A target detection method and a marine electronic fence system

Abstract

The application discloses a target detection method and a marine electronic fence system. The method comprises the following steps: acquiring continuous point cloud information collected by a laser radar group; fusing point cloud frames of different laser radars according to the continuous point cloud information to obtain processed continuous point cloud frames; matching target objects in each point cloud frame in the continuous point cloud frames to extract a to-be-detected target; calculating the moving speed of the to-be-detected target, performing mean filtering operation on the to-be-detected target according to the moving speed of the to-be-detected target, and taking the to-be-detected target with an appearance probability greater than a preset probability threshold after filtering as a suspicious target; wherein the appearance probability refers to the proportion of the appearance time of the to-be-detected target within a preset time. The application solves the problems of short action range of a portrait camera and many points in the prior art, and effectively reduces the false positive rate.

Description

Technical Field

[0001] This invention relates to the field of ship target perception and detection, and in particular to a target detection method and a shipboard electronic fence system. Background Technology

[0002] To meet the needs of maritime security and patrol, and to reduce the workload of patrol personnel, it is necessary to set up a perimeter cordon around ships to prevent unauthorized boarding. Current shipboard electronic fences typically consist of a physical infrared electronic fence installed on the main deck, equipped with facial recognition cameras for secondary security. Traditional infrared electronic fences trigger an alarm immediately upon physical contact, resulting in a high false alarm rate. Furthermore, the facial recognition cameras have a short effective range and require a large number of locations. Summary of the Invention

[0003] This invention provides a target detection method and a marine electronic fence system to solve the technical problem of high false alarm rate in traditional marine electronic fences.

[0004] To address the aforementioned technical problems, embodiments of the present invention provide a target detection method for a data processing subsystem of a marine electronic fence system, comprising:

[0005] Acquire continuous point cloud information collected by the lidar array;

[0006] Based on the continuous point cloud information, the point cloud frames of different lidars are fused to obtain the processed continuous point cloud frames.

[0007] The target objects in each point cloud frame of the continuous point cloud frame are matched to extract the target to be detected;

[0008] Calculate the moving speed of the target to be detected, and perform a mean filtering operation on the target to be detected based on the moving speed of the target to be detected. Targets to be detected with a probability of occurrence greater than a preset probability threshold after filtering are identified as suspicious targets. The occurrence probability refers to the proportion of time the target to be detected appears within a preset time period.

[0009] As a preferred embodiment, the target detection method further includes:

[0010] Calculate the movement information of the suspected target, and determine the threat level of the suspected target based on the movement information; wherein, the movement information includes: the distance of the suspected target from the ship and the speed of the suspected target;

[0011] Early warning and control measures are implemented based on the threat level of suspected targets.

[0012] As a preferred embodiment, the lidar group includes: a lidar mounted on the bow and a lidar mounted on the stern.

[0013] The step of fusing point cloud frames from different lidars based on the continuous point cloud information to obtain processed continuous point cloud frames includes:

[0014] Based on the mechanical angles in the point cloud information, determine the point cloud information belonging to the same point cloud frame;

[0015] Based on the acquisition time and acquisition source of the point cloud information, two frames of point cloud information with the same acquisition time but different acquisition sources are fused into a fused point cloud frame; wherein, the acquisition source includes: the lidar at the bow of the ship and the lidar at the stern of the ship.

[0016] Based on the acquisition time of the point cloud frame before fusion corresponding to the fused point cloud frame, the fused point cloud frames are sorted according to the acquisition time to obtain the processed continuous point cloud frames.

[0017] As a preferred embodiment, the step of matching target objects in each point cloud frame of the continuous point cloud frames to extract the target to be detected includes:

[0018] The continuous point cloud frames are rasterized to remove raster point clouds whose relative height difference is less than a first preset threshold or whose number of raster point clouds is less than a second preset threshold.

[0019] The target objects of each point cloud frame in the continuous point cloud frames are obtained through cluster detection.

[0020] Based on the centroid of the target object in each point cloud frame, construct the KD tree of the target object;

[0021] The target objects in the first frame of the continuous point cloud frames are numbered;

[0022] The second frame of the continuous point cloud frames is taken as the current point cloud frame;

[0023] Repeat the target object numbering operation until all target objects in the consecutive point cloud frames have been numbered.

[0024] All target objects with the same number in the continuous point cloud frames are taken as the same target to be detected, and all targets to be detected are obtained;

[0025] The target numbering operation includes:

[0026] For each target object in the current point cloud frame, based on the KD tree of the current target object, search in the previous point cloud frame whether there is a target object whose distance from the current target object is less than or equal to a first preset distance;

[0027] If they exist, continue to search for several target objects in the previous point cloud frame that are closest to the current target object, and use them as reference targets; calculate the point cloud intersection between the current target object and the several reference targets in the previous frame, select the reference target with the most point cloud intersections and the highest intersection-union ratio as the corresponding target of the current target object in the previous frame, and make the current target object inherit the number of the corresponding target;

[0028] If it does not exist, assign a new number to the current target object;

[0029] If all target objects in the current point cloud frame are numbered, then the next frame after the current point cloud frame will be used as the updated current point cloud frame.

[0030] As a preferred embodiment, the step of calculating the moving speed of the target to be detected and performing a mean filtering operation on the target to be detected based on the moving speed includes:

[0031] For each point cloud frame, the current point cloud frame and several consecutive point cloud frames before and after the current point cloud frame are pushed into the first queue; based on the first and last point cloud frames of the first queue, the average speed of the target to be detected in the point cloud frames of the first queue is calculated; the average speed is used as the moving speed of the target to be detected in the current point cloud frame.

[0032] Each point cloud frame containing velocity information is pushed into the second queue in chronological order.

[0033] The point cloud frames in the second queue are grouped into several point cloud frame groups, and mean filtering is performed on the target to be detected in each point cloud frame group.

[0034] The mean filtering operation includes:

[0035] When the number of invalid point cloud frames corresponding to the target to be detected in the point cloud frame group exceeds a preset number, or when the velocity angle between the first and last frames of the target to be detected in the point cloud frame group is greater than a preset angle, the target to be detected is deleted; wherein, the invalid point cloud frames corresponding to the target to be detected in the point cloud frame group are: point cloud frames where the velocity of the target to be detected is zero.

[0036] As a preferred embodiment, the movement information further includes: the location and heading of the suspected target;

[0037] The marine electronic fence system also includes: a display and control subsystem;

[0038] The aforementioned early warning and control based on the threat level of suspicious targets includes:

[0039] Based on the threat level of the suspected target, generate a control signal for an audible and visual alarm corresponding to the threat level;

[0040] The three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspected target from the ship, and the control signal of the suspected target are transmitted to the display and control subsystem. The display and control subsystem displays the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspected target from the ship, and issues an audible and visual alarm corresponding to the threat level according to the control signal.

[0041] Based on the above embodiments, another embodiment of the present invention provides a marine electronic fence system, characterized in that it includes: a data processing subsystem;

[0042] The data processing subsystem includes:

[0043] The information fusion module is used to acquire continuous point cloud information collected by the lidar group, and to fuse the point cloud frames of different lidars according to the continuous point cloud information to obtain the processed continuous point cloud frame.

[0044] The algorithm processing module is used to match the target objects in each point cloud frame in the continuous point cloud frames and extract the target to be detected.

[0045] The suspicious target tracking module is used to calculate the moving speed of the target to be detected, and to perform mean filtering on the target to be detected based on the moving speed of the target to be detected, and to identify the targets to be detected whose probability of appearing after filtering is greater than a preset probability threshold as suspicious targets.

[0046] As a preferred embodiment, the data processing subsystem further includes: a threat assessment module and an early warning control module;

[0047] The threat assessment module is used to calculate the movement information of the suspicious target and determine the threat level of the suspicious target based on the movement information; wherein, the movement information includes: the distance of the suspicious target from the ship and the speed of the suspicious target;

[0048] The early warning module is used to perform early warning control based on the threat level of suspicious targets.

[0049] As a preferred embodiment, the marine electronic fence system further includes: a data acquisition subsystem;

[0050] The data acquisition subsystem is used to acquire continuous point cloud information through a lidar array and transmit the acquired continuous point cloud information to the information fusion module; wherein, the lidar array includes: a lidar installed at the bow and a lidar installed at the stern.

[0051] As a preferred embodiment, the marine electronic fence system further includes: a display and control subsystem;

[0052] The movement information includes: the location and heading of the suspicious target;

[0053] The early warning control module performs early warning control based on the threat level of the suspected target, including:

[0054] Based on the threat level of the suspected target, generate a control signal for an audible and visual alarm corresponding to the threat level;

[0055] The three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspected target from the ship, and the control signal of the suspected target are transmitted to the display and control system, so that the display and control system can display the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspected target from the ship, and issue an audible and visual alarm corresponding to the threat level according to the control signal.

[0056] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0057] Existing marine electronic fence technology typically uses physical infrared electronic fences, equipped with a wide array of facial recognition cameras. Firstly, physical infrared electronic fences trigger alarms upon contact with any physical object, without further detection and analysis, leading to a high false alarm rate. Secondly, facial recognition cameras have a short effective range and require extensive deployment.

[0058] This invention provides a target detection method and a marine electronic fence system. The method includes: acquiring continuous point cloud information collected by a LiDAR array; fusing point cloud frames from different LiDAR arrays based on the continuous point cloud information to obtain processed continuous point cloud frames; matching target objects in each point cloud frame of the continuous point cloud frames to extract targets to be detected; calculating the movement speed of the targets to be detected; and performing mean filtering on the targets to be detected based on their movement speed, identifying targets whose probability of occurrence after filtering is greater than a preset probability threshold as suspicious targets; wherein the probability of occurrence refers to the proportion of time the targets to be detected occur within a preset time period. By using LiDAR to collect point cloud information around the ship, the problem of short effective range and large number of deployment points of human face cameras in the prior art is solved. By further processing the collected point cloud information, analyzing whether the targets to be detected are suspicious targets, and then executing alarm actions based on confirmed suspicious targets, alarm actions are avoided for all targets to be detected, which can effectively reduce the false alarm rate. Attached Figure Description

[0059] Figure 1This is a flowchart illustrating a target detection method according to the present invention;

[0060] Figure 2 This is a schematic diagram of the data processing subsystem of a marine electronic fence system according to the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1

[0063] Please refer to Figure 1 The present invention provides a flowchart illustrating a target detection method for a data processing subsystem of a marine electronic fence system, comprising the following specific steps:

[0064] S1. Acquire continuous point cloud information collected by the lidar group.

[0065] In a preferred embodiment, the lidar group includes a lidar mounted on the bow and a lidar mounted on the stern. The two lidars are mounted on the bow and stern respectively to obtain the maximum field of view, and registration and calibration are performed on the lidars to determine their specific positions relative to each other. Using the center point of the line segment connecting the two lidars as the origin, a planar affine coordinate system for the ship is established using relevant algorithms. Offline calibration is employed, using manually measured position and attitude of the lidars as initial values. Then, an offline calibration object, such as a rectangular calibration plate, is placed at different locations outside the ship for calibration to determine the pose of the two lidars relative to the ship, forming a transformation matrix.

[0066] It should also be noted that, in a specific embodiment, the transmission of point cloud information is via UDP. Due to the instability of UDP transmission, packet loss may occur when acquiring point cloud information. The solution is to enable the UDP packet sequence number function for the two lidar sensors and check the sequence number continuity when acquiring point cloud information. If discontinuous sequence numbers are found, the acquired point cloud information with discontinuous sequence numbers is recorded, and the packet loss rate is calculated. When the packet loss rate exceeds a preset threshold, the size of the data acquisition buffer is adjusted.

[0067] S2. Based on the continuous point cloud information, the point cloud frames of different lidars are fused to obtain the processed continuous point cloud frames.

[0068] In a preferred embodiment, the step of fusing point cloud frames from different lidars based on the continuous point cloud information to obtain processed continuous point cloud frames includes: determining point cloud information belonging to the same point cloud frame based on the mechanical angle in the point cloud information; fusing two frames of point cloud information with the same acquisition time but different acquisition sources into a fused point cloud frame based on the acquisition time and acquisition source of the point cloud information; wherein, the acquisition source includes: lidar at the bow and lidar at the stern of the ship; and sorting the fused point cloud frames according to the acquisition time of the point cloud frames before fusion corresponding to the fused point cloud frame to obtain processed continuous point cloud frames.

[0069] It should be noted that the continuous point cloud information refers to the point cloud information acquired sequentially according to the acquisition time order of the lidar group. Based on the acquired continuous point cloud information, the position of each point cloud in the continuous point cloud information is calculated through the correction table of the lidar corresponding to the continuous point cloud information. According to the acquisition time order of the point cloud information, the position of each point cloud corresponding to the point cloud information, and the mechanical angle corresponding to the point cloud information, the point cloud information between two "zero crossings" of the mechanical angle is determined as the same frame of point cloud information. The continuous point cloud information is divided into temporally continuous point cloud frames according to the above method. Then, two point cloud frames acquired at the same time from the lidar of the bow and the stern of the ship are merged and stitched into one point cloud frame according to the above-pre-calibrated transformation matrix. The stitched point cloud frame is then sorted according to the acquisition time of the corresponding point cloud frame before fusion to obtain the processed continuous point cloud frame.

[0070] S3. Match the target objects in each point cloud frame in the continuous point cloud frames and extract the target to be detected.

[0071] In a preferred embodiment, the step of matching target objects in each point cloud frame of the continuous point cloud frames to extract the target to be detected includes: performing a rasterization operation on the continuous point cloud frames to remove raster point clouds with a relative height difference less than a first preset threshold or a number of raster point clouds less than a second preset threshold; obtaining the target objects in each point cloud frame of the continuous point cloud frames through cluster detection; constructing a KD tree for the target objects based on the centroids of the target objects in each point cloud frame; numbering the target objects in the first frame of the continuous point cloud frames; taking the second frame of the continuous point cloud frames as the current point cloud frame; repeating the target object numbering operation until all target objects in the continuous point cloud frames have been numbered; taking all target objects with the same number in the continuous point cloud frames as the same target to be detected, thus obtaining all targets to be detected; wherein, The target numbering operation includes: for each target object in the current point cloud frame, based on the KD tree of the current target object, searching for whether there exists a target object in the previous point cloud frame whose distance is less than or equal to a first preset distance; if it exists, continuing to search for several target objects in the previous point cloud frame that are closest to the current target object, as reference targets; calculating the point cloud intersection between the current target object and the several reference targets in the previous frame respectively, selecting the reference target with the most point cloud intersections and the highest intersection-union ratio as the corresponding target of the current target object in the previous frame, and making the current target object inherit the number of the corresponding target; if it does not exist, assigning a new number to the current target object; if all target objects in the current point cloud frame have been numbered, then using the next frame of the current point cloud frame as the updated current point cloud frame.

[0072] It should be noted that a rasterization operation is performed on all point cloud frames in the continuous point cloud frames. The relative height difference and the number of point clouds in each raster are counted. Raster filtering is then used to remove raster point clouds with small relative height differences or few point clouds. This operation is used to filter out water waves and tails, and can also be used to filter out ground points. Through the above raster filtering operation, the point cloud clusters corresponding to different objects on the water surface will be well separated, which is beneficial for subsequent clustering detection operations to successfully detect each target.

[0073] Continue to perform clustering detection on the point cloud frames that have undergone the above raster filtering operation. Based on the positional relationship between each point cloud, determine several point cloud clusters. Each point cloud cluster points to a target object. In this way, the target object of each point cloud frame in the continuous point cloud frames is obtained.

[0074] After obtaining the target objects of each point cloud frame through the above processing, a KD tree of the target objects is constructed based on the position of the centroid of the target objects. All target objects in the first frame of the continuous point cloud frames are numbered. After the numbering is completed, the second frame of the continuous point cloud frames is taken as the current point cloud frame. Then, a target numbering operation is performed on the current point cloud frame. During the target numbering operation, when the numbering of the current point cloud frame is completed, the next frame is taken as the updated current point cloud frame. The target numbering operation is repeated on the updated current point cloud frame until all target objects in all the continuous point cloud frames have been numbered, that is, the last frame in the continuous point cloud frames has also completed the target numbering operation. Finally, all target objects with the same number in the continuous point cloud frames are taken as the same target to be detected, and all targets to be detected are obtained.

[0075] It should also be noted that, in a specific embodiment, the target numbering operation is detailed as follows: Target object A, target object B, and target object C are detected in the current point cloud frame; for target object A, a mapping of target object A in the previous frame of the current point cloud frame is established; based on the mapping of the target object and the KD tree constructed for target object A, it is searched in the previous frame of the current point cloud frame to find whether there exists a target object whose distance to the mapping of target object A is less than or equal to 5 meters; if so, it continues to search in the previous frame of the current point cloud frame for the 5 target objects whose mapping distance is closest to that of target object A, and these 5 target objects are used as reference targets; then, the mapping of target object A and these 5 reference targets are calculated respectively. The point cloud intersections are used to select the reference target with the most intersections as its corresponding target. If there are multiple reference targets with the most intersections, the reference target with the highest intersection-union ratio (IU) is selected as the corresponding target. After determining the corresponding target A' of target object A, target object A inherits the number of A'. If there is no target object with a distance of less than or equal to 5 meters between its mapping and target object A in the previous frame of the current point cloud frame, target object A is considered a newly appearing target and is assigned a new number. After completing the numbering operation for target object A, the numbering operations for target object B and target object C are completed in sequence. Finally, all target objects in the current point cloud frame have been numbered, and the next frame of the current point cloud frame is taken as the updated current point cloud frame.

[0076] S4. Calculate the moving speed of the target to be detected, and perform mean filtering on the target to be detected based on the moving speed of the target to be detected. Targets to be detected with a probability of occurrence greater than a preset probability threshold after filtering are identified as suspicious targets. The occurrence probability refers to the proportion of time the target to be detected appears within a preset time period.

[0077] It should be noted that, in a specific embodiment, a target to be detected with a probability greater than 20% within 0.5 seconds is considered a suspicious target, that is, a target to be detected with an appearance time greater than 0.1 seconds is considered a possible target.

[0078] In a preferred embodiment, calculating the moving speed of the target to be detected and performing mean filtering on the target to be detected based on the moving speed includes: for each point cloud frame, pushing the current point cloud frame and several consecutive point cloud frames before and after the current point cloud frame into a first queue; calculating the average speed of the target to be detected in the point cloud frames in the first queue based on the first and last point cloud frames of the first queue; using the average speed as the moving speed of the target to be detected in the current point cloud frame; and pushing each point cloud frame with speed information into a second queue in chronological order. The second queue is divided into several point cloud frame groups, with a number of consecutive point cloud frames in the second queue as a group. A mean filtering operation is then performed on the target to be detected in each point cloud frame group. The mean filtering operation includes deleting the target when the number of invalid point cloud frames corresponding to the target in the point cloud frame group exceeds a preset number, or when the velocity angle between the first and last frames of the target in the point cloud frame group is greater than a preset angle. The invalid point cloud frames corresponding to the target in the point cloud frame group are those where the target's velocity is zero.

[0079] It should be noted that, in a specific embodiment, ten consecutive point cloud frames—the current point cloud frame, the four point cloud frames preceding the current point cloud frame, and the five point cloud frames following the current point cloud frame—are pushed into a first queue. Based on the first and last point cloud frames of the first queue, the average velocity of the target to be detected in the point cloud frames of the first queue is calculated. The average velocity is used as the moving velocity of the target to be detected in the current point cloud frame. The point cloud frames with velocity information are pushed into a second queue in chronological order. The point cloud frames in the second queue are divided into several point cloud frame groups, with five consecutive point cloud frames forming a group. When the velocity of the target to be detected is zero in at least one point cloud frame of the point cloud frame group, or when the velocity angle between the first and last frames of the point cloud frame group is greater than 60 degrees, the target to be detected is deleted.

[0080] In a preferred embodiment, the target detection method further includes: calculating the movement information of the suspicious target, determining the threat level of the suspicious target based on the movement information; wherein the movement information includes: the distance of the suspicious target from the ship and the speed of the suspicious target; and performing early warning control based on the threat level of the suspicious target.

[0081] It should be noted that, in a specific embodiment, a suspicious target within 50 meters of the ship and moving at a speed greater than 0.5 knots is classified as having a threat level of Level 1, while a suspicious target more than 50 meters away from the ship and moving at a speed greater than 0.5 knots is classified as having a threat level of Level 2.

[0082] In a preferred embodiment, the movement information further includes: the bearing and heading of the suspicious target; the marine electronic fence system further includes: a display and control subsystem; the early warning control based on the threat level of the suspicious target includes: generating a control signal for an audible and visual alarm corresponding to the threat level of the suspicious target; transmitting the three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspicious target from the ship, and the control signal of the suspicious target to the display and control subsystem, so that the display and control subsystem displays the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspicious target from the ship, and issues an audible and visual alarm corresponding to the threat level based on the control signal.

[0083] It should be noted that, in a specific embodiment, the display and control subsystem issues a red light alarm for suspicious targets with a threat level of 1 and a yellow light alarm for suspicious targets with a threat level of 2.

[0084] Example 2

[0085] Please refer to Figure 2 This is a schematic diagram of the data processing subsystem of a marine electronic fence system according to an embodiment of the present invention. The system includes:

[0086] The information fusion module is used to acquire continuous point cloud information collected by the lidar group, and to fuse the point cloud frames of different lidars according to the continuous point cloud information to obtain the processed continuous point cloud frame.

[0087] The algorithm processing module is used to match the target objects in each point cloud frame in the continuous point cloud frames and extract the target to be detected.

[0088] The suspicious target tracking module is used to calculate the moving speed of the target to be detected, and to perform mean filtering on the target to be detected based on the moving speed of the target to be detected, and to identify the targets to be detected whose probability of appearing after filtering is greater than a preset probability threshold as suspicious targets.

[0089] In a preferred embodiment, the data processing subsystem further includes: a threat assessment module and an early warning control module;

[0090] The threat assessment module is used to calculate the movement information of the suspicious target and determine the threat level of the suspicious target based on the movement information; wherein, the movement information includes: the distance of the suspicious target from the ship and the speed of the suspicious target;

[0091] The early warning module is used to perform early warning control based on the threat level of suspicious targets.

[0092] In a preferred embodiment, the marine electronic fence system further includes: a data acquisition subsystem;

[0093] The data acquisition subsystem is used to acquire continuous point cloud information through a lidar array and transmit the acquired continuous point cloud information to the information fusion module; wherein, the lidar array includes: a lidar installed at the bow and a lidar installed at the stern.

[0094] In a preferred embodiment, the marine electronic fence system further includes: a display and control subsystem;

[0095] The movement information includes: the location and heading of the suspicious target;

[0096] The early warning control module performs early warning control based on the threat level of the suspected target, including:

[0097] Based on the threat level of the suspected target, generate a control signal for an audible and visual alarm corresponding to the threat level;

[0098] The three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspected target from the ship, and the control signal of the suspected target are transmitted to the display and control system, so that the display and control system can display the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspected target from the ship, and issue an audible and visual alarm corresponding to the threat level according to the control signal.

[0099] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A target detection method, applicable to the data processing subsystem of a marine electronic fence system, characterized in that, include: Acquire continuous point cloud information collected by the lidar array; Based on the continuous point cloud information, the point cloud frames of different lidars are fused to obtain the processed continuous point cloud frames. The target objects in each point cloud frame of the continuous point cloud frame are matched to extract the target to be detected; For each point cloud frame, the current point cloud frame and several consecutive point cloud frames before and after the current point cloud frame are pushed into the first queue; based on the first and last point cloud frames of the first queue, the average speed of the target to be detected in the point cloud frames of the first queue is calculated; the average speed is used as the moving speed of the target to be detected in the current point cloud frame. Each point cloud frame containing velocity information is pushed into the second queue in chronological order. The point cloud frames in the second queue are grouped into several point cloud frame groups, and mean filtering is performed on the target to be detected in each point cloud frame group. The mean filtering operation includes: When the number of invalid point cloud frames corresponding to the target to be detected in the point cloud frame group exceeds a preset number, or when the velocity angle between the first and last frames of the target to be detected in the point cloud frame group is greater than a preset angle, the target to be detected is deleted; wherein, the invalid point cloud frames corresponding to the target to be detected in the point cloud frame group are: point cloud frames where the velocity of the target to be detected is zero. Targets whose appearance probability after filtering is greater than a preset probability threshold are identified as suspicious targets; wherein, the appearance probability refers to the proportion of time the target appears within a preset time period.

2. The target detection method as described in claim 1, characterized in that, Also includes: Calculate the movement information of the suspected target, and determine the threat level of the suspected target based on the movement information; wherein, the movement information includes: the distance of the suspected target from the ship and the speed of the suspected target; Early warning and control measures are implemented based on the threat level of suspected targets.

3. The target detection method as described in claim 2, characterized in that, The lidar group includes: a lidar installed at the bow and a lidar installed at the stern. The step of fusing point cloud frames from different lidars based on the continuous point cloud information to obtain processed continuous point cloud frames includes: Based on the mechanical angles in the point cloud information, determine the point cloud information belonging to the same point cloud frame; Based on the acquisition time and acquisition source of the point cloud information, two frames of point cloud information with the same acquisition time but different acquisition sources are fused into a fused point cloud frame; wherein, the acquisition source includes: the lidar at the bow of the ship and the lidar at the stern of the ship. Based on the acquisition time of the point cloud frame before fusion corresponding to the fused point cloud frame, the fused point cloud frames are sorted according to the acquisition time to obtain the processed continuous point cloud frames.

4. The target detection method as described in claim 3, characterized in that, The step of matching the target objects in each point cloud frame of the continuous point cloud frames and extracting the target to be detected includes: The continuous point cloud frames are rasterized to remove raster point clouds whose relative height difference is less than a first preset threshold or whose number of raster point clouds is less than a second preset threshold. The target objects of each point cloud frame in the continuous point cloud frames are obtained through cluster detection. Based on the centroid of the target object in each point cloud frame, construct the KD tree of the target object; The target objects in the first frame of the continuous point cloud frames are numbered; The second frame of the continuous point cloud frames is taken as the current point cloud frame; Repeat the target numbering operation until all target objects in the consecutive point cloud frames have been numbered. All target objects with the same number in the continuous point cloud frames are taken as the same target to be detected, and all targets to be detected are obtained; The target numbering operation includes: For each target object in the current point cloud frame, based on the KD tree of the current target object, search in the previous point cloud frame whether there is a target object whose distance from the current target object is less than or equal to a first preset distance; If they exist, continue to find the target objects in the previous point cloud frame that are closest to the current target object, and use them as reference targets; calculate the point cloud intersection between the current target object and the reference targets in the previous frame, select the reference target with the most point cloud intersections and the highest intersection-union ratio as the corresponding target of the current target object in the previous frame, and make the current target object inherit the number of the corresponding target; If it does not exist, assign a new number to the current target object; If all target objects in the current point cloud frame are numbered, then the next frame after the current point cloud frame will be used as the updated current point cloud frame.

5. The target detection method as described in claim 4, characterized in that, The movement information also includes: the location and heading of the suspicious target; The marine electronic fence system also includes: a display and control subsystem; The aforementioned early warning and control based on the threat level of suspicious targets includes: Based on the threat level of the suspected target, generate a control signal for an audible and visual alarm corresponding to the threat level; The three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspected target from the ship, and the control signal of the suspected target are transmitted to the display and control subsystem. The display and control subsystem displays the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspected target from the ship, and issues an audible and visual alarm corresponding to the threat level according to the control signal.

6. A marine electronic fence system, characterized in that, include: Data processing subsystem; The data processing subsystem includes: The information fusion module is used to acquire continuous point cloud information collected by the lidar group, and to fuse the point cloud frames of different lidars according to the continuous point cloud information to obtain the processed continuous point cloud frame. The algorithm processing module is used to match the target objects in each point cloud frame in the continuous point cloud frames and extract the target to be detected. The suspicious target tracking module is used to, for each point cloud frame, push the current point cloud frame and several consecutive point cloud frames before and after it into a first queue; calculate the average velocity of the target in the first queue based on the first and last point cloud frames; use the average velocity as the moving velocity of the target in the current point cloud frame; push the point cloud frames with velocity information into a second queue in chronological order; divide the point cloud frames in the second queue into several point cloud frame groups, taking several consecutive point cloud frames as a group, and perform tracking on the target in each point cloud frame group. A mean filtering operation is performed; wherein the mean filtering operation includes: deleting the target when the number of invalid point cloud frames corresponding to the target in the point cloud frame group exceeds a preset number, or when the velocity angle between the first and last frames of the target in the point cloud frame group is greater than a preset angle; wherein, the invalid point cloud frames corresponding to the target in the point cloud frame group are: point cloud frames where the velocity of the target is zero; and identifying targets whose appearance probability after filtering is greater than a preset probability threshold as suspicious targets; wherein, the appearance probability refers to the proportion of the time the target appears within a preset time period.

7. The marine electronic fence system as described in claim 6, characterized in that, The data processing subsystem further includes: a threat assessment module and an early warning module; The threat assessment module is used to calculate the movement information of the suspicious target and determine the threat level of the suspicious target based on the movement information; wherein, the movement information includes: the distance of the suspicious target from the ship and the speed of the suspicious target; The early warning module is used to perform early warning control based on the threat level of suspicious targets.

8. The marine electronic fence system as described in claim 7, characterized in that, Also includes: Data acquisition subsystem; The data acquisition subsystem is used to acquire continuous point cloud information through a lidar array and transmit the acquired continuous point cloud information to the information fusion module; wherein, the lidar array includes: a lidar installed at the bow and a lidar installed at the stern.

9. The marine electronic fence system as described in claim 8, characterized in that, Also includes: Display and control subsystem; The movement information includes: the location and heading of the suspicious target; The early warning control module performs early warning control based on the threat level of the suspected target, including: Based on the threat level of the suspected target, generate a control signal for an audible and visual alarm corresponding to the threat level; The three-dimensional point cloud image information, bearing, heading, speed, threat level, distance of the suspected target from the ship, and the control signal of the suspected target are transmitted to the display and control subsystem. The display and control subsystem displays the three-dimensional point cloud image information, bearing, heading, speed, threat level, and distance of the suspected target from the ship, and issues an audible and visual alarm corresponding to the threat level according to the control signal.

Images