A method and device for detecting over-width vehicles during road maintenance and construction

Abstract

The present invention is applicable to the field of intelligent transportation technology, and provides a method and device for detecting over-width vehicles during road maintenance and construction. In this embodiment, for the construction area where road maintenance is being carried out, a vehicle width detection device equipped with a laser radar is first placed around the area, and the device is used to monitor a fixed area away from the construction area. When a vehicle is detected in the monitored area, the point cloud information of the vehicle is obtained, and the target point cloud information moving toward the vehicle width detection device is judged by the point cloud information frames at two adjacent moments. The target point cloud information is then clustered into several point cloud blocks, and the width of the point cloud blocks is calculated by using PointRCNN and a PCA method based on deviation weighting to determine the width of the vehicle. When it is determined that the width of the vehicle is greater than the preset alarm width, an alarm is processed. This solves the problem that the existing method of judging whether a vehicle is over-width based on manual visual inspection not only has safety risks, but is also labor-intensive and inefficient.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a method and device for detecting over-width vehicles during road maintenance and construction. Background Art

[0002] Highways require regular upkeep and maintenance. Currently, during maintenance, construction is typically performed on one side of the road, requiring width detection for vehicles passing through that lane. When a vehicle passes by, manual visual inspection is used to determine if the lane is excessively wide. The vehicle is then instructed to slow down, and safety reminders are provided to construction workers via remote intercoms. This method not only poses safety risks but is also labor-intensive and inefficient. Summary of the Invention

[0003] In view of this, the present application provides a method and device for detecting over-width vehicles during road maintenance construction, so as to realize automatic detection of the width of passing vehicles at the construction site, and avoid the problem of over-width vehicles affecting construction equipment and causing traffic accidents.

[0004] In a first aspect, the present application provides a method for detecting over-width vehicles during road maintenance and construction, which is applied to an intelligent transportation system. The system includes a vehicle width detection device and an alarm device, wherein the vehicle width detection device is equipped with a laser radar. The method includes:

[0005] Scanning the monitoring area by using the laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames;

[0006] Comparing the current point cloud information frame with the point cloud information frame at the previous moment, determining target point cloud information moving toward the vehicle width detection device, and clustering the target point cloud information into a plurality of point cloud blocks;

[0007] For each point cloud block, PointRCNN is used to extract the candidate region of the point cloud block, and the candidate bounding box of the candidate region is regressed to output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) is the center coordinate of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle;

[0008] The yaw angle θ is optimized by a deviation-weighted PCA method, and the width w of the point cloud block is determined based on the optimized yaw angle θ. The vehicle width of the target vehicle is determined based on the width w of each point cloud block, and when the vehicle width is greater than a preset alarm width, the target vehicle is determined as an over-width vehicle and an early warning is issued through the alarm device.

[0009] Optionally, comparing the current point cloud information frame with the point cloud information frame at the previous moment to determine the target point cloud information moving toward the vehicle width detection device includes:

[0010] Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device;

[0011] The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

[0012] Optionally, clustering the target point cloud information into a plurality of point cloud blocks includes:

[0013] Clustering the point cloud information in a horizontal direction to obtain a point cloud array consisting of a plurality of horizontal clusters;

[0014] The point cloud array is vertically clustered, and point cloud arrays within the same vertical range are clustered into a point cloud block.

[0015] Optionally, before comparing the current point cloud information frame with the point cloud information frame at the previous moment, the method further includes:

[0016] A point cloud matrix is ​​generated according to the point cloud information, and the point cloud matrix is ​​pushed into a stack of preset capacity.

[0017] Optionally, the vehicle width detection device is placed at any position between the construction area and the monitoring area.

[0018] A second aspect of the present application provides an over-width vehicle detection device during road maintenance and construction, which is applied to an intelligent transportation system. The system includes a vehicle width detection device and an alarm device. The vehicle width detection device is equipped with a laser radar. The device includes:

[0019] a point cloud information acquisition unit, configured to scan the monitoring area using a laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames;

[0020] a point cloud block acquisition unit, configured to compare a current point cloud information frame with a point cloud information frame at a previous moment, determine target point cloud information moving toward the vehicle width detection device, and cluster the target point cloud information into a plurality of point cloud blocks;

[0021] The bounding box determination unit is used to extract the candidate area of ​​the point cloud block using PointRCNN, regress the candidate bounding box of the candidate area, and output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) is the center coordinate of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle;

[0022] An over-width vehicle determination unit is used to optimize the yaw angle θ through a deviation-weighted PCA method, determine the width w of the point cloud block according to the optimized yaw angle θ, and determine the vehicle width of the target vehicle according to the width w of each point cloud block, and when the vehicle width is greater than a preset warning width, determine the target vehicle as an over-width vehicle and issue an early warning through the alarm device.

[0023] Optionally, the step of comparing the current point cloud information frame with the point cloud information frame at the previous moment in the point cloud block acquisition unit to determine the target point cloud information moving toward the vehicle width detection device includes:

[0024] Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device;

[0025] The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

[0026] Optionally, the step of clustering the target point cloud information into a plurality of point cloud blocks in the point cloud block acquiring unit includes:

[0027] Clustering the point cloud information in a horizontal direction to obtain a point cloud array consisting of a plurality of horizontal clusters;

[0028] The point cloud array is vertically clustered, and point cloud arrays within the same vertical range are clustered into a point cloud block.

[0029] Optionally, before the point cloud block acquisition unit compares the current point cloud information frame with the point cloud information frame at the previous moment, the device further includes:

[0030] A stack processing unit is used to generate a point cloud matrix according to the point cloud information and push the point cloud matrix into a stack of preset capacity.

[0031] Optionally, the vehicle width detection device in the point cloud information acquisition unit is placed at any position between the construction area and the monitoring area.

[0032] In the embodiment provided in the present application, for a construction area where road maintenance is being carried out, a vehicle width detection device equipped with a laser radar is first placed around the area, and the device is used to monitor a fixed area away from the construction area. When a vehicle is detected in the monitored area, the point cloud information of the vehicle is obtained, and the target point cloud information moving toward the vehicle width detection device is judged by the point cloud information frames at two adjacent moments. The target point cloud information is then clustered into several point cloud blocks, and the width of the point cloud blocks is calculated by using PointRCNN and a PCA method based on deviation weighting to determine the width of the vehicle. When it is determined that the width of the vehicle is greater than the preset alarm width, an alarm is processed. This solves the problem that the existing method of judging whether a vehicle is overwidth based on manual visual inspection not only has safety risks, but is also labor-intensive and inefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flow chart of the method provided in the embodiment of the present application;

[0034] Figure 2 A structural diagram of a vehicle width detection device provided in an embodiment of the present application;

[0035] Figure 3 A structural diagram of a vehicle width detection device provided in an embodiment of the present application;

[0036] Figure 4 A diagram showing the placement of a vehicle width detection device provided in an embodiment of the present application;

[0037] Figure 5 A diagram of the device structure provided in an embodiment of the present application;

[0038] Figure 6 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0040] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0041] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0042] The present application provides a method and device for detecting over-width vehicles during road maintenance and construction, so as to detect the width of passing vehicles at the construction site and prevent over-width vehicles from affecting construction equipment and causing traffic accidents.

[0043] The following specific embodiments are used to describe the technical solution of the present application in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0044] like Figure 1 FIG. 1 is a flow chart of a method for detecting an over-width vehicle during road maintenance and construction provided by the present application. The method is applied to an intelligent transportation system. The system includes a vehicle width detection device and an alarm device. The process may include the following steps:

[0045] In step S101 , the monitoring area is scanned by a laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames.

[0046] In this embodiment, the structure of the vehicle width detection device is as follows: Figure 2 , Figure 3As shown, label 1 is the equipment caster, which is mainly used to adjust the direction of the laser radar to ensure that the central axis of the laser radar is facing the road sideline; label 2 is the equipment rack; label 3 is the industrial computer; label 4 is the push-pull handrail; label 5 is the laser radar, which requires a horizontal scanning angle of more than 60 degrees and a vertical scanning line number of more than 120 lines. The laser radar is powered by a battery and connected to the industrial computer using a network cable; label 6 is the communication antenna of the LORA protocol; label 7 is the wire hook; label 8 is the lithium iron phosphate battery; label 9 is the cooling fan; label 10 is the side handle; label 11 is the 4 / 5G communication antenna of the mobile network; label 12 is the sound and light alarm device; label 13 is the fastening and adjustment device of the laser radar.

[0047] The vehicle width detection device can be placed on the hard shoulder about 1.8 km before the construction area (vehicles at this location have more than 60 seconds to reach the construction area, and the construction area can take corrective measures), and the center line of the laser radar is aligned with the edge line of the road. After the device is turned on, the laser radar continuously scans the road ahead. Figure 4 As shown, when a vehicle enters the laser scanning area (i.e., the monitoring area), several point cloud information frames of the vehicle at different times can be obtained, thereby forming the real-time point cloud information of the vehicle.

[0048] In another embodiment, the vehicle width detection device is placed at any position between the construction area and the monitoring area. The specific placement is as follows: Figure 4 As shown, placing the vehicle width detection device between the construction area and the monitoring area allows for more precise control and management of traffic flow between the two areas. This setup helps ensure that only vehicles that meet the required dimensions can enter the construction area, reducing potential damage and safety risks. Furthermore, compared to other placements, placing the device between the construction and monitoring areas offers a cost-effective and comprehensive detection solution, providing a simple, low-cost, and comprehensive detection solution.

[0049] Step S102 : Compare the current point cloud information frame with the point cloud information frame at the previous moment, determine the target point cloud information moving toward the vehicle width detection device, and cluster the target point cloud information into a plurality of point cloud blocks.

[0050] In this embodiment, when the LiDAR scans the monitoring area to obtain point cloud information, it will contain a lot of point cloud information other than vehicles, such as point cloud information of objects such as trees and rocks. However, relative to the vehicle, these other point cloud information does not move. Therefore, by removing all non-moving point cloud information, the target point cloud information containing only the vehicle can be obtained.

[0051] The specific method is as follows:

[0052] Each time a point cloud information frame is received, the first distance value between each point in that point cloud information frame and the vehicle width detection device is determined and compared with the second distance value between the corresponding point and the vehicle width detection device in the point cloud information frame at the previous moment. If the first distance value of the current point is less than the second distance value, it indicates that the point is moving toward the vehicle width detection device. To more clearly determine the distance values ​​of each point at different times, a time interval of 1 second can be selected. That is, each received point cloud information frame is compared with the point cloud information frame 1 second ago. This determines the point moving toward the vehicle width detection device and thus determines the target point cloud information.

[0053] Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device;

[0054] The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

[0055] After determining the target point cloud information, it can be clustered into several point cloud blocks using density-based clustering methods. Density-based clustering algorithms, such as DBSCAN, can cluster dense datasets of any shape and detect outliers during clustering, making them ideal for millimeter-wave radar point cloud data processing.

[0056] In another embodiment, clustering the target point cloud information into a plurality of point cloud blocks includes:

[0057] First, all target point cloud information is clustered horizontally to form an array of several horizontal clusters. The horizontally clustered cluster array is then clustered vertically, turning the horizontal clusters within a certain vertical range into a point cloud block. Several point cloud blocks represent the point cloud traces of several moving objects.

[0058] In this embodiment, the horizontal-first, then vertical clustering method allows for more accurate identification of object boundaries in point clouds. Furthermore, step-by-step clustering can reduce computational complexity, particularly when processing large-scale point cloud data. Clustering target point cloud information into multiple point cloud blocks not only improves clustering accuracy and computational efficiency, but also enhances adaptability to complex environments, improves the stability of clustering results, and facilitates subsequent target detection and processing.

[0059] In another embodiment, before comparing the current point cloud information frame with the point cloud information frame at the previous moment, the method further includes:

[0060] A point cloud matrix is ​​generated according to the point cloud information, and the point cloud matrix is ​​pushed into a stack of preset capacity.

[0061] In this embodiment, the stack capacity is set based on the performance of the LiDAR. For example, if the LiDAR can send 10 frames of point cloud information per second, 10 point cloud matrices can be pushed into the stack per second. Therefore, the stack capacity can be set to accommodate 30 frames of point cloud information, thus saving memory costs. This embodiment generates a point cloud matrix from point cloud information and pushes it into a stack of preset capacity, which improves data processing efficiency and flexibility, and optimizes storage and transmission.

[0062] Step S103: For each point cloud block, use PointRCNN to extract the candidate region of the point cloud block, regress the candidate bounding box of the candidate region, and output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) are the center coordinates of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle.

[0063] In this embodiment, for any point cloud block, the point cloud in the point cloud block is represented by P i , where i∈[1,M]. M is the detected vehicle target point cloud. The steps to extract the bounding box of this point cloud block through PointRCNN are as follows:

[0064] 1. Combine PointNet++ and Pyramid Network (FPN) to form a multi-scale point cloud feature extraction framework. PointNet++ first normalizes the coordinates of each point cloud to the input range [0,1] of the model, and uses layered sampling and grouping (Set Abstraction, SA) modules to divide the input point cloud into multiple subsets, each of which learns local geometric features. Each layer outputs a feature set of points, which is recorded as Among them, l represents the number of layers, N l is the number of points in this layer, is the feature of point i. The FPN refers to fusing feature layers of different scales and obtaining a feature map F by upsampling and horizontal connection. p Then, the center position of the candidate box is calculated through the features of the point to generate the candidate area.

[0065] 2. Regress the bounding box of the candidate region based on the loss function. The loss function is GIoU loss. IoU 3Dis the intersection-union ratio of the three-dimensional prediction box A and the real box B of the candidate area, C is the volume of the three-dimensional bounding box containing A and B, and A∪B is the union volume of the prediction box and the real box. By regressing the bounding box of the candidate area through this loss function, the optimal bounding box can be determined, and the optimal bounding box is determined as the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) are the center coordinates of the bounding box, w, l, h are the width, length, and height, respectively, corresponding to the geometric characteristics of the vehicle, and θ is the yaw angle.

[0066] By traversing each point cloud block through the above steps, the bounding box of each point cloud block can be determined.

[0067] In step S104, the yaw angle θ is optimized by a deviation-weighted PCA method, and the width w of the point cloud block is determined according to the optimized yaw angle θ. The vehicle width of the target vehicle is determined according to the width w of each point cloud block, and when the vehicle width is greater than a preset alarm width, the target vehicle is determined as an over-width vehicle and an early warning is issued through the alarm device.

[0068] In this embodiment, the point cloud centroid c of a point cloud block is calculated by averaging the positions of all points in the block. This refers to the average position of all points in the point cloud and can be understood as the center or center of gravity of the point cloud. The point cloud centroid c is calculated by adding the coordinates of all points in the point cloud and then dividing by the number of points in the point cloud block.

[0069] Then the bias-weighted PCA method assigns a weight w to each point i , the weight is based on the distance from the point to the point cloud centroid and the fit of the bounding box, which is Where c is the point cloud centroid, σ is the attenuation of the control weight, and IoU(p i ,B) is point p i The bounding box fit of the unit, σ is selected based on the statistics of the point cloud distribution, σ = α·std(‖p i -c‖),α∈[1,3],std(‖p i -c‖) is the standard deviation of the distance from the point to the centroid. Then the weighted covariance matrix and the PCA main direction angle φ are calculated. The calculation formula of the weighted covariance matrix is: The PCA main direction angle φ refers to the angle between the PCA main direction and the coordinate axis parallel to the lane, where the PCA main direction Combine PointRCNN output θ and PCA main direction angle φ to optimize the yaw angle. The optimization formula is θ final =λθ+(1-λ)φ,λ∈[0,1]. In the above formula, the bounding box changes as the yaw angle is optimized. Solving the formula can obtain the width w of the point cloud block.

[0070] The width of each point cloud block is calculated separately, and the widest one is taken as the width of the target vehicle. When the width exceeds the alarm width, an over-width signal is generated and sent to the alarm device deployed in the construction area via the LoRa wireless protocol, activating an audible and visual alarm. The over-width signal can also be reported to the cloud platform via remote communication.

[0071] So far, completed Figure 1 The process shown.

[0072] In an embodiment of the present application, for a construction area where road maintenance is being carried out, a vehicle width detection device equipped with a laser radar is first placed around the area, and a fixed area away from the construction area is monitored by the device. When a vehicle is detected in the monitored area, the point cloud information of the vehicle is obtained, and the target point cloud information moving toward the vehicle width detection device is judged by the point cloud information frames at two adjacent moments. The target point cloud information is then clustered into several point cloud blocks, and the width of the point cloud blocks is then calculated by using PointRCNN and a PCA method based on deviation weighting to determine the width of the vehicle. When it is determined that the width of the vehicle is greater than the preset alarm width, an alarm is processed. This solves the problem that the existing method of judging whether a vehicle is overwidth based on manual visual inspection not only has safety risks, but is also labor-intensive and inefficient.

[0073] like Figure 5 As shown, the present application also provides an over-width vehicle detection device during road maintenance and construction, which is applied to an intelligent transportation system. The system includes a vehicle width detection device and an alarm device. The vehicle width detection device is equipped with a laser radar. The device includes:

[0074] The point cloud information acquisition unit 501 is configured to scan the monitoring area using a laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames;

[0075] a point cloud block acquiring unit 502 for comparing a current point cloud information frame with a point cloud information frame at a previous moment, determining target point cloud information moving toward the vehicle width detection device, and clustering the target point cloud information into a plurality of point cloud blocks;

[0076] The bounding box determination unit 503 is used to extract the candidate area of ​​the point cloud block using PointRCNN, and regress the candidate bounding box of the candidate area to output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) is the center coordinate of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle;

[0077] The over-width vehicle determination unit 504 is used to optimize the yaw angle θ through the deviation-weighted PCA method, determine the width w of the point cloud block according to the optimized yaw angle θ, and determine the vehicle width of the target vehicle according to the width w of each point cloud block. When the vehicle width is greater than a preset alarm width, the target vehicle is determined as an over-width vehicle and an early warning is issued through the alarm device.

[0078] In another embodiment, the point cloud block acquisition unit compares the current point cloud information frame with the point cloud information frame at the previous moment to determine the target point cloud information moving toward the vehicle width detection device, including:

[0079] Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device;

[0080] The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

[0081] In another embodiment, clustering the target point cloud information into a plurality of point cloud blocks in the point cloud block acquiring unit includes:

[0082] Clustering the point cloud information in a horizontal direction to obtain a point cloud array consisting of a plurality of horizontal clusters;

[0083] The point cloud array is vertically clustered, and point cloud arrays within the same vertical range are clustered into a point cloud block.

[0084] In another embodiment, before the point cloud block acquisition unit compares the current point cloud information frame with the point cloud information frame at the previous moment, the device further includes:

[0085] The stack processing unit 505 is configured to generate a point cloud matrix according to the point cloud information and push the point cloud matrix into a stack of a preset capacity.

[0086] In another embodiment, the vehicle width detection device in the point cloud information acquisition unit is placed at any position between the construction area and the monitoring area.

[0087] The above-mentioned embodiment of the present invention provides a method for detecting over-width vehicles during road maintenance construction, and based on this method, provides a device for detecting over-width vehicles during road maintenance construction. Through the above-mentioned method and device, it is possible to automatically detect the width of passing vehicles at the construction site, thereby avoiding the problem of over-width vehicles affecting construction equipment and causing traffic accidents.

[0088] This embodiment also discloses a computer device, such as Figure 3 As shown, the computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement any of the above-mentioned methods for detecting over-width vehicles during road maintenance and construction.

[0089] In addition, in the above-mentioned example implementation of the over-width vehicle detection device during road maintenance and construction, the logical division of each program module is only an example. In actual application, the above-mentioned functions can be assigned to different program modules as needed, for example, for the configuration requirements of the corresponding hardware or the convenience of software implementation. That is, the internal structure of the over-width vehicle detection device during road maintenance and construction is divided into different program modules to complete all or part of the functions described above.

[0090] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for detecting extra-wide vehicles during road maintenance and construction, characterized in that: Applied to an intelligent transportation system, the system includes a vehicle width detection device and an alarm device, the vehicle width detection device is equipped with a laser radar; the method includes: Scanning the monitoring area by using the laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames; Comparing the current point cloud information frame with the point cloud information frame at the previous moment, determining target point cloud information moving toward the vehicle width detection device, and clustering the target point cloud information into a plurality of point cloud blocks; For each point cloud block, PointRCNN is used to extract the candidate region of the point cloud block, and the candidate bounding box of the candidate region is regressed to output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) is the center coordinate of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle; The yaw angle θ is optimized by a deviation-weighted PCA method, and the width w of the point cloud block is determined based on the optimized yaw angle θ. The vehicle width of the target vehicle is determined based on the width w of each point cloud block, and when the vehicle width is greater than a preset alarm width, the target vehicle is determined as an over-width vehicle and an early warning is issued through the alarm device.

2. The method according to claim 1, characterized in that The step of comparing the current point cloud information frame with the point cloud information frame at the previous moment to determine the target point cloud information moving toward the vehicle width detection device includes: Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device; The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

3. The method according to claim 1, characterized in that Clustering the target point cloud information into a plurality of point cloud blocks includes: Clustering the point cloud information in a horizontal direction to obtain a point cloud array consisting of a plurality of horizontal clusters; The point cloud array is vertically clustered, and point cloud arrays within the same vertical range are clustered into a point cloud block.

4. The method according to claim 1, wherein Before comparing the current point cloud information frame with the point cloud information frame at the previous moment, the method further includes: A point cloud matrix is ​​generated according to the point cloud information, and the point cloud matrix is ​​pushed into a stack of preset capacity.

5. The method according to claim 1, wherein The vehicle width detection device is placed at any position between the construction area and the monitoring area.

6. An over-width vehicle detection device during road maintenance and construction, characterized in that: Applied to an intelligent transportation system, the system includes a vehicle width detection device and an alarm device, wherein the vehicle width detection device is equipped with a laser radar; the device includes: a point cloud information acquisition unit, configured to scan the monitoring area using a laser radar on the vehicle width detection device placed around the construction area to obtain real-time point cloud information of the target vehicle in the monitoring area, wherein the point cloud information includes a plurality of point cloud information frames; a point cloud block acquisition unit, configured to compare a current point cloud information frame with a point cloud information frame at a previous moment, determine target point cloud information moving toward the vehicle width detection device, and cluster the target point cloud information into a plurality of point cloud blocks; The bounding box determination unit is used to extract the candidate area of ​​the point cloud block using PointRCNN, regress the candidate bounding box of the candidate area, and output the bounding box B of the point cloud block. j ={(c x ,c y ,c z ),(w,l,h),θ}, where (c x ,c y ,c z ) is the center coordinate of the bounding box, w, l, h are the width, length, and height of the target vehicle, and θ is the yaw angle; An over-width vehicle determination unit is used to optimize the yaw angle θ through a deviation-weighted PCA method, determine the width w of the point cloud block according to the optimized yaw angle θ, and determine the vehicle width of the target vehicle according to the width w of each point cloud block, and when the vehicle width is greater than a preset warning width, determine the target vehicle as an over-width vehicle and issue an early warning through the alarm device.

7. The device according to claim 6, characterized in that The step of comparing the current point cloud information frame with the point cloud information frame at the previous moment in the point cloud block acquisition unit to determine the target point cloud information moving toward the vehicle width detection device includes: Determining a first distance value between the current point cloud information frame and the vehicle width detection device, and a second distance value between the point cloud information at a previous moment and the vehicle width detection device; The first distance value and the second distance value are compared. When the first distance value is smaller than the second distance value, the point cloud information corresponding to the current point cloud information frame is determined as the target point cloud information.

8. The device according to claim 6, characterized in that Clustering the target point cloud information into a plurality of point cloud blocks in the point cloud block acquisition unit includes: Clustering the point cloud information in a horizontal direction to obtain a point cloud array consisting of a plurality of horizontal clusters; The point cloud array is vertically clustered, and point cloud arrays within the same vertical range are clustered into a point cloud block.

9. The device according to claim 6, characterized in that Before the point cloud block acquisition unit compares the current point cloud information frame with the point cloud information frame at the previous moment, the device further includes: A stack processing unit is used to generate a point cloud matrix according to the point cloud information and push the point cloud matrix into a stack of preset capacity.

10. The device according to claim 6, characterized in that The vehicle width detection device in the point cloud information acquisition unit is placed at any position between the construction area and the monitoring area.

Images