Road pavement disease detection method, device and system

Abstract

The invention relates to the technical field of image detection, in particular to a road pavement disease detection method, device and system. According to the method, visual information and sensor information are subjected to fusion analysis, theoretical rolling data when a detection vehicle passes through a transverse crack are obtained, the time and vertical acceleration of the detection vehicle in the actual rolling process are further collected, and actual rolling data are obtained. On the basis of difference information fed back by theoretical rolling data and actual rolling data, a unit delay time impact index of the vehicle during rolling can be analyzed and detected, and a data difference corresponding to an optimal matching result between an actual rolling data sequence and a theoretical rolling data sequence is obtained in a matching mode. And judging the type of the transverse crack according to the data difference, the signal form difference and the unit delay time impact index. According to the method, the visual information and the sensor information are combined, the impact response when the vehicle passes through the transverse crack is analyzed and detected, and then the bridge expansion joint and the transverse defect crack are accurately distinguished.

Description

Technical Field

[0001] This invention relates to the field of image detection technology, and specifically to a method, apparatus, and system for detecting road surface defects. Background Technology

[0002] Automated road inspection systems commonly employ vehicle-mounted high-definition cameras, combining image processing and deep learning models to identify road surface defects such as cracks. The underlying technology lies in the algorithm's ability to learn and summarize visual patterns of defects from massive amounts of image data. However, this paradigm faces an inherent bottleneck when dealing with a common road scenario: when detecting vehicles crossing bridges, the system is prone to misclassifying functional bridge expansion joints as transverse cracks. The root of this problem is that, in the dimension of two-dimensional images, the linear shape, color contrast, and edge texture of bridge expansion joints are highly similar to certain transverse cracks. Algorithms relying solely on visual patterns lack the information dimension needed to distinguish the essential differences between the two, making effective differentiation difficult. Existing improvements, such as introducing 3D laser scanning technology, can assist in differentiation by acquiring road surface elevation information, but its high equipment cost and complex data processing flow limit its large-scale application. Furthermore, methods that continuously optimize two-dimensional visual models have reached their limit in terms of classification accuracy because they fail to fundamentally introduce new discriminative dimensions. These frequent false alarms not only increase the burden of subsequent manual review, but also reduce the efficiency and data quality of the entire automated detection system. Summary of the Invention

[0003] To address the technical problem of existing technologies in effectively and cost-efficiently distinguishing between transverse cracks and bridge expansion joints during road pavement distress detection, the present invention aims to provide a method, apparatus, and system for road pavement distress detection. The specific technical solution adopted is as follows: This invention proposes a method for detecting road surface defects, the method comprising: The system acquires images of transverse cracks collected by the inspection vehicle during road inspection; predicts the theoretical compaction data sequence of the inspection vehicle over the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; and obtains the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle. The theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration. The vehicle suspension response delay is obtained based on the time difference between the actual compaction data sequence and the theoretical compaction data sequence; the impact index per unit delay time is obtained based on the vertical acceleration under the actual compaction time and the vehicle suspension response delay. Obtain the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; match the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result; The transverse crack type is determined based on the data differences, the signal morphology differences, and the impact index per unit delay time.

[0004] Furthermore, the method for identifying transverse cracks in the transverse crack image includes: A target detection neural network is used to identify a transverse crack image in a detection image. The transverse crack image contains a target region detected by the target detection neural network. After performing a bird's-eye view transformation on the transverse crack image, the local minimum brightness points in the target region are taken as transverse crack pixels, and the transverse crack pixels constitute the transverse crack.

[0005] Furthermore, the method for obtaining the actual compaction data sequence includes: The peak and trough points of the vertical acceleration sequence are identified using a peak detection algorithm. The time corresponding to the peak and trough points and the absolute value of the vertical acceleration are used to construct the corresponding actual compaction data to obtain the actual compaction data sequence.

[0006] Furthermore, the method for obtaining the vehicle suspension response delay includes: For each theoretical compaction data point in the theoretical compaction data sequence, the theoretical compaction time is multiplied by the brightness of the corresponding theoretical compaction location to obtain the weighted theoretical compaction time; the sum of all weighted theoretical compaction times is used as the numerator, and the sum of all theoretical compaction location brightness is used as the denominator to obtain the theoretical compaction time center. For each actual compaction data point in the actual compaction data sequence, the actual compaction time is multiplied by the corresponding vertical acceleration to obtain the weighted actual compaction time; the sum of all weighted actual compaction times is used as the numerator, and the sum of the corresponding vertical accelerations is used as the denominator to obtain the center of actual compaction time. The difference between the actual compaction time center and the theoretical compaction time center is used as the vehicle suspension response delay.

[0007] Furthermore, the method for obtaining the unit delay time impact index includes: The total impact energy is obtained by taking the sum of squares of vertical accelerations in the actual compaction data sequence and dividing the total impact energy by the vehicle suspension response delay.

[0008] Furthermore, the method for obtaining the signal morphology difference includes: Obtain the skewness difference and kurtosis difference between the weighted theoretical compaction time series and the weighted actual compaction time series; use the sum of the skewness difference and the kurtosis difference as the signal morphology difference.

[0009] Furthermore, the method for obtaining the optimal matching result includes: A cost matrix is ​​obtained by combining data points from the actual compaction time series and the theoretical compaction time series. Each element in the cost matrix represents the cost between two data points. The method for obtaining the cost includes: for each element, adding the theoretical compaction time of the corresponding data point to the vehicle suspension response delay to obtain the predicted time point; using the difference between the predicted time point and the actual compaction time corresponding to the data point as the time scale difference; adding the difference between the normalized theoretical compaction position brightness and the normalized vertical acceleration between the corresponding data points to a positive integer 1 to obtain the weighted scale difference; and using the product of the time scale difference and the weighted scale difference as the cost. The optimal matching result is obtained by solving the cost matrix using the Hungarian algorithm; the sum of costs corresponding to the optimal matching result is the data difference.

[0010] Further, the step of determining the transverse crack type based on the data difference, the signal morphology difference, and the impact index per unit delay time includes: The product of the data difference and the signal morphology difference is used as the geometric response matching error. If the geometric response matching error is greater than or equal to the preset error threshold, it is judged as a transverse defect crack. Otherwise, it is judged whether the unit delay time impact index is greater than the preset impact index threshold. If it is greater than the preset impact index threshold, it is judged as a candidate expansion joint. If it is less than or equal to the preset impact index threshold, it is judged as a transverse defect crack. The distance between the candidate expansion joint and the previously detected bridge expansion joint is obtained, and the candidate expansion joint is determined to be a bridge expansion joint based on the distance.

[0011] The present invention also proposes a road pavement distress detection device, the device comprising: The road surface inspection data acquisition module is used to acquire images of transverse cracks collected by the inspection vehicle during road inspection; predict the theoretical compaction data sequence of the inspection vehicle for the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; and obtain the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle. The theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration. The delayed impact analysis module is used to obtain the vehicle suspension response delay based on the time difference between the actual rolling data sequence and the theoretical rolling data sequence; and to obtain the impact index per unit delay time based on the vertical acceleration under the actual rolling time and the vehicle suspension response delay. The delay difference analysis module obtains the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; it matches the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result; The crack category determination module is used to determine the category of transverse cracks based on the data differences, the signal morphology differences, and the impact index per unit delay time.

[0012] The present invention also proposes a road surface defect detection system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any one of the road surface defect detection methods described above.

[0013] The present invention has the following beneficial effects: This invention does not employ a purely visual method but fuses and analyzes visual and sensor information. Considering that transverse cracks and bridge expansion joints exhibit different impact responses to vehicles, this invention statistically integrates the temporal and visual information from the image feedback to obtain theoretical compaction data when a vehicle passes over a transverse crack. Furthermore, it collects the actual compaction time and vertical acceleration during the actual compaction process to obtain actual compaction data. Based on the differences between the theoretical and actual compaction data, the impact index per unit delay time during compaction can be analyzed. This index assesses the impact response of the transverse crack to the vehicle and serves as a key feature for distinguishing bridge expansion joints from transverse cracks. Furthermore, considering that a structurally regular bridge expansion joint will exhibit a high degree of consistency in its visual morphology and dynamic response, while an irregular crack cannot satisfy this consistency, this invention further obtains the data difference corresponding to the optimal matching result between the actual and theoretical compaction data sequences through a matching method. The data difference and the signal morphology difference between the two sequences are used as key features for distinguishing bridge expansion joints from transverse cracks. Furthermore, the type of transverse crack can be determined based on data differences, signal morphology differences, and the impact index per unit delay time. This invention combines visual and sensor information to analyze and detect the impact response of vehicles passing over transverse cracks, thereby accurately distinguishing between bridge expansion joints and transverse defect cracks. Attached Figure Description

[0014] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a road surface defect detection method provided in one embodiment of the present invention. Detailed Implementation

[0016] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a road pavement distress detection method, apparatus, and system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0018] The following description, in conjunction with the accompanying drawings, details the specific scheme of a road surface defect detection method, device, and system provided by the present invention.

[0019] Please see Figure 1 The diagram illustrates a flowchart of a road surface distress detection method according to an embodiment of the present invention, the method comprising: Step S1: Obtain images of transverse cracks collected by the inspection vehicle during road inspection; predict the theoretical compaction data sequence of the inspection vehicle for the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; obtain the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle; the theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration.

[0020] In this embodiment of the invention, a high-definition camera for image acquisition and detection is deployed on the top of the detection vehicle, and an inertial measurement unit is deployed on the front axle of the vehicle. The detection vehicle moves forward at a constant speed. The high-definition camera acquires images of the transverse crack in real time, and the inertial measurement unit acquires vertical acceleration information in real time.

[0021] Preferably, in this embodiment of the invention, the method for identifying transverse cracks in a transverse crack image includes: A target detection neural network is used to identify transverse crack images in a detection image, wherein the transverse crack image contains the target region detected by the target detection neural network. Considering that the image captured by the camera has perspective distortion and cannot directly reflect the true transverse geometry of the target, and that the structural information of the target is mainly reflected in the grooves or edges where the brightness changes drastically, which will appear as obvious dark areas in the image brightness dimension, after performing a bird's-eye view transformation on the transverse crack image, the local minimum brightness points in the target region are taken as transverse crack pixels, and the transverse crack pixels constitute the transverse crack.

[0022] In this embodiment of the invention, the target detection network may be a YOLO network; the bird's-eye view transformation is an image geometric correction method known in the art for eliminating perspective distortion, thereby generating a top view that is linearly proportional to the real-world coordinates in size; the local minima selection method, YOLO network, bird's-eye view transformation, etc. are all technical means well known to those skilled in the art, and the specific process will not be described in detail.

[0023] It should be noted that the image acquisition process for the detected vehicle in this embodiment of the invention is a real-time, continuous process. After identifying the transverse crack image, the acquisition time point is recorded before crack category identification is performed. Simultaneously, the vehicle speed needs to be recorded, and the distance between the transverse crack and the detected vehicle needs to be obtained based on the transverse crack image, thereby obtaining the theoretical crushing time when the detected vehicle runs over the transverse crack. It should be noted that, since the transverse crack has width, it is necessary to calculate the theoretical crushing time for all transverse crack pixels involved in the vehicle's direction of travel, thereby obtaining the theoretical crushing time corresponding to all theoretical crushing positions. For example, the time for the detected vehicle to acquire the transverse crack image is... The time at which the vehicle theoretically begins to run over the transverse crack is then detected. for ,in for The distance between the vehicle and the nearest point of the transverse crack at any given time. Let be the vehicle speed. Then, for the i-th pixel in the direction of vehicle travel that detects a transverse crack, the theoretical compaction time is... for ,in Let be the distance between the i-th transverse crack pixel and the first crushing position of the vehicle. Further considering that the smaller the brightness of the transverse crack pixel, the greater the crack depth at the corresponding location, the brightness value at each theoretical crushing position can be obtained. This brightness value can be used as crack weight information under the theoretical crushing time. Finally, each transverse crack pixel detected in the vehicle's travel direction is considered a theoretical crushing position. Each theoretical crushing position corresponds to a theoretical crushing time and a brightness value in the transverse crack image. All the above parameters can be statistically obtained using existing image analysis algorithms, and each physical quantity can be adaptively transformed using existing coordinate system calibration methods, which will not be elaborated upon here.

[0024] For actual compaction data, this embodiment of the invention obtains the data through the variation characteristics of vertical acceleration. When driving on a smooth road surface, the vehicle's vertical acceleration signal should also be stable and without drastic fluctuations. However, when compacting over cracks, the vehicle experiences significant shaking due to bumps, leading to drastic changes in vertical acceleration. Therefore, this embodiment of the invention can obtain the actual compaction data sequence based on the vertical acceleration sequence.

[0025] It should be noted that, in order to avoid unnecessary data calculations, the embodiments of the present invention can collect and calculate vertical acceleration for a period of time after detecting the transverse crack image. The specific settings can be made according to the actual implementation scenario, and will not be elaborated or limited here.

[0026] Preferably, in this embodiment of the invention, the method for obtaining the actual compaction data sequence includes: The peak and trough points of the vertical acceleration sequence are identified using a peak detection algorithm. The time corresponding to the peak and trough points and the absolute value of the vertical acceleration are used to construct the corresponding actual compaction data to obtain the actual compaction data sequence.

[0027] It should be noted that the peak detection algorithm is a well-known technique in the art. In this embodiment of the invention, a threshold screening process is further built into the peak detection algorithm, and the peak and trough points whose absolute amplitude values ​​of the initially detected peak points are greater than the preset amplitude threshold are taken as the final detection points. The amplitude threshold can be set to 0.5, and the unit is the unit of acceleration.

[0028] It should be noted that, in order to facilitate the joint calculation of subsequent data, the data obtained in step S1 can be considered only in terms of numerical value, without considering the unit. That is, in the process of data transmission and storage, only the numerical value of the data is stored, so that subsequent steps can directly perform numerical calculations.

[0029] Step S2: Obtain the vehicle suspension response delay based on the time difference between the actual compaction data sequence and the theoretical compaction data sequence; obtain the impact index per unit delay time based on the vertical acceleration under the actual compaction time and the vehicle suspension response delay.

[0030] Because the transmission and absorption of force by a vehicle's suspension system upon impact is not instantaneous, but involves a response process related to the obstacle's shape and vehicle speed, the impact center actually perceived by the vehicle body, as measured by the inertial measurement unit, has a time lag compared to the theoretical impact center at the wheel's initial contact. Furthermore, the impact energy transmitted by a structurally robust target (such as a steel expansion joint) differs significantly from that of a loosely structured target (such as an asphalt crack) within the same response time. Therefore, by accurately calculating this time lag and correlating it with the total impact energy, two macroscopic dynamic features reflecting the vehicle's response mode and the target's equivalent impact intensity can be decoupled and extracted. These features can serve as important distinguishing characteristics between bridge expansion joints and transverse pavement cracks. Since bridge expansion joints are steel structures, they exhibit a stronger impact index per unit time delay; while transverse pavement cracks, being asphalt cracks, have a relatively loose structure and less severe morphology, resulting in a weaker impact index per unit time delay.

[0031] Since the actual compaction data sequence and the theoretical compaction data sequence already contain theoretical compaction time information and actual compaction time information, the vehicle suspension response delay can be obtained by comparing the time difference between the two types of information. This feature is characterized as a time quantity, representing the delay time caused by the suspension system.

[0032] By further analyzing the vertical acceleration during actual rolling time and combining it with the vehicle's suspension response delay, the impact index per unit delay time can be obtained. The stronger this index, the greater the impact the vehicle experiences per unit delay time.

[0033] Preferably, in this embodiment of the invention, the method for obtaining the vehicle suspension response delay includes: For each theoretical compaction data point in the theoretical compaction data sequence, the theoretical compaction time is multiplied by the brightness of the corresponding theoretical compaction location to obtain the weighted theoretical compaction time; the sum of all weighted theoretical compaction times is used as the numerator, and the sum of all theoretical compaction location brightness is used as the denominator to obtain the theoretical compaction time center. For each actual compaction data point in the actual compaction data sequence, the actual compaction time is multiplied by the corresponding vertical acceleration to obtain the weighted actual compaction time; the sum of all weighted actual compaction times is used as the numerator, and the sum of the corresponding vertical accelerations is used as the denominator to obtain the center of actual compaction time. The difference between the actual compaction time center and the theoretical compaction time center is used as the vehicle suspension response delay.

[0034] It should be noted that, due to the different statistical methods used, the theoretical compaction time and the actual compaction time do not correspond in terms of quantity and position in the time series. Therefore, the process of obtaining the theoretical compaction time center and the actual compaction time center in the above process is a process of weighted analysis under their respective dimensions. The denominator is the sum of the corresponding weights. The final time center is the comprehensive time result with the strongest response under their respective dimensions. Therefore, it can be used as a reference time to obtain the vehicle suspension response delay.

[0035] Preferably, in this embodiment of the invention, the method for obtaining the impact index per unit delay time includes: The total impact energy is obtained by taking the sum of squares of vertical accelerations in the actual compaction data sequence and dividing the total impact energy by the vehicle suspension response delay.

[0036] Step S3: Obtain the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; match the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result.

[0037] This invention further considers that bridge expansion joints, as a standard industrial component, exhibit high symmetry and concentration in their geometric structure and the resulting impact response over time, meaning they display a sharp impact pattern in their signal morphology. Naturally formed cracks, on the other hand, exhibit more randomness and asymmetry. Therefore, to effectively distinguish bridge expansion joints from transverse defect cracks, the signal morphology difference between actual and theoretical compaction data sequences is obtained. A greater difference in signal morphology indicates stronger randomness in the impact generated by the transverse crack, making it more likely to be a transverse defect crack.

[0038] Preferably, in this embodiment of the invention, the skewness and kurtosis of the signal are used as signal shape features, and the method for obtaining signal shape differences includes: The skewness difference and kurtosis difference between the weighted theoretical compaction time series and the weighted actual compaction time series are obtained; the sum of the skewness difference and the kurtosis difference is taken as the signal morphology difference. The methods for obtaining skewness and kurtosis are well-known techniques to those skilled in the art and will not be elaborated or limited here.

[0039] This invention further considers that, for a real, single physical target, its visual characteristics and impact response should not only correspond in time and space, but also be highly coupled in terms of intensity (e.g., deeper cracks correspond to stronger impacts) and structural morphology (e.g., symmetrical visual morphology corresponds to symmetrical impact responses). Therefore, this invention uses a matching method to match data points between actual compaction data sequences and theoretical compaction data sequences to find the optimal matching result. Then, it calculates the data difference under the optimal matching result; that is, the greater the data difference, the lower the coupling degree, and the greater the probability that the transverse crack belongs to a transverse defect crack.

[0040] Preferably, in this embodiment of the invention, the method for obtaining the optimal matching result includes: The cost matrix is ​​obtained by combining the data points between the actual compaction time series and the theoretical compaction time series, where each element in the cost matrix represents the cost between two data points.

[0041] The methods for obtaining the cost include: (1) For each element, due to the delay caused by the suspension system between the theoretical compaction time and the actual compaction time, when calculating the difference, it is necessary to add the theoretical compaction time of the corresponding data point to the vehicle suspension response delay to obtain the predicted time point and achieve time alignment. The difference between the predicted time point and the actual compaction time corresponding to the data point is taken as the time scale difference.

[0042] (2) Further compare the weight information between data points, add the absolute value of the difference between the normalized theoretical crushing position brightness and the normalized vertical acceleration between the corresponding data points to the positive integer 1 to obtain the weight scale difference. That is, the weight scale difference is at least 1. The greater the difference between weights, the more mismatched the data points are, and the greater the weight scale difference.

[0043] (3) The product of the time scale difference and the weight scale difference is used as the cost.

[0044] The optimal matching result is obtained by solving the cost matrix using the Hungarian algorithm; the sum of costs corresponding to the optimal matching result is the data difference.

[0045] Step S4: Determine the type of transverse crack based on data differences, signal morphology differences, and impact index per unit delay time.

[0046] By statistically analyzing the data differences, signal morphology differences, and impact index per unit delay time obtained in the above steps, the current type of transverse crack can be identified. Bridge expansion joints, being steel structures, exhibit smaller data differences, smaller signal morphology differences, and larger impact per unit delay time. In contrast, transverse cracks show the opposite characteristics, exhibiting larger data differences, larger signal morphology differences, and smaller impact per unit delay time. Therefore, transverse crack types can be accurately and effectively distinguished based on these characteristics.

[0047] Preferably, in this embodiment of the invention, a three-level discrimination process is designed based on the data difference, the signal morphology difference, and the impact index per unit delay time to determine the type of transverse crack, specifically including: (1) First level: The product of the data difference and the signal morphology difference is taken as the geometric response matching error; if the geometric response matching error is greater than or equal to the preset error threshold, it is judged as a transverse defect crack; otherwise, it enters the second level of discrimination process. In this embodiment of the invention, the error threshold is set to 0.05. If it is less than the error threshold, it means that the observed impact response pattern can be well predicted by the visual geometry of the target through kinematic principles, which is consistent with the characteristics of the engineering component of the bridge expansion joint; if it is greater than or equal to the error threshold, it means that the morphology of the transverse crack is irregular and should be a transverse defect crack.

[0048] (2) Second level: Determine whether the impact index per unit delay time is greater than the preset impact index threshold. If it is greater than the preset impact index threshold, it is judged as a candidate expansion joint and enters the third level judgment process. If it is less than or equal to the preset impact index threshold, it is judged as a transverse defect crack. It should be noted that the impact index threshold can be specifically set based on parameters such as vehicle suspension parameters and road surface conditions. This embodiment of the invention will not elaborate or limit it.

[0049] (3) Third level: Since the design and construction of bridge engineering follow the standardization principle, the layout of multiple expansion joints on a bridge in the longitudinal direction of the road usually follows a periodic pattern of equal spacing, while the appearance of transverse cracks is random and irregular. Therefore, by analyzing the distribution pattern of the target group initially identified as candidate expansion joints in the spatial sequence, a verification layer derived from prior engineering knowledge can be introduced to confirm the correct identification results and correct accidental misjudgments. That is, the distance between the candidate expansion joint and the previously detected bridge expansion joints is obtained, and the candidate expansion joint is judged as a bridge expansion joint based on the distance. In the embodiment of the present invention, the distance between the candidate expansion joint and all other adjacent objects in the bridge expansion joints within a range of one kilometer before and after is counted. If the difference between the distance between the current candidate expansion joint and the adjacent bridge expansion joint and the average distance is greater than the preset distance difference threshold, it indicates that the current candidate expansion joint is isolated and abnormal in space, and is considered as a transverse defect crack; otherwise, it indicates that the current candidate expansion joint can belong to a local sequence with high periodicity, indicating that it is a similar object among the previous bridge expansion joints.

[0050] In summary, this invention fuses and analyzes visual and sensor information to obtain theoretical compaction data when a vehicle passes over a transverse crack. It further collects the actual compaction time and vertical acceleration during the actual compaction process to obtain actual compaction data. Based on the differences between the theoretical and actual compaction data, the impact index per unit delay time during compaction can be analyzed. The optimal matching result between the actual and theoretical compaction data sequences is obtained through matching to identify the data difference. The transverse crack type is determined based on the data differences, signal morphology differences, and the impact index per unit delay time. This invention combines visual and sensor information to analyze the impact response of a vehicle passing over a transverse crack, thereby accurately distinguishing between bridge expansion joints and transverse defect cracks.

[0051] Based on the same inventive concept, the present invention also proposes a road surface distress detection device, the device comprising: The road surface inspection data acquisition module is used to acquire images of transverse cracks collected by the inspection vehicle during road inspection; predict the theoretical compaction data sequence of the inspection vehicle for the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; and obtain the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle. The theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration. The delayed impact analysis module is used to obtain the vehicle suspension response delay based on the time difference between the actual rolling data sequence and the theoretical rolling data sequence; and to obtain the impact index per unit delay time based on the vertical acceleration under the actual rolling time and the vehicle suspension response delay. The delay difference analysis module obtains the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; it matches the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result; The crack category determination module is used to determine the category of transverse cracks based on the data differences, the signal morphology differences, and the impact index per unit delay time.

[0052] The present invention also proposes a road surface defect detection system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any one of the road surface defect detection methods described above.

[0053] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for detecting road surface defects, characterized in that, The method includes: The system acquires images of transverse cracks collected by the inspection vehicle during road inspection; predicts the theoretical compaction data sequence of the inspection vehicle over the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; and obtains the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle. The theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration. The vehicle suspension response delay is obtained based on the time difference between the actual compaction data sequence and the theoretical compaction data sequence; the impact index per unit delay time is obtained based on the vertical acceleration under the actual compaction time and the vehicle suspension response delay. Obtain the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; match the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result; The transverse crack type is determined based on the data differences, the signal morphology differences, and the impact index per unit delay time.

2. The method for detecting road surface defects according to claim 1, characterized in that, The method for identifying transverse cracks in the transverse crack image includes: A target detection neural network is used to identify a transverse crack image in a detection image. The transverse crack image contains a target region detected by the target detection neural network. After performing a bird's-eye view transformation on the transverse crack image, the local minimum brightness points in the target region are taken as transverse crack pixels, and the transverse crack pixels constitute the transverse crack.

3. The method for detecting road surface defects according to claim 1, characterized in that, The method for obtaining the actual compaction data sequence includes: The peak and trough points of the vertical acceleration sequence are identified using a peak detection algorithm. The time corresponding to the peak and trough points and the absolute value of the vertical acceleration are used to construct the corresponding actual compaction data to obtain the actual compaction data sequence.

4. The method for detecting road surface defects according to claim 1, characterized in that, The method for obtaining the vehicle suspension response delay includes: For each theoretical compaction data point in the theoretical compaction data sequence, the theoretical compaction time is multiplied by the brightness of the corresponding theoretical compaction location to obtain the weighted theoretical compaction time; the sum of all weighted theoretical compaction times is used as the numerator, and the sum of all theoretical compaction location brightness is used as the denominator to obtain the theoretical compaction time center. For each actual compaction data point in the actual compaction data sequence, the actual compaction time is multiplied by the corresponding vertical acceleration to obtain the weighted actual compaction time; the sum of all weighted actual compaction times is used as the numerator, and the sum of the corresponding vertical accelerations is used as the denominator to obtain the center of actual compaction time. The difference between the actual compaction time center and the theoretical compaction time center is used as the vehicle suspension response delay.

5. The method for detecting road surface defects according to claim 1, characterized in that, The method for obtaining the impact index per unit delay time includes: The total impact energy is obtained by taking the sum of squares of vertical accelerations in the actual compaction data sequence and dividing the total impact energy by the vehicle suspension response delay.

6. The method for detecting road surface defects according to claim 4, characterized in that, The method for obtaining the signal morphology differences includes: Obtain the skewness difference and kurtosis difference between the weighted theoretical compaction time series and the weighted actual compaction time series; use the sum of the skewness difference and the kurtosis difference as the signal morphology difference.

7. The method for detecting road surface defects according to claim 1, characterized in that, The method for obtaining the optimal matching result includes: A cost matrix is ​​obtained by combining data points from the actual compaction time series and the theoretical compaction time series. Each element in the cost matrix represents the cost between two data points. The method for obtaining the cost includes: for each element, adding the theoretical compaction time of the corresponding data point to the vehicle suspension response delay to obtain the predicted time point; using the difference between the predicted time point and the actual compaction time corresponding to the data point as the time scale difference; adding the difference between the normalized theoretical compaction position brightness and the normalized vertical acceleration between the corresponding data points to a positive integer 1 to obtain the weighted scale difference; and using the product of the time scale difference and the weighted scale difference as the cost. The optimal matching result is obtained by solving the cost matrix using the Hungarian algorithm; the sum of costs corresponding to the optimal matching result is the data difference.

8. The method for detecting road surface defects according to claim 1, characterized in that, The step of determining the transverse crack type based on the data differences, the signal morphology differences, and the impact index per unit delay time includes: The product of the data difference and the signal morphology difference is used as the geometric response matching error. If the geometric response matching error is greater than or equal to the preset error threshold, it is judged as a transverse defect crack. Otherwise, it is judged whether the unit delay time impact index is greater than the preset impact index threshold. If it is greater than the preset impact index threshold, it is judged as a candidate expansion joint. If it is less than or equal to the preset impact index threshold, it is judged as a transverse defect crack. The distance between the candidate expansion joint and the previously detected bridge expansion joint is obtained, and the candidate expansion joint is determined to be a bridge expansion joint based on the distance.

9. A road surface defect detection device, characterized in that, The device includes: The road surface inspection data acquisition module is used to acquire images of transverse cracks collected by the inspection vehicle during road inspection; predict the theoretical compaction data sequence of the inspection vehicle for the transverse cracks based on the distance between the transverse cracks and the inspection vehicle; and obtain the actual compaction data sequence based on the vertical acceleration sequence on the inspection vehicle. The theoretical compaction data includes the theoretical compaction time and the brightness of the theoretical compaction position in the transverse crack image; the actual compaction data includes the actual compaction time and its corresponding vertical acceleration. The delayed impact analysis module is used to obtain the vehicle suspension response delay based on the time difference between the actual rolling data sequence and the theoretical rolling data sequence; and to obtain the impact index per unit delay time based on the vertical acceleration under the actual rolling time and the vehicle suspension response delay. The delay difference analysis module obtains the signal morphology difference between the actual compaction data sequence and the theoretical compaction data sequence; it matches the data points between the actual compaction data sequence and the theoretical compaction data sequence to obtain the data difference corresponding to the optimal matching result; The crack category determination module is used to determine the category of transverse cracks based on the data differences, the signal morphology differences, and the impact index per unit delay time.

10. A road surface defect detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the road surface defect detection method as described in any one of claims 1 to 8.

Images