Road surface deformation judgment method for identifying spikes based on artificial intelligence

Abstract

The invention discloses a road surface deformation judgment method for identifying spikes based on artificial intelligence, which comprises the following steps of: acquiring image data of initial positions and sizes of the spikes in a preset area, detecting the image data by utilizing artificial intelligence, and acquiring and recording initial pixel coordinates and size data of each spike; periodically collecting the image data, detecting the image data by utilizing artificial intelligence, and obtaining and recording current pixel coordinates and size data of each spike; and by calculating the variable quantity between the current pixel coordinate and size data and the initial pixel coordinate and size data, whether the spike has position offset or size change is judged.

Description

Technical Field

[0001] The present invention is applied to the field of artificial intelligence, and specifically relates to a method for judging pavement deformation based on artificial intelligence to identify road studs. Background Art

[0002] With the continuous growth of the highway mileage in China, the roadbed conditions passed by the highways are various, such as soft roadbed sections. The road conditions of these highways will be affected by the external environment and cause different degrees of pavement deformation or even damage. The pavement maintenance company conducts regular pavement inspections. Due to the long interval between inspections, when there are abnormalities on the road surface, the pavement conditions cannot be sensed in a timely manner. Only when a traffic accident occurs can it intervene and handle the situation afterwards. Similarly, the background monitor polls the road surface conditions through the camera and also cannot detect the minute pavement deformation in a timely manner. Only when the pavement undergoes a large degree of deformation and the polling time reaches the road section can the problem be discovered. The timeliness of problem discovery is relatively low. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method for judging pavement deformation based on artificial intelligence to identify road studs in view of the deficiencies of the prior art.

[0004] To solve the above technical problem, a method for judging pavement deformation based on artificial intelligence to identify road studs of the present invention at least includes the following steps:

[0005] Collect image data of the initial positions and sizes of road studs in a preset area, and use artificial intelligence to detect the image data to obtain the initial pixel coordinates and size data of each road stud;

[0006] Periodically collect the above image data, and use artificial intelligence to detect the image data to obtain the current pixel coordinates and size data of each road stud;

[0007] By calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data, judge whether the road studs have a position offset or a size change.

[0008] As a possible implementation manner, further, it further includes the following steps:

[0009] Output warning information of different levels according to the change amount.

[0010] As a possible implementation manner, further, the step of collecting image data of the initial positions and sizes of road studs in a preset area and using artificial intelligence to detect the image data to obtain the initial pixel coordinates and size data of each road stud includes:

[0011] Select a low - traffic period without vehicle occlusion and use a camera to collect static images or video frames of the covered area for recording the initial state of the road studs;

[0012] Eliminate noise through pre - processing and adjust the brightness and contrast of the image to enhance the edge clarity of the road studs;

[0013] Use a deep - learning object - detection model to automatically identify the positions of the road studs in the image. The input of the deep - learning object - detection model is the pre - processed road image, and the output is the rectangular detection box and size of each road stud, which serve as the initial state of the road studs.

[0014] As a possible implementation, further, the steps of periodically collecting the above - mentioned image data, using artificial intelligence to detect the image data, and obtaining the current pixel coordinates and size data of each road stud include:

[0015] Preset a detection period. Within each detection period, use a camera to collect static images or video frames of the covered area for recording the current state of the road studs;

[0016] Eliminate noise through pre - processing and adjust the brightness and contrast of the image to enhance the edge clarity of the road studs;

[0017] Use a deep - learning object - detection model to automatically identify the positions of the road studs in the image. The input of the deep - learning object - detection model is the pre - processed road image, and the output is the rectangular detection box and size of each road stud, which serve as the current state of the road studs.

[0018] As a possible implementation, further, the steps of judging whether the road studs have position offset or size change by calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data specifically include:

[0019] Quantify the position change of the road studs by calculating the intersection - over - union (IOU) of the two detection boxes. The IOU is defined as the ratio of the intersection area to the union area of the two detection boxes;

[0020] Divide the IOU values into multiple intervals, and preset multiple threshold intervals corresponding to different warning levels;

[0021] When the IOU is 1, no warning is triggered. The IOU within each interval range corresponds to low - to - high warning levels.

[0022] As a possible implementation, further, the steps of periodically collecting the above - mentioned image data, using artificial intelligence to detect the image data, and obtaining the current pixel coordinates and size data of each road stud also include:

[0023] In cycle detection, if the position of the spike is blocked and cannot be detected, set consecutive detection intervals. If the spike still cannot be detected, trigger the highest-level alarm;

[0024] If detection can be restored after a single occlusion, determine whether there are changes based on the detection data.

[0025] As a possible implementation, further, the step of determining whether the spike has a position offset or size change by calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data further includes:

[0026] When multiple spikes all have position offsets or size changes, accumulate the alarm levels of each spike and calculate the overall alarm level of the area;

[0027] When the detected number of spikes decreases, trigger a continuous alarm mechanism, set a multiple detection threshold to increase the alarm level until the highest-level alarm is triggered.

[0028] As a possible implementation, further, alarm output and handling: Output alarm information of different levels to the background monitoring platform according to the alarm level of the spike position change and the alarm situation of the decrease in the number of spikes; The background monitoring platform generates a warning based on the alarm information and notifies relevant users. The users make a secondary judgment in combination with the camera video image or dispatch personnel for on-site inspection; Execute a preset handling plan according to the alarm level and the result of the secondary judgment.

[0029] As a possible implementation, further, the image data is collected by a camera. The installation position of the camera is the median strip or the roadside of a two-way lane to ensure that all spikes within the camera coverage are within the detection range.

[0030] A pavement deformation judgment system for identifying spikes based on artificial intelligence

[0031] An initial data acquisition module collects image data of the initial positions and sizes of spikes in a preset area, uses artificial intelligence to detect the image data, and obtains the initial pixel coordinates and size data of each spike;

[0032] A current data acquisition module periodically collects the above image data, uses artificial intelligence to detect the image data, and obtains the current pixel coordinates and size data of each spike;

[0033] A change status calculation module determines whether the spike has a position offset or size change by calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data;

[0034] An alarm module outputs alarm information of different levels according to the change amount.

[0035] The present invention adopts the above technical solutions and has the following beneficial effects:

[0036] 1. Real-time monitoring and early warning: By regularly collecting road images and using artificial intelligence technology to automatically detect the displacement of road spikes, it is possible to discover tiny road surface deformations in real time and automatically, greatly improving the timeliness and accuracy of road surface inspections. Compared with traditional manual inspection methods, the present invention can issue alarms at the initial stage of problems, avoiding the limitation of discovering problems only after traffic accidents.

[0037] 2. High-precision positioning and dimension detection: Adopting high-resolution image processing technology based on cameras and combining deep learning object detection algorithms, it is possible to accurately locate the position and dimension changes of each road spike. Through an accurate IOU calculation method, it is possible to effectively quantify the displacement of road spikes and divide different alarm levels according to the degree of displacement, thereby providing detailed data support for road surface maintenance.

[0038] 3. Intelligent alarm system: The present invention automatically generates alarm information according to the displacement size, quantity, and disappearance of road spikes, and takes corresponding handling measures according to different alarm levels, reducing the need for manual intervention and improving the automation degree and emergency response speed of the system.

[0039] 4. Dynamic adaptation and fault tolerance: The system can effectively handle problems such as road spikes being blocked and misrecognized. Through multiple detections and trend analysis, the system can distinguish whether it is a real displacement or occlusion, reducing the false alarm rate and ensuring the accuracy and reliability of alarm information.

[0040] 5. Reducing maintenance costs: The present invention can effectively reduce the lag of road repair and maintenance by improving the discovery efficiency of road surface deformations, avoiding potential safety hazards and greater scope of repair needs caused by road surface deformations, thereby reducing maintenance costs and improving the safety and service life of roads. Specific implementation manners

[0041] To make the purposes, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0042] Example 1

[0043] The present invention provides a method for judging road surface deformation based on artificial intelligence recognition of road spikes, which at least includes the following steps:

[0044] Collect image data of the initial positions and dimensions of road spikes in a preset area, and use artificial intelligence to detect the image data to obtain the initial pixel coordinates and dimension data of each road spike;

[0045] Periodically collect the above image data, use artificial intelligence to detect the image data, and obtain the current pixel coordinates and size data of each spike;

[0046] By calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data, determine whether the spike has a position offset or a size change.

[0047] Output warning messages of different levels according to the change amount.

[0048] Among them, the steps of collecting the image data of the initial positions and sizes of the spikes in the preset area, using artificial intelligence to detect the image data, and obtaining the initial pixel coordinates and size data of each spike include:

[0049] Select a low-traffic period without vehicle occlusion, and use a camera to collect static images or video frames of the covered area for recording the initial state of the spikes;

[0050] Eliminate noise through preprocessing, and adjust the brightness and contrast of the image to enhance the edge sharpness of the spikes;

[0051] Use a deep learning object detection model to automatically identify the positions of the spikes in the image. The input of the deep learning object detection model is the preprocessed road image, and the output is the rectangular detection frame and size of each spike, which are used as the initial state of the spikes.

[0052] Among them, the steps of periodically collecting the above image data, using artificial intelligence to detect the image data, and obtaining the current pixel coordinates and size data of each spike include:

[0053] Preset a detection period. In each detection period, use a camera to collect static images or video frames of the covered area for recording the current state of the spikes;

[0054] Eliminate noise through preprocessing, and adjust the brightness and contrast of the image to enhance the edge sharpness of the spikes;

[0055] Use a deep learning object detection model to automatically identify the positions of the spikes in the image. The input of the deep learning object detection model is the preprocessed road image, and the output is the rectangular detection frame and size of each spike, which are used as the current state of the spikes.

[0056] Real-time monitoring and early warning: By regularly collecting road images and using artificial intelligence technology to automatically detect the displacement of spikes, small deformations of the road surface can be discovered in real time and automatically, greatly improving the timeliness and accuracy of road surface inspections. Compared with the traditional manual inspection method, the present invention can issue a warning at the initial stage of the problem, avoiding the limitation of discovering the problem only after a traffic accident.

[0057] Among them, the steps of judging whether the road stud has a position offset or size change by calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data specifically include:

[0058] By calculating the intersection over union (IOU) of the two detection frames, the position change of the road stud is quantified. The IOU is defined as the ratio of the intersection area to the union area of the two detection frames;

[0059] The IOU values are divided into multiple intervals, and multiple threshold intervals are preset corresponding to different alarm levels;

[0060] When the IOU is 1, no alarm is triggered. The IOU corresponds to alarm levels from low to high within each interval range.

[0061] Among them, the steps of periodically collecting the above image data, using artificial intelligence to detect the image data, and obtaining the current pixel coordinates and size data of each road stud also include:

[0062] In the periodic detection, if the position of the road stud is blocked and cannot be detected, a continuous detection interval is set. If the road stud still cannot be detected, the highest-level alarm is triggered;

[0063] If the detection can be restored after a single occlusion, it is judged whether there is a change according to the detection data.

[0064] High-precision positioning and size detection: By adopting the high-resolution image processing technology based on a camera and combining with the deep learning object detection algorithm, the position and size change of each road stud can be accurately located. Through the precise IOU calculation method, the displacement of the road stud can be effectively quantified, and different alarm levels are divided according to the displacement degree, so as to provide detailed data support for road maintenance.

[0065] Among them, the steps of judging whether the road stud has a position offset or size change by calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data also include:

[0066] When multiple road studs all have position offsets or size changes, the alarm levels of each road stud are accumulated to calculate the overall alarm level of the area;

[0067] When the number of detected road studs decreases, a continuous alarm mechanism is triggered, and the alarm level is increased by setting multiple detection thresholds until the highest-level alarm is triggered.

[0068] Dynamic adaptation and fault tolerance ability: The system can effectively handle problems such as road stud occlusion and misidentification. Through multiple detections and trend analysis, the system can distinguish whether it is a real displacement or occlusion, reduce the false alarm rate, and ensure the accuracy and reliability of the alarm information.

[0069] Among them, alarm output and handling: According to the alarm level of the spike position change and the alarm situation of the decrease in the number of spikes, different levels of alarm information are output to the background monitoring platform; the background monitoring platform generates a warning based on the alarm information and notifies relevant users. The users make a secondary judgment in combination with the camera video image or dispatch personnel for on-site inspection; according to the alarm level and the result of the secondary judgment, a preset handling plan is executed.

[0070] Intelligent alarm system: According to the displacement size, quantity and disappearance situation of the spikes, this invention automatically generates alarm information and takes corresponding handling measures according to different alarm levels, reducing the need for manual intervention and improving the automation degree and emergency response speed of the system.

[0071] Among them, the image data is collected by the camera, and the installation position of the camera is the median strip or the roadside of the two-way lane, ensuring that all spikes within the camera coverage range are within the detection range.

[0072] Road surface deformation judgment system based on artificial intelligence for identifying spikes

[0073] Initial data acquisition module: Collect the image data of the initial positions and sizes of the spikes in the preset area, and use artificial intelligence to detect the image data to obtain the initial pixel coordinates and size data of each spike.

[0074] Current data acquisition module: Periodically collect the above-mentioned image data, and use artificial intelligence to detect the image data to obtain the current pixel coordinates and size data of each spike.

[0075] Change status calculation module: By calculating the change amount between the current pixel coordinates and size data and the initial pixel coordinates and size data, judge whether the spikes have position offset or size change.

[0076] Alarm module: Output different levels of alarm information according to the change amount.

[0077] Embodiment 2

[0078] A method for judging road surface deformation based on artificial intelligence for identifying spikes is provided, and the difference from Embodiment 1 is that:

[0079] The installation height of the camera is 5 meters for ordinary roads and 12 meters for highways. Taking the highway as an example, the interval is 100 meters (inside the tunnel) to 120 meters (outside the tunnel).

[0080] Camera parameters are prioritized: Pixel: 4 million; Maximum resolution: 2688×1520; Minimum illumination: 0.001 Lux (color mode); 0.0001 Lux (black and white mode); 0 Lux (fill light on); Maximum fill light distance: 120 m (infrared); Lens type: motorized zoom; Lens focal length: 3.5 - 12 mm; Field of view: Horizontal: 114°~47°; Vertical: 62°~26°; Diagonal: 136°~54°;

[0081] The camera can be installed in the middle of the two-way lane or on the roadside according to the road conditions. Ensure that the road studs at the farthest end from the camera are within...

[0082] The actual displacement of the warning registration is associated with the displacement of the road studs. The larger the single displacement, the more severe the road deformation. At the same time, the more road studs with displacement, the more severe the overall road deformation. If the number of road studs decreases, an immediate warning is required.

[0083] Road studs appear as pixels on the camera lens. Due to the different distances of road studs from the lens, road studs with the same actual physical size will have fewer pixels when they are farther from the lens and more pixels when they are closer.

[0084] Therefore, to judge the displacement of a single road stud, the IOU data of the position of the road stud in the previous detection, the previous ten detections, and the detection at the same time period of the previous day compared with the current detection position of the road stud can be used. The definition of IOU is the intersection over union, that is, the quotient of the intersection area of the two detection frames divided by the union area of the detection frames. If they completely overlap, it is 1; if they are completely separated, it is 0. We define 0.9, 0.8, 0.7, 0.6, and use these four values as thresholds to divide five intervals of IOU values. The warning levels range from low to high as 1 - 5. Only when the IOU is 1, there is no warning. In other cases, it indicates displacements of different sizes.

[0085] When multiple road studs all have displacement warnings, it indicates that the overall road surface has deformed. The warning levels of multiple road studs are accumulated to obtain the final warning level of this area.

[0086] If it is found that the number of road studs decreases, for 2 consecutive detections, the warning level is 1; for 3 consecutive detections, the warning level is 1. This continues until the highest warning level.

[0087] After the above warnings occur, the user, based on the actual video situation seen, or dispatches personnel to the scene for secondary research and judgment, and processes according to the subsequent disposal plan.

[0088] Embodiment 3

[0089] A method for judging road surface deformation based on artificial intelligence recognition of road studs, including:

[0090] Initial deployment stage:

[0091] Collect the initial position and size image data of road studs within the coverage area through cameras installed in the road area;

[0092] Use artificial intelligence technology to detect the collected road stud images, obtain the pixel coordinates and size data of the road studs, and record the initial positions and sizes of each road stud;

[0093] The artificial intelligence technology includes a deep learning object detection algorithm for identifying and locating the positions of road studs in camera images.

[0094] Periodic detection stage:

[0095] Set a predetermined time interval and periodically collect road stud images in the camera coverage area;

[0096] Use artificial intelligence algorithms to detect the positions and sizes of road studs in the periodically collected images;

[0097] By calculating the change amount between the detected positions and sizes of the road studs and the initial recorded data, determine whether the road studs have shifted in position or changed in size.

[0098] Quantitative calculation of road stud displacement:

[0099] Quantify the position change of road studs by calculating the Intersection over Union (IOU) of two detection frames. The IOU is defined as the ratio of the intersection area to the union area of the two detection frames;

[0100] Divide the IOU values into multiple intervals, and preset multiple threshold intervals (such as 0.9, 0.8, 0.7, 0.6) corresponding to different alarm levels;

[0101] When the IOU is 1, no alarm is triggered. The IOU within each interval corresponds to a low to high alarm level.

[0102] Multiple occlusion confirmation mechanism:

[0103] In periodic detection, if the position of a road stud is blocked by a vehicle or the like and cannot be detected, set a continuous multiple detection interval (such as 3 times). If the road stud still cannot be detected, trigger the highest-level alarm;

[0104] If detection can be restored after a single occlusion, determine whether there is a change based on the detection data.

[0105] Judgment of overall regional deformation:

[0106] When multiple road studs have all shifted in position or changed in size, accumulate the alarm levels of each road stud and calculate the overall regional alarm level;

[0107] When the number of road studs is detected to decrease, a continuous alarm mechanism is triggered, and the detection threshold is set for multiple times (for example, 2 or 3 times continuously) to increase the alarm level until the highest-level alarm is triggered.

[0108] Alarm output and handling:

[0109] According to the alarm level of the change in the position of the road studs and the alarm situation of the decrease in the number of road studs, alarm information of different levels is output to the background monitoring platform;

[0110] The background monitoring platform generates a warning based on the alarm information and notifies the relevant users. The users make a secondary judgment in combination with the camera video images or dispatch personnel for on-site inspection;

[0111] According to the alarm level and the results of the secondary judgment, a preset handling plan is executed.

[0112] Among them,

[0113] The detection of the road stud size includes calculating the pixel size of the road studs in the images collected by the camera and correcting the actual physical size change of the road studs according to their positions in the camera's field of view.

[0114] Artificial intelligence technology uses deep convolutional neural networks (such as YOLO or Faster R-CNN) to achieve real-time detection and position recording of road studs.

[0115] The time interval of periodic detection is set to 1 hour, 6 hours or 12 hours according to the road surface conditions, and the specific time interval can be dynamically adjusted according to the actual situation.

[0116] The quantization rules for the change in the position of road studs include the following steps:

[0117] Extract the detection data of the previous detection, the previous ten detections and the same time period of the previous day;

[0118] Calculate the IOU value between the current detection result and the above historical data to judge the change trend of the road stud position.

[0119] The alarm information includes the alarm level, the change amplitude of the road stud position, the decrease situation of the road stud number and the real-time image collected by the camera.

[0120] The alarm level of the overall deformation of the area is calculated according to the following formula:

[0121] Area alarm level = Σ (alarm level of individual road studs).

[0122] Among them,

[0123] Detailed description of the initial deployment stage:

[0124] 1. Camera installation and initial image data collection

[0125] Reasonably deploy cameras in the road area so that the cameras cover the distribution range of road studs within the entire monitoring area.

[0126] The cameras should meet the following requirements:

[0127] Resolution: It should be able to clearly capture the image details of road studs, at least 1080p high definition.

[0128] Installation location: According to the road conditions, the cameras can be installed on the median strip or the roadside between two-way lanes, ensuring that the images of the road studs at the farthest end are clearly visible.

[0129] Viewing angle range: The cameras need to have a sufficient wide angle to cover multiple road studs, and at the same time ensure that the positions of the road studs in the picture will not be distorted due to blind spots or excessive viewing angles.

[0130] Collect the first set of initial image data:

[0131] Start the cameras and continuously collect multiple sets of video frames, and intercept clear frames as static images;

[0132] Ensure that all road studs in the images are not blocked by vehicles or the like.

[0133] 2. Detection of the initial positions and sizes of road studs

[0134] Use artificial intelligence technology to process the collected images. The specific steps are as follows:

[0135] Image preprocessing: Perform denoising processing on the collected images to eliminate the influence of light changes or blurring on the recognition accuracy. Perform geometric correction on the images to ensure that the camera viewing angle deviation or distortion will not affect the recognition of the positions and sizes of road studs.

[0136] Object detection algorithm: Use deep learning object detection algorithms such as YOLO (You Only Look Once), Faster R-CNN (Region-based Convolutional Neural Network), etc. to identify the road studs in the images. Mark the positions of the road studs as rectangular detection frames and obtain their pixel coordinates (for example, represent the position of the detection frame with the pixel coordinates of the upper left and lower right pixel points).

[0137] Initial record of road stud sizes: Record the width, height, and center point coordinates of the detection frames to represent the initial sizes and positions of each road stud. Considering the possible size changes caused by the distance from the camera, use the calibration formula of the actual physical size and pixel size of the road studs for calibration:

[0138] S real =S pixel ×f(d)

[0139] Among them, S real is the actual physical size of the spike, and S pixel is the pixel size. f(d) is a calibration function for the relationship between distance and size.

[0140] 3. Initial Record Data Generation

[0141] Store all detected spike information (including position and size) in the database to form "initial record data". The recorded content includes:

[0142] Spike label: Each spike is assigned a unique identification number for subsequent tracking.

[0143] Initial position: including the pixel coordinates of the spike in the image and the physical position coordinates (such as longitude and latitude or relative position).

[0144] Initial size: the actual physical width and height of the spike.

[0145] Perform a one-time verification on the recorded spike position and size data:

[0146] Ensure that all spikes can be completely detected in the image;

[0147] Set a threshold in the detection algorithm to filter out interference targets misidentified (such as fallen leaves, stains, etc.).

[0148] 4. System Configuration and Monitoring Area Determination

[0149] Determine the parameters of the monitoring area:

[0150] Road length, width, and spike distribution interval.

[0151] Maximum monitoring distance and monitoring coverage width of the camera.

[0152] Configure the time interval for periodic detection:

[0153] It is recommended to set the initial detection interval according to the importance of the road and the vehicle flow, for example, once per hour.

[0154] Set the initial alarm parameters:

[0155] Preset the threshold for spike offset (such as the IOU value range), corresponding to the alarm levels from low to high.

[0156] 5. Manual Confirmation of the Initial State

[0157] The initial data of the spike position and size automatically recorded by the system needs to be manually confirmed to ensure its accuracy:

[0158] Manually compare and verify the recognition results of each spike in combination with the real-time video collected by the camera.

[0159] If there is an inaccurate recognition, the initial position record of the spike can be manually adjusted.

[0160] 6. Initial Deployment Data Saving and System Startup

[0161] Save all the confirmed initial position and size data of the spikes to the system background, and enable the system to enter the periodic detection mode.

[0162] Through the above steps, the initial deployment stage completes the complete modeling and recording of the spikes in the monitoring area, providing basic data support for subsequent periodic detection and alarm determination.

[0163] Among them,

[0164] Detailed description of the displacement quantization calculation of the spike:

[0165] IOU is a commonly used metric to measure the overlap degree between two rectangular boxes (detection boxes), and is widely used in object detection tasks. It is defined as the ratio of the intersection area to the union area of the two detection boxes.

[0166] To accurately quantify the displacement of the spike, it is necessary to calculate the detection box of each spike frame by frame:

[0167] Coordinate extraction of the detection box:

[0168] Extract the rectangular box coordinates of the position of the spike detected each time from the output of the object detection algorithm (such as YOLO or Faster R-CNN).

[0169] Intersection area calculation:

[0170] Calculate the coordinates of the intersection area of the two detection boxes;

[0171] Union area calculation:

[0172] The total area of the two detection boxes minus the intersection area.

[0173] The intersection area is divided by the union area to obtain the IOU value.

[0174] According to the size of the IOU value, the displacement of the spike can be quantified and corresponding to the alarm level:

[0175]

[0176]

[0177] Among them, the smaller the IOU, the lower the overlap degree between the detection box of the spike and the initial position, and the more significant the displacement of the spike. When the IOU is lower than 0.6, the highest-level alarm should be triggered immediately, and the monitoring platform or relevant personnel should be notified for handling.

[0178] Multiple detections and trend analysis: To avoid false alarms caused by errors in single detections, multiple detections and trend analysis can be introduced:

[0179] Reference of multiple historical data: Extract the detection results of the same time period in the previous 1 time, the previous 10 times, and the previous 1 day, and calculate the difference between the IOU value and the current detection result. Combine the changing trends of multiple IOUs to determine whether the displacement of the spike continues to occur.

[0180] Setting of time interval: Set the time interval for spike position detection, for example, detect once per hour. If the IOU value continues to be lower than a certain threshold (such as 0.7) in three consecutive detections, the alarm level is raised.

[0181] Optimization of IOU value calculation and boundary conditions

[0182] Handling of occlusion situations: If the spike is occluded by a vehicle or fails to be recognized for other reasons during a certain detection, it is necessary to detect continuously for multiple times (such as 3 times) to confirm whether it is a real displacement or occlusion. If the spike is not detected in three consecutive times, it should be determined that the spike may be lost and the highest-level alarm is directly triggered.

[0183] Error correction: Changes in the installation position of the camera or ambient light may cause detection errors. Errors can be reduced by dynamically smoothing the coordinates of the detection frame (such as weighted average).

[0184] Output of alarm information:

[0185] When the calculated IOU value corresponds to a specific alarm level, the system will generate the following alarm information and send it to the background: Alarm level: indicating the severity of the displacement; Spike number: marking the specific spike with displacement; Displacement details: including the initial position, current detection position, IOU value, and alarm level of the spike; Image data: attaching the currently detected image for manual verification.

[0186] The above are the embodiments of the present invention. For those of ordinary skill in the art, according to the teachings of the present invention, all equivalent changes, modifications, substitutions, and variations made within the scope of the patent application of the present invention without departing from the principles and spirit of the present invention shall fall within the scope of the present invention.

Claims

1. A road surface deformation judgment method based on artificial intelligence recognition of road spikes, characterized in that: At least the following steps are included: Collect image data of the initial position and size of road spikes in a preset area, use artificial intelligence to detect the image data, and obtain and record the initial pixel coordinates and size data of each road spike; The above image data is collected periodically, and the image data is detected by using artificial intelligence to obtain and record the current pixel coordinates and size data of each road spike; By calculating the change between the current pixel coordinates and size data and the initial pixel coordinates and size data, it is determined whether the road spike has a position shift or a size change.

2. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 1, characterized in that: The following steps are also included: Output different levels of warning information according to the change amount.

3. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 2, characterized in that: The steps of collecting image data of the initial position and size of road spikes in a preset area, detecting the image data using artificial intelligence, and obtaining and recording the initial pixel coordinates and size data of each road spike include: Select a low traffic period without vehicle obstruction, use the camera to collect static images or video frames of the coverage area for recording the initial state of the road studs; The noise is eliminated through preprocessing, and the brightness and contrast of the image are adjusted to enhance the edge clarity of the road studs; A deep learning target detection model is used to automatically identify the position of road spikes in an image. The deep learning target detection model inputs a preprocessed road image and outputs a rectangular detection frame and size of each road spike as an initial state of the road spike.

4. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 2, characterized in that: The steps of periodically collecting the image data, detecting the image data using artificial intelligence, and obtaining and recording the current pixel coordinates and size data of each road spike include: Preset detection cycle. In each detection cycle, the camera collects static images or video frames of the coverage area to record the current status of the road studs. The noise is eliminated through preprocessing, and the brightness and contrast of the image are adjusted to enhance the edge clarity of the road studs; A deep learning target detection model is used to automatically identify the position of road spikes in an image. The deep learning target detection model inputs a preprocessed road image and outputs a rectangular detection box and size of each road spike as the current state of the road spike.

5. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 2, characterized in that: The step of determining whether the road spike has positional displacement or size change by calculating the change between the current pixel coordinates and size data and the initial pixel coordinates and size data specifically includes: The change in the position of the road spikes is quantified by calculating the intersection over union (IOU) of the two detection frames. IOU is defined as the ratio of the intersection area to the union area of ​​the two detection frames. The IOU value is divided into multiple intervals, and multiple threshold intervals are preset to correspond to different alarm levels; When IOU is 1, no alarm is triggered. IOU corresponds to low to high alarm levels within each interval.

6. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 2, characterized in that: The step of periodically collecting the image data, detecting the image data using artificial intelligence, and obtaining and recording the current pixel coordinates and size data of each road spike also includes: If the road spike position is blocked and cannot be detected during periodic detection, set multiple detection intervals. If the road spike still cannot be detected, the highest level alarm will be triggered; If detection can be restored after a single occlusion, determine whether there is a change based on the detection data.

7. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 5, characterized in that: The step of determining whether the road spike has positional displacement or size change by calculating the change between the current pixel coordinates and size data and the initial pixel coordinates and size data further includes: When multiple road spikes are offset or have size changes, the warning levels of each road spike are accumulated to calculate the overall warning level of the area; When it is detected that the number of road spikes is reduced, a continuous alarm mechanism is triggered, and multiple detection thresholds are set to increase the alarm level until the highest level alarm is triggered.

8. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 1, characterized in that: Also includes: Alarm output and disposal: According to the alarm level of the change of the position of the road studs and the alarm situation of the reduction of the number of road studs, different levels of alarm information are output to the background monitoring platform; the background monitoring platform generates an early warning based on the alarm information and notifies the relevant users, and the users make a secondary judgment based on the camera video image, or send personnel to conduct on-site inspections; according to the alarm level and the results of the secondary judgment, the preset disposal plan is executed.

9. The method for determining road deformation based on artificial intelligence identification of road spikes according to claim 1, characterized in that: The image data is collected by a camera, and the camera is installed in the middle isolation zone of a two-way lane or on the road side to ensure that all road studs within the camera coverage area are within the detection range.

10. A road surface deformation judgment system based on artificial intelligence recognition of road spikes, characterized in that: The initial data acquisition module collects the image data of the initial position and size of the road spikes in the preset area, detects the image data using artificial intelligence, and obtains and records the initial pixel coordinates and size data of each road spike; The current data acquisition module periodically collects the above image data, detects the image data using artificial intelligence, and obtains and records the current pixel coordinates and size data of each road spike; The change state calculation module determines whether the road spike has positional displacement or size change by calculating the change between the current pixel coordinates and size data and the initial pixel coordinates and size data; The alarm module outputs alarm information of different levels according to the change amount.

Images