A channel ship flow rate counting method and system based on laser radar

Abstract

The application discloses a kind of channel ship flow statistics method and system based on laser radar, it is related to the processing method of laser radar point cloud data in wisdom channel field.The method comprises the following steps: 1) utilize laser radar to realize channel all-weather automatic monitoring and ship data acquisition;2) the point cloud data of channel scene is preprocessed by space coordinate filtering method;3) ship target detection is realized by considering the DBSCAN algorithm of weight;4) by extracting and analyzing the edge features of ship, the statistics of channel ship flow and the judgment of ship sailing direction are completed;5) by constructing 3D boundary box model, the size of ship is calculated;6) combined with the prior knowledge of standard ship type of inland river lockage transport ship, the standard ship type and tonnage are matched using similar neighbor method.The application utilizes laser radar data acquisition and analysis technology, combined with machine learning and other processing means, realizes intelligent, digital channel ship flow statistics, improves the accuracy and work efficiency of ship detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of laser radar technology and intelligent waterway, and particularly relates to a waterway ship flow statistical method and system based on laser radar. BACKGROUND

[0002] The ship flow observation and statistical work of inland waterway is of great significance to improve the observation, statistics, analysis and service level of water transportation, strengthen the prediction and early warning function, and provide solid data guarantee for the economic situation of water transportation. Accurate waterway ship flow data is an important basis for waterway development positioning, infrastructure construction and waterway management strategy.

[0003] At present, the method for statistical observation of inland waterway ship flow mainly relies on manual real-time observation and video monitoring playback. The observation and statistical method of the two methods is original, and the informatization degree is low. Not only the labor and time cost is high, but also the observation and statistical error is large, and the data accuracy is low. At the same time, the data form of different observation points is not unified, and the integration and analysis cost is large. In summary, the digital, network and intelligent waterway ship flow statistics are imminent. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to construct a waterway ship flow statistical method and system based on laser radar. Through the informationization, digitization and intelligentization, the ship flow, heading, size, tonnage and other information are monitored, counted and analyzed, which not only saves manpower and material resources, but also greatly reduces the artificial observation error, and further improves the efficiency of statistical results.

[0005] In order to achieve the above purpose, the present application provides a waterway ship flow statistical method based on laser radar, which specifically includes the following steps:

[0006] S1: waterway ship point cloud data acquisition based on laser radar. The laser radar data acquisition system is deployed on the bridge of inland waterway, and the ship point cloud data in the waterway scene is collected.

[0007] S2: waterway point cloud data background filtering based on spatial coordinate filtering method. According to the installation position of laser radar and the side position of waterway, the coordinates of waterway space in laser radar coordinate system under corresponding inland waterway scene are determined through distance calculation, and the point cloud data background filtering is realized based on waterway space coordinates, only the ship point cloud in waterway is reserved.

[0008] S3: Ship target detection based on weight-considered DBSCAN algorithm. To solve the problem of multiple ships being close to each other and difficult to distinguish, a weight-considered DBSCAN algorithm is proposed to cluster the discrete point cloud data in the channel into different ship targets, realizing ship target detection.

[0009] S4: Ship flow statistics and sailing direction judgment based on edge features. A section is used as a detection means to detect and extract the edge features of the ship bow and the edge features of the ship stern. The number of ships passing through the section is counted through the extraction of edge features, and the sailing direction of the ship is judged by comparing the time difference of the appearance of the edge features of the two sections.

[0010] S5: Ship size extraction based on 3D bounding box model. Based on S3 ship target detection, a 3D bounding box model of the ship is constructed to regularize the shape of the ship, and then more accurate micro information such as the position, size, and height of the ship is calculated.

[0011] S6: Standard ship type matching based on channel prior knowledge and similar neighbor method. Combined with the prior knowledge of standard ship types for inland river transit ships, a standard ship type database is constructed with the standard sizes of all ship types. The ship size calculated in S5 is compared in the standard ship type database through the similar neighbor method to match the accurate standard ship size and tonnage information.

[0012] Further, the laser radar collection system in S1 includes four modules: a mechanical rotating laser radar, a camera, a computer, and a mobile power supply. The laser radar is responsible for collecting channel ship point cloud data, the camera is responsible for collecting ship video data to provide data basis for subsequent comparison with the real scene, the computer is responsible for real-time monitoring and storing data, and the mobile power supply is responsible for powering the radar.

[0013] Further, the channel space in S2 refers to the water surface area where ships can travel. In the channel scene, the point cloud background is filtered out, i.e., the point cloud data on both sides of the river and the bridge (including bridge deck, ground, trees, etc. background objects, as well as pedestrians, vehicles, etc. dynamic targets) is filtered out, and the ship point cloud data on the water surface is retained.

[0014] Further, the S3 considers the weight of DBSCAN algorithm. Since the angular resolution of laser radar in horizontal and vertical directions is 0.1° and 2° respectively, the density of point cloud data of the same ship target in X direction is much greater than that in Y direction in the returned point cloud data. Based on this feature, different weights are given to the distances in X and Y directions in the process of calculating the distance between point clouds, and the weight in X direction is selected to be greater than that in Y direction, that is, more attention is paid to the closeness of ship point cloud in X direction, so as to perform DBSCAN algorithm clustering and realize multi-ship target detection.

[0015] Further, the S4 includes three steps: (1) detecting and extracting the bow edge feature and the stern edge feature; (2) ship flow statistics; (3) ship sailing direction judgment.

[0016] Further, the S4 detects and extracts the bow edge feature and the stern edge feature. When the point cloud data is detected from nothing to something on the selected cross section, the edge feature of the bow is detected, and the time T head is recorded. end .

[0017] Further, the S4 of ship flow statistics. The number of times that the bow or stern edge feature appears on the detection cross section within a certain time range is counted, that is, the number of ships sailing through the cross section within the time range.

[0018] Further, the S4 of ship sailing direction judgment. Two detection cross sections are set, which are the upper cross section and the lower cross section. The time when the ship head edge feature passes through the upper cross section is recorded as T up , and the time when the ship head edge feature passes through the lower cross section is recorded as T down . When T up <T down , it indicates that the direction of ship sailing is from top to bottom, and when T up >T down , it indicates that the direction of ship sailing is from bottom to top.

[0019] Further, the S5 includes two steps: (1) constructing a 3D bounding box model of the ship; (2) calculating the length, width and height of the ship.

[0020] Further, the construction of the ship 3D bounding box model in S5. First, project the clustered ship target point cloud data to the XOY plane, simplify the three-dimensional problem to a two-dimensional problem; then extract the convex hull points of the target point cloud, and construct the 2D bounding box model of the ship according to the minimum area method principle; further consider the Z coordinate size of the points in the 2D bounding box range, find the maximum and minimum Z value, and extend the 2D bounding box model to a 3D bounding box model.

[0021] Further, the calculation of the length, width and height of the ship in S5. The ship length L ship , ship width W ship and ship height H ship are directly extracted by the size of the ship 3D bounding box.

[0022] Further, the channel prior knowledge in S6 refers to the standard size of inland river ships specified in the national standard, that is, “Inland River Transit Ship Standard Ship Type Main Dimension Series” (GB38030-2019), which includes standard ship type main dimensions of oil tankers, barges, container ships, etc.

[0023] Further, the standard ship type matching of the similar neighbor method in S6. This method first performs difference operation on the ship width extracted in S5 and each standard ship width in the standard ship type database, and uses the size of the difference value to measure the similarity. The smaller the difference value, the higher the similarity, so that a most similar standard ship width can be matched. Then the same method is used to match the ship length, and then the standard size (standard ship length and standard ship width) of the ship is obtained. Finally, the tonnage information of the ship is obtained according to the standard ship type.

[0024] Compared with the prior art, the technical scheme of the present application has the following advantages:

[0025] 1. Advanced laser radar channel data acquisition technology and point cloud data processing technology based on machine learning are adopted to realize automatic and intelligent ship flow statistics, which not only saves manpower and resources, but also greatly reduces manual observation errors.

[0026] 2. Compared with video sensors, the laser radar sensor used in the present application has high measurement accuracy and is not affected by environmental light, and can realize all-weather channel monitoring and ship flow statistics.

[0027] 3. The ship flow statistics method constructed in the present application not only has high accuracy, but also can obtain more rich ship information (passing number, sailing direction, size, tonnage, etc.), and realizes unified and informatized management of ship data. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the embodiments and technical solutions of the present application, the drawings needed in the specific embodiments will be briefly introduced below. In the drawings, the same reference signs represent the same or similar components or elements throughout the drawings unless otherwise specified. These drawings only depict some embodiments according to the present application, and should not be considered as limiting the scope of the present application.

[0029] Figure 1 Flow chart for channel ship flow statistics method based on laser radar

[0030] Figure 2 Schematic diagram of laser radar data acquisition system

[0031] Figure 3 Schematic diagram of original point cloud data in channel scene

[0032] Figure 4 Effect diagram of ship point cloud data processed by spatial coordinate filtering method

[0033] Figure 5 Ship target detection result diagram based on DBSCAN algorithm considering weight

[0034] Figure 6 Schematic diagram of ship number change curve in monitoring area, including ship number change and total ship number change in different navigation directions

[0035] Figure 7 Channel ship flow statistics bar chart

[0036] Figure 8 Schematic diagram of ship 3D bounding box model DETAILED DESCRIPTION

[0037] The present application will be further described in detail below with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description.

[0038] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and embodiments.

[0039] Laser radar has been widely used in road traffic perception field with its active distance detection, high space-time resolution and strong anti-interference ability. Therefore, the application aims at the ship flow observation and statistics problem of inland waterway, and constructs a waterway ship flow statistics method and system based on laser radar. The laser radar is deployed on the inland waterway bridge to realize 24-hour automatic monitoring and ship data acquisition. Through machine learning and other information technology means, the active perception of real-time ship number, direction, size and tonnage information is completed.

[0040] Reference Figure 1 As shown in the figure, the waterway ship flow statistics method based on laser radar provided by the embodiment of the application comprises the following processes: first, the laser radar is deployed on the waterway bridge to collect waterway ship point cloud data; second, the spatial coordinate range of the waterway is obtained to complete background filtering of the original point cloud data; further, the DBSCAN algorithm considering weight is used to realize ship target detection; on the one hand, the ship edge features are extracted, the number of ships passing through the monitoring area is counted, and the sailing direction of the ship is judged; on the other hand, the 3D boundary box model of the ship is constructed to extract the size information of the ship; finally, the prior knowledge of the standard ship type of the inland transit ship is combined, and the similar neighbor method is used to match the standard size and tonnage of the ship.

[0041] The specific implementation method of the embodiment is as follows:

[0042] Step S1: waterway ship point cloud data collection and preprocessing based on laser radar. The detailed content of step S1 is described as follows S101-S103:

[0043] S101: laser radar data collection system construction. Referring to Figure 2 As shown in the figure, the laser radar data collection system comprises four modules of laser radar, camera, computer and mobile power supply. The laser radar is responsible for collecting ship point cloud data of the waterway scene, the camera is responsible for collecting ship video data, the computer is responsible for real-time monitoring and storing laser radar and video data, and the mobile power supply is responsible for power supply for the laser radar.

[0044] S102: waterway ship point cloud data collection. The laser radar is deployed on the bridge, and the laser radar is installed in the middle of the bridge body, so as to cover the entire waterway and obtain complete waterway ship point cloud data. Since the laser beam is scanned from the bridge to the lower river, this deployment scheme avoids the mutual shielding problem between ships, and can also obtain relatively complete ship features, thereby providing good quality data for subsequent ship target detection and size calculation.

[0045] S103: Data preprocessing. In order to more conveniently carry out background filtering and construct the 3D bounding box model of the ship, simplify the calculation difficulty of the size of the ship, before carrying out point cloud data processing, through the method of coordinate rotation, the point cloud data is rotated by a corresponding angle around the Z axis, so that the channel direction is parallel to the coordinate Y axis in the new coordinate system. The rotation matrix formula is as follows:

[0046]

[0047] Where (X, Y, Z) is the original coordinate of the point cloud data, (X', Y', Z') is the coordinate of the point cloud data after rotation around the Z axis, and θ is the rotation angle. Referring to Figure 3 The original point cloud data of the channel after coordinate rotation preprocessing is shown in

[0048] Step S2: channel point cloud data background filtering based on spatial coordinate filtering method. The detailed content of step S2 is described as follows S201-S202:

[0049] S201: channel space coordinate acquisition. According to the installation position of the laser radar and the shore position of the channel, the coordinate range of the channel space in the laser radar coordinate system is determined through distance calculation. Through measurement, the channel space range of X direction and Y direction of this embodiment is taken as [-40, 30] and [10, 120] respectively, and since the distance between the bridge and the water surface is 10 m, the channel space range of Z direction is taken as [-10, 0], with the unit of m.

[0050] S202: point cloud data background filtering. Based on the obtained channel space coordinate, the point cloud data background filtering is carried out, the point cloud data on the river bank and the bridge is removed, and the ship point cloud data in the channel is reserved. Referring to Figure 4 The ship point cloud data processed by the spatial coordinate filtering method is shown in

[0051] Step S3: ship target detection based on DBSCAN algorithm considering weight. The detailed content of step S3 is described as follows S301-S303:

[0052] S301: algorithm parameter setting. DBSCAN algorithm is a classic clustering algorithm, which characterizes the tightness of sample distribution through neighborhood radius Eps and minimum number of points in neighborhood MinPts, determines the connectivity between samples from the close relationship between samples, divides samples with connectivity into the same cluster, and then continuously expands the cluster in the space data with noise to finally generate the clustering result. The selection of parameters Eps and MinPts directly affects the result of target detection, in this embodiment, based on the distribution characteristics of the ship point cloud, the neighborhood radius Eps is set to 1.5 m, and the minimum number of points in the neighborhood MinPts is set to 20.

[0053] S302: Algorithm weight setting. According to the characteristics of different angular resolutions of the laser radar in the horizontal and vertical directions, the density of the point cloud data of the same ship target in the X direction is much greater than that in the Y direction. Therefore, the distance measurement method in the DBSCAN algorithm is improved, and different weight values are assigned to the distances in the X and Y directions, and then the clustering algorithm is performed. The distance measurement formula of the DBSCAN algorithm considering the weight is as follows:

[0054]

[0055] where D new is the distance between points in a three-dimensional space, used to represent the closeness of the distribution of two points, and D new is smaller, indicating that the two points are closer in three-dimensional space. Δx, Δy and Δz are the distances between points in the X, Y and Z directions, respectively, and α and β are the weight values of the distances in the X and Y directions. In this embodiment, based on the resolution difference of the laser radar and the distribution characteristics of the ship point cloud data, the weight α of the X direction is set to 1, and the weight β of the Y direction is set to 0.01.

[0056] S303: Ship target detection. In view of the problem that multiple ships are close and difficult to distinguish, the DBSCAN algorithm considering the weight is used for ship target detection. Referring to FIG. 3, the result of ship target detection is shown. Figure 5

[0057] Step S4: Ship flow statistics and sailing direction judgment based on edge features. The detailed content of step S4 is described as follows S401-S404:

[0058] S401: Ship edge feature detection and extraction. The edge features of the ship include the bow edge feature and the stern edge feature. A cross section is selected in the monitoring area to detect the edge features of the ship. When the point cloud data is detected from nothing to something, the bow edge feature is detected, and the time T head is recorded; when the point cloud data is detected from something to nothing, the stern edge feature is detected, and the time T end is recorded.

[0059] S402: Ship flow statistics. In order to avoid the influence of noise on the calculation of the number of ships, the difference between T head and T end is calculated. If T end -T head > 5s, it indicates that a ship passes through the detection cross section, and if T end -T head < 5s, it indicates that it is affected by noise and does not count the ship. In this embodiment, in the 40-minute channel data, there are 21 times that satisfy T end -T head ​> 5s, so the number of vessels passing through the monitoring area in this time range is 21.

[0060] S403: Vessel sailing direction judgment. In order to judge the sailing direction of the vessel, two detection sections are set, which are the upper section and the lower section. The time when the bow edge feature passes through the upper section is recorded as T up , and the time when the bow edge feature passes through the lower section is recorded as T down . When T up < T down , it indicates that the vessel is sailing from top to bottom, and when t up > T down , it indicates that the vessel is sailing from bottom to top.

[0061] S404: Vessel flow statistics visualization. Referring to FIG. 4B, it is a curve of the number of vessels passing through the monitoring area, which includes the number of vessels of different sailing directions and the total number of vessels. Referring to FIG. 4C, it is a column chart of the channel vessel flow statistics in different time periods. Figure 6 Figure 7

[0062] Step S5: Vessel size extraction based on 3D bounding box model. The details of step S5 are described as follows S501-S503:

[0063] S501: Vessel 2D bounding box model construction. In the process of vessel sailing, its motion trajectory is parallel to the water surface, so the vessel target point cloud can be projected to the XOY plane, and then the convex hull point extraction is performed on the target point cloud, and the vessel 2D bounding box model is obtained according to the minimum area method principle.

[0064] S502: Vessel 3D bounding box model construction. The maximum and minimum values of the Z coordinate in the 2D bounding box model point cloud data are calculated, and then the rectangle vertices obtained after the point cloud data is projected to the XOY plane are combined to extend the 2D bounding box model to a 3D bounding box model. Referring to FIG. 5B, it is a schematic diagram of the vessel 3D bounding box model. Figure 8

[0065] S503: Vessel size extraction. The length L ship , the width W ship and the height H ship of the vessel are directly extracted through the size of the vessel 3D bounding box. As shown in Table 1, it is the detected channel passing vessel information, which includes the vessel number, the vessel size (length, width, height) and the vessel sailing direction.

[0066] Table 1 Channel vessel detection information

[0067] Ship number Ship heading Ship length (m) Ship width (m) Ship height (m) 1 1 44.15 8.06 1.23 2 0 66.18 11.89 1.91 3 1 43.86 8.26 1.52 4 1 43.48 10.20 1.39 5 1 47.13 7.51 1.59 6 0 82.41 15.80 1.90 7 0 47.04 9.27 1.88 … … … … …

[0068] ​​​In the navigation direction, 1 indicates that the ship is sailing from top to bottom, and 2 indicates that the ship is sailing from bottom to top.

[0069] Step S6: Standard ship type matching based on prior knowledge of the waterway and the similar nearest neighbor method. The detailed description of step S6 is as follows: S601~S602:

[0070] S601: Construction of Standard Ship Type Database. In this embodiment, the deployment scenario for the lidar is the Beijing-Hangzhou Grand Canal. Therefore, when constructing the standard ship type database, the main dimensions of standard ship types such as oil tankers, barges, and container ships were extracted from the national standard GB38030.2-2019: "Series of Main Dimensions of Standard Ship Types for Inland Waterway Lock-Passing Transport Vessels Part 2: Beijing-Hangzhou Grand Canal and Huaihe River System" to construct a standard ship type database suitable for this embodiment.

[0071] S602: Matching Standard Ship Dimensions and Tonnage. Using the similar nearest neighbor method to match standard ship types, first, the ship's beam W extracted from S5 is... ship The width is compared with the standard ship widths in the standard ship type database using a differential calculation. The width with the smallest error is selected as the most similar standard ship width, denoted as W. GB Then filter out all results related to W from the standard library. GB For the corresponding standard ship beam, the same method is used to find and extract the ship length L. ship The most similar standard ship length is denoted as L. GB Finally, according to W GB and L GB Matching vessel tonnage information. As shown in Table 2, the vessel information after matching with the standard vessel type has been updated with the standard dimensions and tonnage information of the vessel.

[0072] Table 2. Standard Information for Vessels in Waterways

[0073]

[0074] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for counting the number of vessels in a channel based on a laser radar, characterized by, The method comprises the steps of: S1: collecting channel ship point cloud data by using a laser radar; S2: based on the collected channel ship point cloud data, performing background filtering on the point cloud data by a spatial coordinate filtering method, and retaining the ship point cloud data; S3: performing clustering on the filtered ship point cloud data by using a DBSCAN algorithm considering weights, and performing multi-ship target detection; S4: detecting and extracting edge features of the ship, and then counting the ship flow based on the edge features and judging the sailing direction of the ship; the edge features of the ship in step S4 are divided into two parts: a bow edge feature and a stern edge feature, and the method for detecting and extracting the edge features of the ship comprises: selecting a section in the monitoring area, when detecting that the section point cloud data changes from nothing to something, the bow edge feature is detected, and the time T head is recorded ; when detecting that the point cloud data changes from something to nothing, the stern edge feature is detected, and the time T end is recorded ; S5: constructing a ship 3D boundary box model to obtain micro information including ship size; S6: combining channel prior knowledge to construct a standard ship type database, and obtaining standard ship size and tonnage information by similarity neighbor method matching. Before the background filtering in step S2, further comprising: channel space coordinate acquisition, wherein the range of the channel space coordinate is obtained by calculating the distance between the installation position of the laser radar and the shore side position.

2. The method of claim 1, wherein, The background filtering in step S2 comprises filtering static background point cloud data and dynamic point cloud data.

3. The method of claim 1, wherein the method is based on a laser radar. The DBSCAN algorithm considering weights in step S3 gives different weights to the distances in X direction and Y direction in the process of measuring the distance between point clouds, and the weight of X direction is set to 1 and the weight of Y direction is set to 0.

01.

4. The method for counting the number of vessels in a channel based on laser radar according to claim 1, wherein, If T end -T head >5s, it indicates that a ship passes through the detection section, and further, the number of ships is counted, and the ship flow statistics in the monitoring area of the channel are completed.

5. The method for counting the number of vessels in a channel based on laser radar according to claim 1, wherein, The judgment of the sailing direction of the ship in step S4 includes: setting two detection sections, which are an upper section and a lower section respectively, recording the time when the bow edge feature passes through the upper section as T up , and recording the time when the bow edge feature passes through the lower section as T down ; when T up <T down , it indicates that the sailing direction of the ship is from top to bottom, and when T up >T down , it indicates that the sailing direction of the ship is from bottom to top.

6. The method for counting the number of vessels in a channel based on laser radar according to claim 1, wherein, The method for constructing the ship 3D boundary box model in step S5 comprises the following steps: first, projecting the ship target point cloud to XOY plane, then extracting convex hull points from the target point cloud, further constructing a 2D boundary box model according to the minimum area method principle, and finally expanding the 2D boundary box model to a 3D boundary box model according to the maximum and minimum values of Z coordinates in the point cloud data in the 2D boundary box model.

7. The method of claim 1, wherein the method is based on a laser radar. The similarity neighbor method in step S6 comprises the following steps: first, performing difference operation on the ship size extracted in S5 and the ship size in the standard ship type database, then finding the size with the smallest difference from the database as the most similar standard size to the ship, and finally matching the corresponding ship tonnage information according to the standard size.

Images