Road crack detection method based on vehicle-mounted sensor

Abstract

The invention relates to a road crack detection method based on a vehicle-mounted sensor, and relates to the technical field of road detection, and the method comprises the steps: S1, loading the vehicle-mounted sensor at the front end of a vehicle, and obtaining a high-quality road surface image and road reflection wave related data; s2, setting parameters of the vehicle-mounted sensor according to the vehicle speed and the preset distance of the vehicle-mounted sensor, and acquiring high-quality road surface images and road reflection wave related data along with the driving of the vehicle; the road crack detection method is constructed through the vehicle-mounted sensor carrying the high-definition camera and the ground penetrating radar, so that the accuracy and the high efficiency of the road crack are ensured through mutual verification of the high-definition camera and the ground penetrating radar, and automatic identification and calibration of the crack after detection are completed based on reinforcement learning. An efficient solution is provided for researching road crack detection, and the method has wide applicability.

Description

Technical Field

[0001] The invention belongs to the technical field of road detection and relates to a road crack detection method based on a vehicle-mounted sensor. Background Art

[0002] In modern urban traffic systems, road safety and reliability are crucial to the lives of drivers and pedestrians. However, over time, cracks and damage on the road surface may cause traffic accidents and vehicle damage, seriously affecting the service life of the road and traffic efficiency. Therefore, timely and accurate detection and calibration of road cracks and their locations have become one of the key issues in road maintenance and management. Traditional road crack detection methods mainly rely on manual inspections or the use of professional equipment, which are usually time-consuming, labor-intensive and costly, and may involve subjective judgments and errors.

[0003] In recent years, the development of vehicle-mounted sensing technologies, such as high-definition cameras and ground-penetrating radars, has provided new possibilities for road crack detection.

[0004] For example, patent announcement number CN 220318324 U is a road crack detection device based on a vehicle-mounted camera and radar, including a shielding component, a positioning component and a shooting component. A support seat is provided on the shielding component, and a connecting base is connected to the middle part of one side of the support seat by bolts. The positioning component and the shooting component cooperate with each other to adjust the camera on the top of the shooting component so that the camera can obtain a wider field of view. On the one hand, it is conducive to the camera to detect road cracks more comprehensively, thereby facilitating maintenance personnel to more thoroughly complete the repair work on the cracks on the road surface, which is not only conducive to extending the service life of the road, but also conducive to ensuring road safety and reducing the risk of traffic accidents. On the other hand, the detection of road cracks by the positioning component and the shooting component saves time and effort, and further improves the efficiency of road crack detection.

[0005] However, the existing technology fails to combine computer vision and machine learning technology to realize automatic detection and calibration of road cracks, thereby improving the accuracy, efficiency and cost-effectiveness of detection. Based on the above background, the present invention aims to develop a road crack detection method based on vehicle-mounted sensors, which uses images obtained by high-definition cameras and data obtained by ground-penetrating radars, combined with advanced technologies such as reinforcement learning, to realize automatic recognition and location calibration of road cracks, and provide reliable technical support for road maintenance and management. Summary of the invention

[0006] The present invention mainly relates to a road crack detection method based on vehicle-mounted sensors, which uses high-definition camera images and ground-penetrating radar data obtained by vehicle-mounted sensors, combined with advanced technologies such as reinforcement learning, to realize automatic identification and position calibration of road cracks. Its main functions include: obtaining images and data of the road surface in real time through vehicle-mounted sensors; analyzing images using computer vision technology to identify the location and shape of road cracks; further verifying and accurately locating cracks by processing ground-penetrating radar data; optimizing crack detection algorithms based on reinforcement learning methods to improve detection accuracy and robustness. The advantages of the present invention are: realizing automatic detection and calibration of road cracks, reducing the time and cost of manual inspections; using vehicle-mounted sensor data to realize real-time monitoring of road cracks, improving the efficiency and timeliness of road maintenance; at the same time, through the application of reinforcement learning methods, the accuracy and robustness of crack detection algorithms are further improved, and it is suitable for road environments of different types and complexities.

[0007] The purpose of the present invention can be achieved by the following technical solutions: The present invention provides a road crack detection method based on a vehicle-mounted sensor, comprising the following steps: S1: The vehicle-mounted sensors are installed on the front of the vehicle. The vehicle-mounted sensors include a high-definition camera with a resolution of more than 8 million pixels and a ground-penetrating radar with a detection distance of 30-500 cm, which are used to obtain high-quality road surface images and road reflection wave related data; S2: Set the parameters of the high-definition camera and ground-penetrating radar according to the vehicle speed and the preset distance of the on-board sensor, and obtain high-quality road surface images and road reflection wave related data as the vehicle travels; S3: Construct a road crack detection database, including a high-definition image database and a reflected wave signal database. The collected high-definition road surface images are extracted to obtain sub-block images at each moment to form an image set, and the reflected wave data is converted into a digital signal sequence to form a data set. S4: performing data preprocessing, preprocessing the acquired image data, including steps such as denoising by using Gaussian filtering, cropping based on edge detection, feature extraction and normalization based on gray-level co-occurrence matrix, and similarly, preprocessing the reflected wave digital signal sequence, including steps such as filtering by using Butterworth bandpass filter, DC offset correction, background removal based on spectral subtraction, gain adjustment based on adaptive gain control algorithm, and data smoothing based on moving average filter; S5: Build a road crack recognition model to identify and calibrate the location of road cracks based on reinforcement learning, including crack detection model training and crack detection testing, to complete the detection and calibration of road cracks.

[0008] Furthermore, in step S1, the on-board sensor installed at the front end of the vehicle is composed of a monocular camera with more than 8 million pixels and a ground-penetrating radar with a detection distance of 30-500 cm, ensuring the high quality of the collected road data. Both the high-definition camera and the ground-penetrating radar can accurately identify the road surface status, facilitating the subsequent identification and calibration of road cracks.

[0009] Furthermore, in step S2, it is necessary to set parameters for the high-definition camera and the ground-penetrating radar to ensure that after the vehicle-mounted sensor is installed on the vehicle, a sufficiently clear image with a suitable size and a reflected wave signal with a high signal-to-noise ratio and good stability can be obtained. The settings of the high-definition camera include shooting area confirmation, optical focus, lens correction, and image overlap detection, etc., to ensure that the vehicle can collect clear and high-quality images during driving, which is convenient for subsequent preprocessing and extraction of images. The settings of the ground-penetrating radar include signal frequency, sampling interval, sampling duration, and scanning depth, etc., to ensure that the sensor can receive a signal sequence with high signal strength and good signal quality. By obtaining high-quality road detection data, subsequent data processing is facilitated, and the accuracy and efficiency of road crack detection are enhanced.

[0010] Furthermore, in step S3, the images captured by the high-definition camera are extracted using Pix4Dmapper software to merge repeated ground features into a single, accurate image to improve the clarity and information content of the image, and are combined to form a high-quality image set, so that the extraction of ground features of the road surface and its surrounding environment will be more accurate and reliable, providing reliable data support for subsequent road crack detection.

[0011] Furthermore, in step S3, for the reflected wave signal obtained by the ground penetrating radar, the continuous analog signal is converted into a digital signal sequence through an analog-to-digital converter, and stored to form a reflected wave signal set, so as to better utilize digital technology to process, transmit and store the obtained reflected wave signal, thereby improving the reliability, accuracy and applicability of the signal, and providing more accurate and reliable information for subsequent road crack detection.

[0012] Further, in step S4, for the high-quality image data set, first, the original image is subjected to Gaussian filtering denoising to eliminate the interference noise in the image, ensuring that the subsequent processing steps can accurately analyze the actual road conditions. Secondly, the denoised image is cropped to cut the image area into 608 pixels × 608 pixels of the road surface part, and the irrelevant area is removed, thereby improving the processing efficiency and reducing the computational burden. Thirdly, on the cropped image, a feature extraction operation is performed, and a texture analysis algorithm based on the gray-level co-occurrence matrix is ​​used to capture the key features of the road cracks, such as texture, shape, and color. Through feature extraction, the road image is converted into a representation with richer information, providing strong support for subsequent machine learning and model training. Finally, the extracted features are normalized to ensure that they are within the same scale range to avoid model training instability caused by differences in the numerical ranges of different features.

[0013] Furthermore, in step S4, for the collected reflected wave signal set, first, a Butterworth bandpass filter is used to filter the digital signal to effectively remove high-frequency noise and low-frequency trends, and retain the frequency components related to road cracks. Secondly, for the DC offset of the signal, correction is performed by calculating and subtracting the mean of the signal to ensure that the signal baseline is stable. Thirdly, spectral subtraction is used to remove background noise, and the influence of fixed reflections in the surface layer is eliminated by comparing the average values ​​of multiple scanning signals, thereby highlighting important underground features. In addition, in response to the problem that the signal strength weakens with depth, adaptive gain adjustment is applied according to the signal depth to enhance the detection capability of deep signals. Finally, the signal is smoothed by applying a moving average or Gaussian smoothing algorithm to reduce random noise and sharp changes to improve the stability and readability of the signal.

[0014] Further, in step S5, it includes environment modeling, strategy selection, model training, model testing, model optimization and detection application. First, the road crack detection problem is modeled as a reinforcement learning environment, and the state space, action space, reward function and learning objectives are defined. The state space is represented by states consisting of camera images and reflected wave signal sequences. These states should contain sufficient information to describe the location, shape and other relevant features of road cracks. The action space is the prediction of the location and size of the cracks. For the crack detection task, the reward function can be based on the difference between the crack location predicted by the model and the actual crack location, and the learning goal is determined so that the action selected in a given state can maximize the cumulative reward. Secondly, the DQN algorithm is selected to train the road crack detection model, and the optimal strategy is learned through interaction with the environment, so as to realize the detection and calibration of road cracks. Thirdly, the model is tested to evaluate its performance. The accuracy and robustness of the model are evaluated by running the model in a simulated environment based on a real experimental scenario and comparing its predicted results with the actual crack location and size, and the model is optimized according to the test results, including adjusting the model architecture, hyperparameter tuning, and increasing training data. Finally, the trained model is deployed on the vehicle sensor data set preprocessed in step S4 to achieve detection and calibration of road cracks.

[0015] Further, in step S5, the detection of road cracks includes inputting the pre-processed image and reflected wave data into the detection model, outputting the location and shape of the cracks, and post-processing the model output to filter false detections, merge adjacent cracks, etc. The calibration of road cracks includes determining the specific location of the cracks on the road based on the detection results, and calibrating the type, location, length, width and depth of the cracks in combination with a geographic information system (GIS) or other positioning system.

[0016] The present invention has at least the following beneficial effects: the present invention uses high-definition cameras and ground-penetrating radars to perform dual detection and mutual verification through automated road crack detection and calibration, timely discovers and repairs road cracks, reduces the possibility of traffic accidents, and improves the safety and reliability of road use. In addition, the present invention reduces the time and cost of manual inspections, improves maintenance efficiency, and avoids road damage and increased repair costs caused by failure to discover cracks in time. By using on-board sensor data, real-time monitoring of road cracks is achieved, improving the efficiency and timeliness of road maintenance; at the same time, through the application of reinforcement learning methods, the accuracy and robustness of the crack detection algorithm are further improved, and it is suitable for road environments of different types and complexities. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Provides an overall method flow chart for an embodiment of the present invention; Figure 2This is a flowchart of high-quality image processing; Figure 3 This is a flow chart of ground penetrating radar signal processing; Figure 4 This is a flow chart of the road crack detection model based on DQN. DETAILED DESCRIPTION

[0018] For a better understanding of the present invention, the present invention is further described below in conjunction with specific examples, but the content of the present invention is not limited to the following embodiments.

[0019] The present invention provides a road crack detection method based on a vehicle-mounted sensor, which includes the following steps: S1: The on-board sensors are installed on the front of the vehicle. The on-board sensors include high-definition cameras and ground-penetrating radars, which are used to obtain high-quality road surface images and road reflection wave related data.

[0020] In this example, a vehicle-mounted sensor consisting of an 8-megapixel monocular camera and a ground-penetrating radar with a detection distance of 30-500 cm is installed on the front end of the vehicle to ensure the high quality of the collected road data.

[0021] S2: Set the parameters of the high-definition camera and ground-penetrating radar according to the vehicle speed and the preset distance of the on-board sensor, and obtain high-quality road surface images and road reflection wave related data as the vehicle travels.

[0022] After the vehicle-mounted sensors are assembled, the parameters of the HD camera and ground-penetrating radar are set in this example to obtain the best field of view and data collection effect. The shooting area of ​​the HD camera is a single lane and the lane lines on both sides of the vehicle × 2.5m, the resolution is set to 1080p, the frame rate is set to 60 frames, and optical focus, lens correction and image overlap detection are performed to ensure that the vehicle can collect clear and high-quality images during driving, which is convenient for subsequent image preprocessing and extraction. The signal frequency of the ground-penetrating radar is 800MHz, and the detection depth is set to 2m to ensure that the sensor can receive a signal sequence with high signal strength and good signal quality. By obtaining high-quality road detection data, subsequent data processing is facilitated, and the accuracy and efficiency of road crack detection are enhanced.

[0023] S3: Construct a road crack detection database, including a high-definition image database and a reflected wave signal database. The collected high-definition road surface images are extracted to obtain sub-block images at each moment to form an image set, and the reflected wave data is converted into a digital signal sequence to form a data set.

[0024] This embodiment uses Pix4Dmapper software to extract the collected high-quality images, merge repeated ground features into a single, accurate image, improve the clarity and information content of the image, and combine them to form a high-quality image set. Through the reflection wave receiver, the received reflection signal is converted from the reflection wave form to the digital signal sequence form through a precise conversion process for subsequent signal processing and analysis. The obtained high-definition image data and reflection wave digital signal sequence data are integrated and stored, and the time, location, lane and other information collected for each piece of data are marked.

[0025] S4: performing data preprocessing, preprocessing the acquired image data, including steps such as denoising by using Gaussian filtering, cropping based on edge detection, feature extraction and normalization based on gray-level co-occurrence matrix, and similarly, preprocessing the reflected wave digital signal sequence, including steps such as filtering by using Butterworth bandpass filter, DC offset correction, background removal based on spectral subtraction, gain adjustment based on adaptive gain control algorithm, and data smoothing based on moving average filter; In this embodiment, for the collected data set, first, Gaussian filtering denoising is performed on the original image to eliminate the interference noise in the image, ensuring that the subsequent processing steps can accurately analyze the actual road conditions. Secondly, the denoised image is cropped to cut the image area into 608 pixels × 608 pixels of the road surface part, and the irrelevant areas are removed, thereby improving the processing efficiency and reducing the computational burden. Thirdly, a feature extraction operation is performed on the cropped image, and a texture analysis algorithm based on the gray-level co-occurrence matrix is ​​used to capture the key features of road cracks, such as texture, shape, and color. Through feature extraction, the road surface image is converted into a representation with richer information, providing strong support for subsequent machine learning and model training. Finally, the extracted features are normalized to ensure that they are within the same scale range to avoid model training instability caused by differences in the numerical ranges of different features.

[0026] Subsequently, the collected reflected wave signal set is filtered using a Butterworth bandpass filter to filter the digital signal, effectively removing high-frequency noise and low-frequency trends, and retaining the frequency components related to road cracks. Correction is performed by calculating and subtracting the mean of the signal to ensure a stable signal baseline. Subsequently, spectral subtraction is used to remove background noise, and the influence of fixed reflections in the surface layer is eliminated by comparing the average values ​​of multiple scanned signals, thereby highlighting important underground features. In response to the problem of signal strength weakening with depth, adaptive gain adjustment is applied according to the signal depth to enhance the detection capability of deep signals. Finally, the signal is smoothed by applying a moving average or Gaussian smoothing algorithm to reduce random noise and sharp changes to improve signal stability and readability.

[0027] The processed image data and signal data are converted into text formats or data formats such as txt and xml that can be recognized by Python language, integrated and stored. At the same time, real calibration data of road cracks obtained through indoor experiments are also carried out for verification during training and testing.

[0028] S5: Build a road crack recognition model to identify and calibrate the location of road cracks based on reinforcement learning, including crack detection model training and crack detection testing, to complete the detection and calibration of road cracks.

[0029] In this embodiment, the road crack detection problem based on reinforcement learning is mainly divided into two parts: training and testing. The road crack detection problem is modeled as a reinforcement learning environment, and the state space, action space, reward function and learning objectives are defined. The state space is represented by the state consisting of camera images and reflected wave signal sequences. These states should contain enough information to describe the location, shape and other relevant features of the road cracks. The action space is the prediction of the location and size of the cracks. For the crack detection task, the reward function can be based on the difference between the crack location predicted by the model and the actual crack location, and the learning goal is determined so that the action selected in a given state can maximize the cumulative reward. Secondly, the DQN (Deep Q Network) algorithm is selected to train the road crack detection model. 100 sub-block images with road cracks and 50 sub-block images without road cracks are randomly selected from the collected data set, as well as 100 segments of reflected wave digital signal sequences with road cracks and 50 segments of reflected wave digital signal sequences without road cracks for model training, and the optimal strategy is learned through interaction with the environment, thereby realizing the detection and calibration of road cracks. Subsequently, the model is tested to evaluate its performance. The accuracy and robustness of the model are evaluated by running the model in a simulation environment based on a real experimental scenario and comparing its predicted results with the location and size of the actual cracks. Based on the test results, the model is optimized, including adjusting the model architecture, hyperparameter tuning, and increasing training data.

[0030] The trained model is deployed on the preprocessed vehicle sensor data set collected in step S4 to achieve the detection and calibration of road cracks. The test inputs 50 sub-block images with road cracks and 50 sub-block images without road cracks, as well as 50 reflected wave digital signal sequences with road cracks and 50 reflected wave digital signal sequences without road cracks into the road crack detection model to determine whether there are cracks and the type of corresponding road cracks, where the road crack types are divided into four types: cracks, block cracks, transverse cracks and longitudinal cracks. Then the model is used to roughly locate and mark the location of the damaged road surface, and the model output is post-processed to filter false detections, merge adjacent cracks, etc. The calibration of road cracks includes determining the specific location of the cracks on the road according to the detection results, and combining the geographic information system (GIS) or other positioning systems to calibrate the type, location, length, width and depth of the cracks. The reflected wave digital signal sequence detected at the location of the road cracks can be used as secondary inspection data to verify the crack detection effect. The specific information of the calibration frame is obtained through the conversion algorithm, and its specific calculation formula is as follows: In the formula, ( x 1, y 1) and ( x 2, y 2) are the coordinates of the upper left and lower right corners of the marked bounding box, img_w and img_h are the width and height of the image, respectively. x and y is the center coordinate of the bounding box obtained by the road crack recognition model, w and h are the width and height of the bounding box.

[0031] After completing the training and testing of the road crack recognition model, the remaining collected data is input into the model, and the model can autonomously identify and calibrate road cracks.

Claims

1. A road crack detection method based on a vehicle-mounted sensor, characterized in that: The steps include: S1: Install the on-board sensor on the front of the vehicle to obtain high-quality road surface images and road reflection wave related data; S2: according to the vehicle speed and the preset distance of the vehicle sensor, the vehicle sensor parameters are set, and high-quality road surface images and road reflection wave related data are obtained as the vehicle travels; S3: Construct a road crack detection database, including a high-definition image database and a reflected wave signal database. The collected high-definition road surface images are extracted to obtain sub-block images at each moment to form an image set, and the reflected wave data is converted into a digital signal sequence to form a data set. S4: perform data preprocessing; S5: Build a road crack recognition model to complete the detection and calibration of road cracks.

2. A road crack detection method based on a vehicle-mounted sensor according to claim 1, characterized in that: In step S1, the on-board sensor mounted on the front end of the vehicle is composed of a monocular camera with more than 8 million pixels and a ground-penetrating radar with a detection distance of 30-500 cm, which ensures the high quality of the collected road data. Both the high-definition camera and the ground-penetrating radar can accurately identify the road surface status, facilitating the subsequent identification and calibration of road cracks.

3. A road crack detection method based on a vehicle-mounted sensor according to claim 2, characterized in that: In the step S2, setting the vehicle-mounted sensor parameters includes setting the parameters of the high-definition camera and the ground-penetrating radar to ensure that after the vehicle-mounted sensor is installed on the vehicle, a sufficiently clear image with a suitable size and a reflected wave signal with a high signal-to-noise ratio and good stability can be obtained, wherein the setting of the high-definition camera includes shooting area confirmation, optical focus, lens correction and image overlap detection to ensure that the vehicle can collect clear and high-quality images during driving, which is convenient for subsequent preprocessing and extraction of the image; the setting of the ground-penetrating radar includes signal frequency, sampling interval, sampling duration and scanning depth to ensure that the sensor can receive a signal sequence with high signal strength and good signal quality; by obtaining high-quality road detection data, subsequent data processing is facilitated, and the accuracy and efficiency of road crack detection are enhanced.

4. A road crack detection method based on a vehicle-mounted sensor according to claim 3, characterized in that: In step S3, the high-definition images captured by the high-definition camera are extracted using Pix4Dmapper software, and repeated ground features are merged into a single, accurate image to improve the clarity and information content of the image, and are combined to form a high-quality image set, so that the extraction of ground features of the road surface and its surrounding environment will be more accurate and reliable, providing reliable data support for subsequent road crack detection.

5. A road crack detection method based on a vehicle-mounted sensor according to claim 4, characterized in that: In step S3, for the reflected wave signal obtained by the ground penetrating radar, the continuous analog signal is converted into a reflected wave digital signal sequence through an analog-to-digital converter, and stored to form a reflected wave signal set, so as to better utilize digital technology to process, transmit and store the obtained reflected wave signal, thereby improving the reliability, accuracy and applicability of the signal, and providing more accurate and reliable information for subsequent road crack detection.

6. A road crack detection method based on a vehicle-mounted sensor according to claim 1, characterized in that: In step S4, for the high-quality image data set, first, the original image is subjected to Gaussian filtering denoising processing to eliminate interference noise in the image, so as to ensure that the subsequent processing steps can accurately analyze the actual road conditions; secondly, the denoised image is cropped to cut the image area into 608 pixels × 608 pixels of road surface parts, and irrelevant areas are removed, thereby improving processing efficiency and reducing computational burden; thirdly, a feature extraction operation is performed on the cropped image, and a texture analysis algorithm based on a gray-level co-occurrence matrix is ​​used to capture key features of road cracks, including texture, shape and color; finally, the extracted features are normalized to ensure that they are within the same scale range, so as to avoid model training instability caused by differences in the numerical ranges of different features.

7. A road crack detection method based on a vehicle-mounted sensor according to claim 6, characterized in that: In step S4, for the collected reflected wave signal set, first, a Butterworth bandpass filter is used to filter the digital signal to effectively remove high-frequency noise and low-frequency trends, and retain the frequency components related to road cracks; secondly, for the DC offset of the signal, correction is performed by calculating and subtracting the mean of the signal to ensure the stability of the signal baseline. Thirdly, spectral subtraction is used to remove background noise, and the influence of fixed reflections of the surface layer is eliminated by comparing the average values ​​of multiple scanning signals, thereby highlighting important underground features; in addition, in order to address the problem of signal strength weakening with depth, adaptive gain adjustment is applied according to the signal depth to enhance the detection capability of deep signals; finally, the signal is smoothed by applying a moving average or Gaussian smoothing algorithm to reduce random noise and sharp changes, so as to improve the stability and readability of the signal.

8. The road crack detection method based on vehicle-mounted sensors according to claim 1, characterized in that: In step S5, the construction of the road crack recognition model includes crack detection model training and crack detection testing: first, the road crack detection problem is modeled as a reinforcement learning environment, and the state space, action space, reward function and learning objectives are defined; the state space is represented by states consisting of camera images and reflected wave signal sequences; these states should contain sufficient information to describe the location, shape and other relevant features of the road cracks; The action space is a prediction of the location and size of the crack. For the crack detection task, the reward function can be based on the difference between the crack location predicted by the model and the actual crack location, and the learning objective is determined so that the action selected in a given state can maximize the cumulative reward. Secondly, the DQN algorithm is selected to train the road crack detection model, and the optimal strategy is learned through interaction with the environment, thereby realizing the detection and calibration of road cracks; Thirdly, the model is tested to evaluate its performance. The accuracy and robustness of the model are evaluated by running the model in a simulated environment based on a real experimental scenario and comparing its predicted results with the location and size of the actual cracks. Based on the test results, the model is optimized, including adjusting the model architecture, hyperparameter tuning, and increasing training data. Finally, the trained model is deployed on the preprocessed vehicle sensor data set in step S4 to achieve detection and calibration of road cracks.

9. A road crack detection method based on a vehicle-mounted sensor according to claim 8, characterized in that: In step S5, the crack detection test includes inputting the preprocessed image and reflected wave data into the detection model, outputting the position and shape of the crack, and post-processing the model output to filter out false detections and merge adjacent cracks; the road crack calibration includes determining the specific position of the crack on the road based on the detection results, and calibrating the type, position, length, width and depth of the crack in combination with a geographic information system (GIS) or other positioning system.

Images