, etc.
[0064] Optionally, the test vehicle is equipped with a camera 1 and an amplitude sensor 2 for detecting the up and down vibration amplitude of the test vehicle, and the up and down vibration parameters of the test vehicle at various positions when the test vehicle is traveling on the calibration section include:
[0065] The road surface video captured by the camera 1 when the test vehicle is traveling on the calibrated road section and the first up and down vibration amplitude of the test vehicle at each position of the calibrated road section detected by the amplitude sensor 2 are obtained.
[0066] Specifically, the camera 1 may be a high-speed camera, and the amplitude sensor 2 may be a gyroscope.
[0067] The road surface video is processed to obtain a second up and down shaking amplitude of the test vehicle at each position of the calibrated road section.
[0068] Optionally, the processing of the road surface video to obtain a second up and down shaking amplitude of the test vehicle at each position of the calibrated road section includes:
[0069] Based on the optical flow method, at least one corner point in the first frame of the road video is extracted, each corner point is tracked in the road video, and the second up and down jitter amplitude of the test vehicle at each position of the calibrated road section is determined according to the position change value of each corner point.
[0070] In this optional embodiment, each corner point is first extracted in the first frame of the road video to determine the initial position of the corner point. Then, each corner point is tracked in the road video and the position of each corner point is determined in each frame. For any frame, the position change value of the corner point is determined based on the position of the corner point in the frame and the initial position of the corner point. The position change value of each corner point is determined, and all the position change values are calculated for processing, such as weighted averaging, to obtain the average position change value of all corner points. The average position change is the second up and down jitter amplitude of the test vehicle at the position corresponding to the frame.
[0071] A weighted average is performed on the first up-and-down jitter amplitude and the second up-and-down jitter amplitude to obtain up-and-down jitter parameters of the test vehicle at various positions on the calibrated road section.
[0072] In this optional embodiment, the timestamps of the first up and down jitter amplitude and the second up and down jitter amplitude are matched, and weighted average is performed according to the confidence of the first up and down jitter amplitude and the second up and down jitter amplitude. The up and down jitter amplitude detected by the camera 1 and the up and down jitter amplitude collected by the amplitude sensor 2 are combined to determine the up and down jitter parameters of the test vehicle, which can reduce the error caused by single device detection and improve the accuracy of data acquisition.
[0073] Optionally, the test vehicle is equipped with a positioning device 3 and an inertial measurement device, and the up and down jitter parameters of the test vehicle at various positions when traveling on the calibration section include:
[0074] Acquire the detection position of the test vehicle on the calibration section collected by the positioning device 3, the motion data of the test vehicle collected by the inertial measurement device, and the up and down jitter parameters corresponding to the detection position;
[0075] The detection position is corrected according to the motion data to obtain the position of the test vehicle, and the position of the test vehicle corresponds to the up and down jitter parameters one by one.
[0076] In this optional embodiment, the positioning device 3 may adopt a GPS (Global Positioning System) positioning device 3 or a Beidou satellite navigation system, etc. The motion data includes the three-axis attitude angle and acceleration of the test vehicle, etc. The inertial measurement device is a device for measuring the three-axis attitude angle and acceleration of the test vehicle. Generally, an inertial measurement device includes three single-axis accelerometers and three single-axis gyroscopes.
[0077] Since satellite positioning signals such as GPS are easily blocked by buildings on urban roads, resulting in large errors, the motion data collected by the inertial measurement device is used to correct the position of the test vehicle determined by the positioning device 3, which can improve the positioning accuracy and reduce the error.
[0078] Step S120, establish a model to be trained based on machine learning, train the model to be trained with the up and down jitter parameters as a training set and the corresponding road surface defects as a test set, and obtain a road surface defect detection model 7, wherein the road surface defect detection model 7 includes a cascaded Lasso model and a linear SVM model.
[0079] Optionally, before training the to-be-trained model using the up-and-down jitter parameters as a training set and the corresponding road surface defects as a test set, the method includes:
[0080] The data in the training set and the test set are preprocessed using the RobustScaler algorithm to obtain a preprocessed training set and a preprocessed test set.
[0081] Specifically, the RobustScaler method is used to standardize the data in the training set and the test set. The RobustScaler method is a general data standardization method, which is mainly used to exclude outliers. Its principle is to sort the data in the data set and divide it into four parts evenly, and reduce the data to the interquartile range, which is the distance between the first quartile (25%) and the third quartile (75%). The specific calculation method is:
[0082] ,
[0083] in, is the standardized data, is any data in the dataset, is the median value of the data in the data set, that is, the median. is the interquartile range.
[0084] In this optional embodiment, when the data contains many outliers, using mean and variance scaling is not a good choice. At this time, using robust_scale and RobustScaler as alternatives can use more robust estimates of the center and range of the data.
[0085] Optionally, the establishing of a model to be trained based on machine learning, using the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to train the model to be trained, and obtaining the road surface defect detection model 7 comprises:
[0086] Lasso model and linear SVM model were established respectively.
[0087] Specifically, the Lasso (Least absolute shrinkage and selection operator) model and the linear SVM (Support Vector Machine) model are both classic linear models in machine learning. The Lasso model is used for regression to compress data and then make predictions. It is a compression estimate that constructs a penalty function to obtain a more refined model, which compresses some regression coefficients, that is, forces the sum of the absolute values of the coefficients to be less than a fixed value, and sets some regression coefficients to zero, retaining the advantages of subset shrinkage. It is a biased estimate for processing data with complex collinearity. The SVM model is used for classification. The premise of using a linear kernel is that the data in the test is linearly separable, and the linear kernel is used to ultimately distinguish between outliers and normal values.
[0088] Based on the cross-validation method, the training set and the test set are used to respectively train the Lasso model and the linear SVM model to obtain a trained Lasso model and a trained linear SVM model.
[0089] Specifically, the preprocessed training set and the preprocessed test set are used to train the Lasso model and the linear SVM model respectively.
[0090] The training set and the test set are segmented according to the calibrated time length. For example, the data in the data set can be segmented in units of 10 seconds. The segmented up and down jitter parameters and the corresponding road surface defects are input into the Lasso model and the linear SVM model for training.
[0091] In this optional embodiment, both the Lasso model and the linear SVM model use the determination coefficient r2 as the final determination coefficient. The determination coefficient r2 is a linear regression determination coefficient. The determination coefficient reflects how much percentage of the fluctuation of the dependent variable y can be described by the fluctuation of the independent variable x, that is, how much percentage of the variation of the dependent variable y can be explained by the controlled independent variable x. The determination coefficient r2 can measure the prediction ability of the model, that is, the correlation between the true value and the prediction. The closer the data predicted by the model is to the true data, the larger r2 is and the closer it is to 1.
[0092] In the analysis method based on computer vision in the prior art, the collected image is analyzed and the image is a three-dimensional signal. The present invention uses the upper and lower jitter parameters as the training set to train the road surface defect detection model 7, and the upper and lower jitter parameters are one-dimensional amplitude signals. Compared with the analysis method based on computer vision, the amount of calculation for model establishment and determination of road surface defects by the model is greatly reduced, the running speed of the model is improved, the number of samples required in the model training process is smaller, and the road surface defect detection model 7 can be converged faster.
[0093] Optionally, the adopting the training set and the test set to respectively train the Lasso model and the linear SVM model comprises:
[0094] Divide the training set into two groups of data, one group is a sub-training set, and the other group is a validation set;
[0095] Using the sub-training set to train the Lasso model and the linear SVM model respectively, to obtain the trained Lasso model and the trained linear SVM model;
[0096] Using the trained Lasso model and the trained linear SVM model to predict the validation set and the test set respectively, to obtain a first prediction set and a second prediction set;
[0097] Establishing a classifier, and using the first prediction set to train the classifier to obtain a trained classifier;
[0098] The trained classifier is used to predict the second prediction set to obtain a first weight of the trained Lasso model and a second weight of the trained linear SVM model.
[0099] The trained Lasso model and the trained linear SVM model are integrated using a blending algorithm to obtain the road surface defect detection model 7.
[0100] Optionally, the adopting a blending algorithm to integrate the trained Lasso model and the trained linear SVM model to obtain the road surface defect detection model 7 comprises:
[0101] The road surface defect detection model 7 is obtained by fusing the trained Lasso model and the trained linear SVM model according to the first weight and the second weight.
[0102] In this embodiment, the road surface defects at various positions on the calibration section are obtained, which can be detected by methods such as manual detection, and the up and down jitter parameters at various positions when the test vehicle is traveling on the calibration section. The up and down jitter parameters can be the up and down jitter amplitude of the body of the test vehicle. The road surface defects at each position correspond to the up and down jitter parameters at that position. The up and down jitter parameters are used as the training set and the corresponding road surface defects are used as the test set to train the machine learning-based training model to obtain the road surface defect detection model 7. The road surface defect detection model 7 reflects the connection between the up and down jitter parameters and the road surface defects. In the future, it is only necessary to detect the up and down jitter parameters of the test vehicle and input them into the road surface defect model to quickly determine the road surface defects at the corresponding position.
[0103] The technical solution of the present invention uses the up and down jitter parameters of the test vehicle and the road surface defects to establish a road surface defect detection model 7. Compared with the prior art, the present invention will not be disturbed by noise such as water stains and stains, can highlight the characteristics of road surface defects, and can greatly improve the accuracy of defect detection. In addition, the present application establishes a road surface defect detection model 7 based on the up and down jitter parameters of the test vehicle. The up and down jitter parameters are one-dimensional signals. Compared with the analysis method based on computer vision, the present invention can greatly reduce the amount of calculation for model establishment and defect detection, and improve the speed of defect detection.
[0104] like Figure 4 As shown, a road surface defect detection method provided by an embodiment of the present invention includes:
[0105] Step S210, obtaining up and down vibration parameters of the test vehicle at various positions when the test vehicle is traveling on the road;
[0106] Step S220, inputting the up and down jitter parameters into the road surface defect detection model 7, and outputting the road surface defects at various positions;
[0107] The pavement defect detection model 7 is established by using the pavement defect detection model establishment method as described above.
[0108] In this embodiment, when the road surface defect detection model 7 established based on the up and down jitter parameters and the road surface defects is used to analyze road surface defects, it has a wide range of applications and can be applied to road surface defect detection in various other road sections.
[0109] Optionally, the road surface defect detection model includes a cascaded Lasso model and a linear SVM model.
[0110] In this optional embodiment, the Lasso model and the linear SVM model are pre-trained. When determining the road surface defects corresponding to each position of the road surface based on the up and down jitter parameters, the up and down jitter parameters are respectively input into the Lasso model and the linear SVM model to obtain a first prediction value output by the Lasso model and a second prediction value output by the linear SVM model. The first prediction value and the second prediction value are weightedly averaged according to a first weight corresponding to the Lasso model and a second weight corresponding to the linear SVM model to obtain a final prediction value. The final prediction value is the prediction value output by the road surface defect detection model based on the up and down jitter parameters.
[0111] like Figure 5 As shown, a road surface defect detection model establishment system provided by an embodiment of the present invention includes:
[0112] The sensor is used to obtain the up and down vibration parameters of the test vehicle at various positions when it is traveling on the calibration section.
[0113] Specifically, the sensor includes an amplitude sensor 2 and a camera 1 or a camera, etc. The amplitude sensor 2 is used to detect the first up and down jitter amplitude of the test vehicle at various positions on the calibrated section, and the camera 1 or the camera is used to collect road surface video when the test vehicle is traveling on the calibrated section.
[0114] The controller 4 is used to obtain the road surface defects at various positions on the calibrated road section, establish a model to be trained based on machine learning, train the model to be trained with the upper and lower jitter parameters as a training set and the corresponding road surface defects as a test set, and obtain a road surface defect detection model 7, wherein the road surface defect detection model 7 includes a cascaded Lasso model and a linear SVM model.
[0115] Specifically, the controller 4 can be used to process the road video to obtain the second up and down jitter amplitude of the test vehicle at various positions on the calibrated road section, and the first up and down jitter amplitude and the second up and down jitter amplitude are weighted averaged to obtain the up and down jitter parameters.
[0116] like Figure 5 As shown, a road surface defect detection system provided by an embodiment of the present invention includes:
[0117] Sensors, used to obtain up and down vibration parameters at various positions when the test vehicle is driving on the road;
[0118] A controller 4, used for inputting the up and down shaking parameters into a road surface defect detection model 7, and outputting road surface defects at various positions;
[0119] The pavement defect detection model 7 is established by using the pavement defect detection model establishment method as described above.
[0120] An electronic device provided by an embodiment of the present invention includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the road surface defect detection model establishment method or the road surface defect detection method as described above when executing the computer program.
[0121] A computer-readable storage medium provided by an embodiment of the present invention stores a computer program, and when the computer program is executed by a processor, the above-mentioned road surface defect detection model establishment method or the above-mentioned road surface defect detection method is implemented.
[0122] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
[0123] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.
Claims
1. A method for establishing a road surface defect detection model, characterized in that: include: Acquire road surface defects at various positions on the calibration road section, and up and down shaking parameters at various positions when the test vehicle travels on the calibration road section, wherein the up and down shaking parameters include the up and down shaking amplitude of the body of the test vehicle; Establishing a model to be trained based on machine learning, using the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to train the model to be trained, and obtaining a road surface defect detection model, wherein the road surface defect detection model includes a cascaded Lasso model and a linear SVM model, and the road surface defect detection model is obtained by integrating the trained Lasso model and the trained linear SVM model using a blending algorithm, and the road surface defect detection model is used to reflect the relationship between the up and down jitter parameters and the road surface defects; The method of establishing a model to be trained based on machine learning, using the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to train the model to be trained, and obtaining a road surface defect detection model comprises: establishing a Lasso model and a linear SVM model respectively; based on a cross-validation method, using the training set and the test set to train the Lasso model and the linear SVM model respectively, to obtain a trained Lasso model and a trained linear SVM model; using a blending algorithm to integrate the trained Lasso model and the trained linear SVM model, to obtain the road surface defect detection model; The method of using the training set and the test set to respectively train the Lasso model and the linear SVM model comprises: dividing the training set into two groups of data, one group being a sub-training set and the other group being a validation set; using the sub-training set to respectively train the Lasso model and the linear SVM model to obtain the trained Lasso model and the trained linear SVM model; using the trained Lasso model and the trained linear SVM model to respectively predict the validation set and the test set to obtain a first prediction set and a second prediction set; establishing a classifier, using the first prediction set to train the classifier to obtain a trained classifier; using the trained classifier to predict the second prediction set to obtain a first weight of the trained Lasso model and a second weight of the trained linear SVM model; The method of integrating the trained Lasso model and the trained linear SVM model using a blending algorithm to obtain the pavement defect detection model includes: fusing the trained Lasso model and the trained linear SVM model according to the first weight and the second weight to obtain the pavement defect detection model.
2. The method for establishing a road surface defect detection model according to claim 1, characterized in that: The test vehicle is equipped with a camera and an amplitude sensor for detecting the up and down shaking amplitude of the test vehicle. The road surface defects at various positions on the calibration section and the up and down shaking parameters at various positions when the test vehicle is traveling on the calibration section include: Acquire a road surface video captured by the camera when the test vehicle travels on the calibrated road section, and a first up and down vibration amplitude of the test vehicle at each position of the calibrated road section detected by the amplitude sensor; Processing the road surface video to obtain a second up and down shaking amplitude of the test vehicle at each position of the calibrated road section; A weighted average is performed on the first up-and-down jitter amplitude and the second up-and-down jitter amplitude to obtain up-and-down jitter parameters of the test vehicle at various positions on the calibrated road section.
3. The method for establishing a road surface defect detection model according to claim 2, characterized in that: The processing of the road surface video to obtain the second up and down shaking amplitude of the test vehicle at each position of the calibrated road section includes: Based on the optical flow method, at least one corner point in the first frame of the road video is extracted, each corner point is tracked in the road video, and the second up and down jitter amplitude of the test vehicle at each position of the calibrated road section is determined according to the position change value of each corner point.
4. The method for establishing a road surface defect detection model according to claim 1, characterized in that: The test vehicle is equipped with a positioning device and an inertial measurement device. The up and down jitter parameters of the test vehicle at various positions when traveling on the calibration section include: Acquire the detection position of the test vehicle on the calibration section collected by the positioning device, the motion data of the test vehicle collected by the inertial measurement device, and the up and down jitter parameters corresponding to the detection position; The detection position is corrected according to the motion data to obtain the position of the test vehicle, and the position of the test vehicle corresponds to the up and down jitter parameters one by one.
5. The method for establishing a road surface defect detection model according to claim 1, characterized in that: Before training the model to be trained using the up-and-down jitter parameters as a training set and the corresponding road surface defects as a test set, the method includes: The data in the training set and the test set are preprocessed using the RobustScaler algorithm to obtain a preprocessed training set and a preprocessed test set.
6. A road surface defect detection method, characterized in that: include: Obtain the up and down vibration parameters of the test vehicle at various positions when it is driving on the road; Input the up and down jitter parameters into a road surface defect detection model, and output road surface defects at various locations; Wherein, the pavement defect detection model is established by using the pavement defect detection model establishment method as described in any one of claims 1 to 5.
7. A road surface defect detection model building system, characterized in that: include: A sensor, used to obtain up and down shaking parameters at various positions of the test vehicle when the test vehicle is traveling on a calibration road section, wherein the up and down shaking parameters include the up and down shaking amplitude of the body of the test vehicle; A controller, used to obtain road surface defects at various positions on the calibrated road section, establish a model to be trained based on machine learning, train the model to be trained with the up and down jitter parameters as a training set and the corresponding road surface defects as a test set, and obtain a road surface defect detection model, wherein the road surface defect detection model includes a cascaded Lasso model and a linear SVM model, the road surface defect detection model is obtained by integrating the trained Lasso model and the trained linear SVM model using a blending algorithm, and the road surface defect detection model is used to reflect the relationship between the up and down jitter parameters and the road surface defects; The method of establishing a model to be trained based on machine learning, using the up and down jitter parameters as a training set and the corresponding road surface defects as a test set to train the model to be trained, and obtaining a road surface defect detection model comprises: establishing a Lasso model and a linear SVM model respectively; based on a cross-validation method, using the training set and the test set to train the Lasso model and the linear SVM model respectively, to obtain a trained Lasso model and a trained linear SVM model; using a blending algorithm to integrate the trained Lasso model and the trained linear SVM model, to obtain the road surface defect detection model; The method of using the training set and the test set to respectively train the Lasso model and the linear SVM model comprises: dividing the training set into two groups of data, one group being a sub-training set and the other group being a validation set; using the sub-training set to respectively train the Lasso model and the linear SVM model to obtain the trained Lasso model and the trained linear SVM model; using the trained Lasso model and the trained linear SVM model to respectively predict the validation set and the test set to obtain a first prediction set and a second prediction set; establishing a classifier, using the first prediction set to train the classifier to obtain a trained classifier; using the trained classifier to predict the second prediction set to obtain a first weight of the trained Lasso model and a second weight of the trained linear SVM model; The method of integrating the trained Lasso model and the trained linear SVM model using a blending algorithm to obtain the pavement defect detection model includes: fusing the trained Lasso model and the trained linear SVM model according to the first weight and the second weight to obtain the pavement defect detection model.
8. A road surface defect detection system, characterized in that: include: Sensors, used to obtain up and down vibration parameters at various positions when the test vehicle is driving on the road; A controller, used for inputting the up and down shaking parameters into a road surface defect detection model and outputting road surface defects at various positions; Wherein, the pavement defect detection model is established by using the pavement defect detection model establishment method as described in any one of claims 1 to 5.
9. An electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the road surface defect detection model establishment method according to any one of claims 1 to 5 or the road surface defect detection method according to claim 6 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the pavement defect detection model establishment method as described in any one of claims 1 to 5 or the pavement defect detection method as described in claim 6 is implemented.
