A multi-frame fusion intrusion target detection method based on elevation grid maps
By adopting a multi-frame fusion intrusion target detection method based on elevation grid maps, the mapping algorithm is simplified, storage space and false alarm rate are reduced, and it can adapt to complex terrain, thus solving the detection problem of perimeter intrusion target detection system in harsh environments.
Patent Information
- Application Number
- CN202410936344.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-07-12
AI Technical Summary
Existing perimeter intrusion detection systems have a high false alarm rate in complex and harsh environments. Traditional laser point cloud mapping algorithms in the field of robotics are complex and costly to deploy, making it difficult to meet the requirements of all-weather, all-time, high detection rate and low false alarm rate.
A multi-frame fusion intrusion target detection method based on elevation raster maps is adopted. By constructing a static elevation raster map and combining it with the Kalman filter algorithm to simplify the mapping algorithm, noise points are filtered out using a threshold for the number of raster unit points and a threshold for the target movement distance between adjacent frames to achieve target detection.
It reduces the complexity of the mapping algorithm, reduces storage space, improves detection performance under harsh weather conditions, reduces false alarm rate, and adapts to complex and varied terrain.
Smart Images

Figure CN118552903B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent security technology, and in particular to a multi-frame fusion intrusion target detection method based on elevation grid maps. Background Technology
[0002] Perimeter intrusion detection is an important area of the intelligent security industry, and it has gone through stages such as video surveillance, intrusion detection based on signal feature analysis, and intrusion detection based on multi-sensor collaborative perception.
[0003] The earliest video surveillance systems evolved towards intelligent video analysis and early warning based on deep neural networks with the advancement of deep learning technology. However, they were not optimized for intrusion target detection scenarios in complex and harsh environments. Limited by the characteristics of visible light imaging itself, they could not meet the requirements of all-weather, all-time, high detection rate and low false alarm rate.
[0004] Intrusion target detection based on signal feature analysis mainly targets the signal characteristics of intrusion targets (e.g., people). It detects intrusion targets by means of microwave blockage, laser blockage, electromagnetic field changes and vibration caused by the intrusion. It is the mainstream product direction abroad. However, under extreme weather conditions such as rain, snow and strong winds, each technology system faces different sources of high false alarms. For example, strong winds will induce a large number of false alarms in vibration detection attached to fences, while rain and snow are prone to induce a large number of false alarms in electromagnetic induction detection mechanisms.
[0005] Multi-sensor collaborative sensing forms a collaborative sensing capability by networking multiple similar sensors, which can effectively filter false alarms generated by individual sensors and achieve target intrusion detection with a low false alarm rate. However, the fusion of similar sensors cannot overcome the performance degradation caused by the influence of complex external environments on the inherent mechanism of these sensors.
[0006] As an important type of sensor in autonomous driving, LiDAR has the ability to measure three-dimensional distance and scan point clouds. It can obtain the distance, orientation and height of the target relative to the LiDAR. It is usually used for the three-dimensional detection of the target - that is, to obtain the target's category and cuboid bounding box. The cuboid bounding box contains the target's distance relative to the LiDAR, the target's orientation and the target's three-dimensional dimensions of length, width and height.
[0007] In the field of intrusion target detection, laser point clouds are relatively sparse, lacking necessary texture information, and the model parameters of 3D convolutional neural networks (CNNs) are sparsely distributed. Directly using deep neural networks to process laser point clouds to obtain intrusion target information presents problems such as high computational cost and large storage space requirements. Furthermore, for perimeter intrusion target detection systems deployed in outdoor environments with undulating terrain and varied vegetation, existing systems either rely on attachment to physical defense facilities (such as fences or walls to address terrain undulations) or are adapted to complex terrain through underground installation, both of which face high deployment costs and long deployment cycles. Traditional laser point cloud mapping algorithms in robotics research primarily target scenarios where robots and their installed LiDARs are moving, neglecting scenarios where LiDARs are fixed in one location. Therefore, the mapping algorithms are highly complex and require subsequent map data editing and improvement. Summary of the Invention
[0008] The purpose of this application is to provide a multi-frame fusion intrusion target detection method based on elevation grid maps, which can reduce the computational complexity of point cloud mapping and the storage space of point cloud maps. By introducing a threshold for the number of grid unit point clouds related to the target distance and a multi-frame fusion mechanism, it can filter out false alarms caused by point cloud noise introduced by severe weather, and achieve perimeter intrusion target detection and accurate positioning based on elevation grid maps.
[0009] To achieve the above objectives, this application provides a multi-frame fusion intrusion target detection method based on elevation raster maps, comprising the following steps: S1: constructing a perimeter environmental elevation raster map to obtain a static elevation raster map; S2: performing multi-frame fusion intrusion target detection on the static elevation raster map to obtain detection output information and outputting it, wherein the detection output information includes at least: the number of targets and the actual location of the targets.
[0010] As described above, the sub-steps for constructing a perimeter environmental elevation raster to obtain a static elevation raster are as follows: S11: Initialize the height, variance, number of observations, cumulative point cloud data falling into the raster cell, and number of received laser point cloud frames for each raster cell in the elevation raster being constructed; S12: Increment the number of received laser point cloud frames by one, and then filter the laser point cloud according to the pre-set perimeter defense zone, thereby filtering out laser point clouds located outside the pre-set perimeter defense zone, and using the laser point clouds located within the pre-set perimeter defense zone as the filtered point cloud data; wherein, the pre-set perimeter defense zone includes at least: length, width, and height; S13: Discretize the filtered point cloud data according to the size of the pre-set raster cell in the elevation raster being constructed, to obtain the height of each raster cell and the number of point clouds falling into each raster cell, forming a temporary elevation raster; S14: Process the temporary elevation raster... The number of point clouds in each grid cell of the raster map is counted. If the number of point clouds in a grid cell in the temporary elevation raster map is greater than the threshold N, it is determined whether the observation count of the corresponding grid cell in the elevation raster map being constructed is 0. If not, proceed to S15; if yes, proceed to S16. If the number of point clouds in each grid cell of the temporary elevation raster map is less than the threshold N, the grid cell is ignored, and the process jumps to step S19. S15: Determine whether the height of the grid cell in the temporary elevation raster map is within the height variance range of the corresponding grid cell in the elevation raster map being constructed, where the height variance range = [height - variance, height + variance]. If yes, proceed to S16; if not, proceed to S17. S16: Based on the height and variance of the grid cell in the temporary elevation raster map and the height and variance of the grid cell in the elevation raster map being constructed, a Kalman filter (KF) is applied. The Filter algorithm updates the height and variance of the raster cell in the currently constructed elevation raster map. After the update is complete, proceed to S18; S17: Do not update the height of the raster cell in the currently constructed elevation raster map, and increment the variance of the raster cell in the currently constructed elevation raster map by outlier_var, where outlier_var is used to represent the change in variance when outliers occur; S18: Update the cumulative point cloud data of the raster cell in the currently constructed elevation raster map based on the number of point clouds of the raster cell in the temporary elevation raster map, and increment the observation count of the raster cell in the currently constructed elevation raster map by one, and proceed to S19; S19: Determine whether each raster cell in the temporary elevation raster map has been traversed. If it has, proceed to S20; if not, re-execute S14.S20: Traverse the raster cells in the currently constructed elevation raster map that have been observed more than 0 times and have not been updated by data from the temporary elevation raster map. If any cell's height is not updated in the currently constructed elevation raster map, increment the variance of that cell by time_var and execute S21. Here, time_var represents the change in variance when no point cloud appears. S21: Determine if the number of received laser point cloud frames is greater than a specified threshold M. If not, receive the next frame of laser point cloud and execute S12. If yes, traverse the raster cells in the currently constructed elevation raster map, record the average point cloud count of each raster cell whose cumulative point cloud count exceeds the specified threshold M, and calculate the average point cloud count, minimum height, and maximum height of each raster cell whose cumulative point cloud count exceeds the specified threshold M. Store the maximum height, minimum height, and average point cloud count of each raster cell as a static elevation raster map. The mapping is now complete.
[0011] As described above, the sub-steps for multi-frame fusion intrusion target detection of the static elevation raster map to obtain detection output information are as follows: T1: Load the static elevation raster map, which includes: the maximum height of each raster unit, the minimum height of each raster unit, and the average number of point clouds; T2: Perform range filtering on each received laser point cloud according to the preset three-dimensional range of the perimeter defense zone, thereby filtering out laser point clouds located outside the preset three-dimensional range of the perimeter defense zone, and using laser point clouds located within the preset three-dimensional range of the perimeter defense zone as filtered point cloud data; wherein, the preset three-dimensional range of the perimeter defense zone includes at least: length, width, and height; T3: According to the preset raster of the static elevation raster map... The size of the grid cell discretizes the filtered point cloud data, obtaining the height of each grid cell and the number of points falling into each grid cell, forming a temporary elevation raster map; T4: Determine the number of points in each grid cell of the temporary elevation raster map. If the number of points in the grid cell is greater than the threshold K, check if the number of points in the corresponding grid cell in the static elevation raster map is 0. If not, proceed to T5; if yes, proceed to T6; if the number of points in the grid cell is less than the threshold K, ignore the grid cell and jump directly to step T8; T5: Determine whether the grid cell in the temporary elevation raster map is within the static height range of the corresponding grid cell in the static elevation raster map, where the static height range... Enclosure = [Height - 2x variance, Height + 2x variance]; If not, it indicates that a new target may have appeared in the grid cell, execute T6; if yes, execute T7; T6: Determine whether the height of the grid cell in the temporary elevation raster is greater than the target height threshold. If yes, set the grid cell in the temporary elevation raster to indicate that an intrusion target has appeared, mark the grid cell in the temporary elevation raster as the target grid cell, and obtain the target raster map of the current frame based on the target grid cell, execute T9; if not, execute T8; T7: Determine whether the number of point clouds in the grid cell in the temporary elevation raster is less than the average number of point clouds in the corresponding grid cell of the static elevation raster plus the threshold K. If yes, execute T8; if not, execute T9. T6; T8: Ignore the raster cell in the temporary elevation raster; T9: Determine if all raster cells in the temporary elevation raster have been traversed. If yes, execute T10; otherwise, re-execute T4; T10: Compare the target raster of the current frame with the target raster of the previous frame. If the difference in index or position of each target raster cell in the target raster of the current frame and the target raster of the previous frame is within a set raster cell size, execute T11; if the difference in index or position of each target raster cell in the target raster of the current frame and the target raster of the previous frame is not within a set raster cell size, it indicates that the corresponding raster cell in the target raster of the current frame may correspond to a newly appearing target, execute T12.T11: Retain the corresponding raster cell in the target raster image of the current frame, and set the detection count of this target raster cell in the target raster image of the current frame to the detection count of the corresponding target raster cell in the target raster image of the previous frame + 1, then execute T13; T12: Set the detection count of this target raster cell in the target raster image of the current frame to 1; T13: Determine whether the cumulative number of detected frames has reached the specified cumulative threshold; if yes, execute T14, otherwise execute T15; T14: Set the detection count in the target raster image of the current frame to be greater than 1. T16: Mark the raster cells in the target raster map of the current frame as valid targets and execute T16; or mark the raster cells with a detection count greater than 1 in the target raster map of the current frame as valid targets, and form a target raster map based on the valid targets, and execute T17; T15: Input a new laser point cloud and execute T2; T16: Aggregate the adjacent raster cells marked as valid targets into a convex polygon, and use this convex polygon as the target convex polygon, and execute T17; T17: Convert the target raster map or target polygon into detection output information and output the detection output information.
[0012] As shown above, the height of each grid cell is initialized to 0, the variance is initialized to a preset positive value, the number of observations is initialized to 0, the cumulative point cloud data falling into the grid cell is initialized to 0, and the number of received laser point cloud frames is initialized to 0.
[0013] As shown above, the default positive value is 1000.0.
[0014] As shown above, the maximum height of the raster cell in the elevation raster map being constructed is the height of the raster cell being constructed plus twice the variance of the raster cells being constructed.
[0015] As shown above, the minimum height of the raster cell in the elevation raster map being constructed is the height of the raster cell minus twice the variance of the raster cells in the elevation raster map being constructed.
[0016] As shown above, the average number of point clouds is equal to the cumulative number of point clouds divided by the number of observations.
[0017] As shown above, the grid cell size of the elevation raster map being constructed is preset to 0.5m.
[0018] As shown above, the target height threshold is preset according to the characteristics of the interest target.
[0019] The beneficial effects achieved by this application are as follows: (1) The multi-frame fusion intrusion target detection method based on elevation grid map of this application simplifies the synchronous localization mapping algorithm in the field of robotics, and proposes an automatic and fast two-dimensional elevation grid map construction algorithm based on Kalman filtering, which does not require manual editing of map data in the later stage. Compared with the traditional point cloud synchronous localization mapping algorithm in the field of robotics (such as three-dimensional NDT (Normal Distribution Transform), the complexity of the mapping algorithm is significantly reduced, the amount of map data and storage space are reduced, and it can quickly adapt to the perimeter environment of complex and varied terrain.
[0020] (2) The multi-frame fusion intrusion target detection method based on elevation grid map of this application adopts a multi-frame fusion mechanism that combines the threshold of the number of grid unit point clouds with the threshold of the target movement distance between adjacent frames on the basis of two-dimensional elevation grid map. This can effectively filter out point cloud noise generated by severe weather such as rain, snow and fog with random characteristics, improve the detection performance of the perimeter intrusion target detection system under severe weather conditions, and reduce the false alarm rate. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.
[0022] Figure 1 This is a flowchart of one embodiment of a multi-frame fusion intrusion target detection method based on elevation grid maps. Detailed Implementation
[0023] 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, not all, of the embodiments of the present invention. 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.
[0024] like Figure 1 As shown, this application provides a multi-frame fusion intrusion target detection method based on elevation grid maps, including the following steps: S1: Construct a perimeter environmental elevation raster map to obtain a static elevation raster map.
[0025] Furthermore, the sub-steps for constructing a perimeter environmental elevation raster to obtain a static elevation raster are as follows: S11: Initialize the height, variance, number of observations, cumulative point cloud data falling into the grid cell, and number of received laser point cloud frames for each grid cell in the elevation grid being constructed.
[0026] Furthermore, the height of each grid cell is initialized to 0, the variance is initialized to a preset positive value, the number of observations is initialized to 0, the cumulative point cloud data falling into the grid cell is initialized to 0, and the number of received laser point cloud frames is initialized to 0.
[0027] The variance of the grid cells is the point cloud measurement variance.
[0028] Furthermore, the specific value of the preset positive value depends on the actual situation, and is preferably 1000.0 in this application.
[0029] S12: After incrementing the number of received laser point cloud frames by one, filter the laser point cloud according to the pre-set perimeter defense zone, thereby filtering out laser point clouds located outside the pre-set perimeter defense zone, and using the laser point clouds located within the pre-set perimeter defense zone as the filtered point cloud data; wherein, the pre-set perimeter defense zone includes at least: length, width and height.
[0030] S13: Discretize the filtered point cloud data according to the size of the grid cells preset in the elevation raster map being constructed, and obtain the height of each grid cell and the number of point clouds falling into each grid cell to form a temporary elevation raster map.
[0031] Specifically, the size of the grid cells preset in the elevation grid map (EGM) being constructed depends on the actual situation, and is preferably 0.5m in this application.
[0032] S14: Count the number of point clouds in each grid cell of the temporary elevation raster. If the number of point clouds in the grid cell of the temporary elevation raster is greater than the threshold N, determine whether the observation count of the grid cell in the elevation raster being constructed is 0. If not, proceed to S15; if yes, proceed to S16. If the number of point clouds in each grid cell of the temporary elevation raster is less than the threshold N, ignore the grid cell and jump to step S19.
[0033] Specifically, if the observation count of a grid cell in the elevation raster map being constructed is 0, it indicates that the grid cell is newly observed. If the observation count of a grid cell in the elevation raster map being constructed is not 0, it indicates that the grid cell is not newly observed.
[0034] Furthermore, the number of observations corresponding to the grid cell in the elevation grid map being constructed is updated in step S18.
[0035] Furthermore, a threshold N is pre-set based on the distance between the grid cells of the elevation raster map being constructed and the size of the target of interest.
[0036] Specifically, the threshold N is only related to the distance between the grid cell and the lidar; the greater the distance, the smaller the value of N.
[0037] S15: Determine whether the height of the grid cell in the temporary elevation grid is within the height variance range of the corresponding grid cell in the elevation grid being constructed, where the height variance range = [height - variance, height + variance]; if yes, proceed to S16; if no, proceed to S17.
[0038] Specifically, the height of a raster cell in a temporary elevation raster map refers to the height of a raster cell in the temporary elevation raster map whose point cloud count is greater than a threshold.
[0039] S16: Based on the height and variance of the raster cell in the temporary elevation raster map and the height and variance of the raster cell in the elevation raster map being constructed, the Kalman filter (KF) algorithm is used to update the height and variance of the raster cell in the elevation raster map being constructed. After the update is completed, S18 is executed.
[0040] Specifically, the height variance of a grid cell in the temporary elevation grid (i.e., the point cloud measurement variance) is an intrinsic parameter of the LiDAR and is provided by the manufacturer. The height of a grid cell depends on the actual height of the object at that grid cell location. For example, if there is an object A at the location of that grid cell, then the height of the grid cell is the height of object A.
[0041] The specific method for updating the height and variance of the grid cell in the elevation raster map being constructed using the Kalman filter (KF) algorithm can be achieved using existing methods.
[0042] S17: Do not update the height of the raster cell in the elevation raster map being built, and increase the variance of the raster cell in the elevation raster map being built by outlier_var, where outlier_var is used to represent the change in variance when outliers occur.
[0043] Specifically, the value of outlier_var depends on the actual situation.
[0044] S18: Update the cumulative point cloud data of the grid cell in the currently constructed elevation raster based on the point cloud count of the grid cell in the temporary elevation raster, and increment the observation count of the grid cell in the currently constructed elevation raster by one, then execute S19.
[0045] S19: Determine whether each grid cell in the temporary elevation grid has been traversed. If it has, execute S20; otherwise, execute S14 again.
[0046] Specifically, the length and width of the grid cells, as well as the number of grid cells, are determined by the perimeter defense zone. As an example, the method for determining whether each grid cell in the temporary elevation grid map has been traversed is as follows: a for loop is used in the code to iterate through the cells; completion of the loop indicates that the traversal is complete.
[0047] S20: Traverse the raster cells in the elevation raster map that have been observed more than 0 times and have not been updated by the data of the temporary elevation raster map. If the height of the raster cell in the elevation raster map is not updated, the variance of the raster cell in the elevation raster map is increased by time_var, and S21 is executed. Here, time_var is used to represent the change in variance when no point cloud appears.
[0048] Specifically, the value of time_var should be set according to the actual situation.
[0049] A raster cell that is not updated by the data in the temporary elevation raster is a raster cell whose point cloud count is less than the threshold N in the temporary elevation raster and therefore will not be updated.
[0050] S21: Determine whether the number of received laser point cloud frames is greater than the specified threshold M. If not, receive the next frame of laser point cloud and execute S12. If yes, traverse the grid cells of the elevation raster map being constructed, record the average number of points in each grid cell whose cumulative point cloud count exceeds the specified threshold M, and calculate the average number of points, minimum height, and maximum height of each grid cell whose cumulative point cloud count exceeds the specified threshold M. Store the maximum height, minimum height, and average number of points in each grid cell as a static elevation raster map (EGM), and the mapping is complete.
[0051] Furthermore, the maximum height of the raster cell in the elevation raster map being constructed is calculated as the height of the raster cell being constructed plus twice the variance of the raster cells being constructed.
[0052] Specifically, the maximum height of each grid cell = height + 2 times the variance; Furthermore, the minimum height of the raster cell in the elevation raster map being constructed is determined by subtracting twice the variance of the raster cells in the elevation raster map being constructed.
[0053] Specifically, the minimum height of each grid cell = height - 2 times the variance.
[0054] Furthermore, the average number of point clouds is equal to the cumulative number of point clouds divided by the number of observations.
[0055] S2: Perform multi-frame fusion intrusion target detection on a static elevation raster map, obtain detection output information, and output it. The detection output information includes at least the number of targets and the actual location of the targets.
[0056] Specifically, the number of targets refers to the total number of targets. If convex polygon aggregation has been performed, it refers to the number of convex polygons; otherwise, it refers to the number of grid cells. The actual position of the target refers to its X and Y coordinates relative to the lidar, obtained from the grid cell number.
[0057] Furthermore, the intrusion target detection is performed by multi-frame fusion on the static elevation raster map to obtain detection output information, and the output sub-steps are as follows: T1: Load a static elevation raster map, which includes: the maximum height of each raster cell, the minimum height of each raster cell, and the average number of point clouds.
[0058] Specifically, the maximum height of each grid cell = height + 2 times the variance; The minimum height of each grid cell = height - 2 times the variance.
[0059] The average number of point clouds is equal to the cumulative number of point clouds divided by the number of observations.
[0060] T2: Perform range filtering on each received laser point cloud according to the pre-set three-dimensional range of the perimeter defense zone, thereby filtering out laser point clouds located outside the pre-set three-dimensional range of the perimeter defense zone, and using laser point clouds located within the pre-set three-dimensional range of the perimeter defense zone as filtered point cloud data; wherein, the pre-set three-dimensional range of the perimeter defense zone includes at least: length, width and height.
[0061] T3: Discretize the filtered point cloud data according to the preset grid cell size of the static elevation raster map to obtain the height of each grid cell and the number of point clouds falling into each grid cell, thus forming a temporary elevation raster map.
[0062] Specifically, the size of the preset grid cell in the static elevation grid map (EGM) depends on the actual situation, and is preferably 0.5m in this application.
[0063] T4: Determine the number of point clouds in each grid cell of the temporary elevation raster map. If the number of point clouds in the grid cell is greater than the threshold K, determine whether the number of point clouds in the corresponding grid cell in the static elevation raster map is 0. If not, proceed to T5; if yes, proceed to T6. If the number of point clouds in the grid cell is less than the threshold K, ignore the grid cell and jump directly to step T8.
[0064] Specifically, if the number of point clouds in a grid cell of a static elevation raster is 0, while the number of point clouds in the corresponding grid cell of a temporary elevation raster is greater than the threshold K, it indicates that a new target may have appeared in that grid cell, and it is necessary to continue to determine whether the grid cell in the temporary elevation raster is greater than the target height threshold.
[0065] The threshold K is related to the distance between the grid cells and the lidar, as well as the size of the target of interest. The greater the distance, the smaller the value of K; the smaller the target of interest, the smaller the value of threshold K.
[0066] Furthermore, the threshold K is different from the threshold N.
[0067] T5: Determine whether the grid cell in the temporary elevation grid is within the static height range of the corresponding grid cell in the static elevation grid, where the static height range = [height - 2 times variance, height + 2 times variance]; if not, it means that a new target may have appeared in the grid cell, and execute T6; if yes, execute T7.
[0068] T6: Determine whether the height of the grid cell in the temporary elevation grid is greater than the target height threshold. If yes, set the grid cell in the temporary elevation grid as the target grid cell, mark the grid cell in the temporary elevation grid as the target grid cell, obtain the target grid cell of the current frame based on the target grid cell, and execute T9; otherwise, execute T8.
[0069] Specifically, a target height threshold can be preset based on the characteristics of the target of interest. For example, if the height of target B (e.g., a person) is greater than i meters, then the target height threshold is set to be greater than i meters.
[0070] Specifically, if the height of a grid cell in the temporary elevation raster is greater than the target height threshold, an intrusion target is identified at that grid cell, and the grid cell is marked as the target grid cell. If the height of the grid cell in the temporary elevation raster is less than or equal to the target height threshold, the grid cell is ignored.
[0071] Furthermore, the target raster map of the current frame is formed from the target raster cells obtained from the temporary elevation raster map, which is: the target raster map of the current frame obtained from the current laser point cloud frame. Each raster cell uses its highest bit (0 or 1) to indicate whether a target exists within that cell; a highest bit of 1 indicates no target, and a highest bit of 0 indicates a target exists.
[0072] Furthermore, 7 bits are maintained for the number of target detections (i.e., the lower 7 bits of a byte are used to maintain how many times the detections were performed, up to a maximum of 127 times). Therefore, each cell grid is represented by one byte, but it is not limited to each cell grid being represented by one byte. In this application, it is preferred that each cell grid be represented by one byte.
[0073] T7: Determine whether the number of point clouds in the temporary elevation raster is less than the average number of point clouds in the corresponding raster in the static elevation raster plus the threshold K. If yes, execute T8; otherwise, execute T6.
[0074] Specifically, if the number of point clouds in a grid cell in the temporary elevation raster map is greater than the average number of point clouds in the corresponding grid cell in the static elevation raster map plus the threshold K, it indicates that a new target may have appeared in the grid cell, and it is necessary to continue to determine whether the grid cell in the temporary elevation raster map is greater than the target height threshold.
[0075] If the number of points in a grid cell in a temporary elevation raster is less than or equal to the average number of points in the corresponding grid cell in a static elevation raster plus a threshold K, it indicates that the fluctuation in the number of points may be due to point cloud measurement errors and raster discretization. Therefore, this grid cell should be ignored.
[0076] T8: This cell in the temporary elevation raster map is ignored.
[0077] T9: Determine whether all areas of the temporary elevation raster have been traversed. If yes, execute T10; otherwise, re-execute T4.
[0078] Specifically, the length and width of the grid cells, as well as the number of grid cells, are determined by the perimeter defense zone. As an example, the method for determining whether each grid cell in the temporary elevation grid map has been traversed is as follows: a for loop is used in the code to iterate through the cells; completion of the loop indicates that the traversal is complete.
[0079] T10: Compare the target raster image of the current frame with the target raster image of the previous frame. If the difference between the index or position of each target raster unit in the target raster image of the current frame and the target raster image of the previous frame is within a set raster unit size, then execute T11; if the difference between the index or position of each target raster unit in the target raster image of the current frame and the target raster image of the previous frame is not within a set raster unit size, it means that the corresponding raster unit in the target raster image of the current frame may correspond to a newly appearing target, then execute T12.
[0080] If the difference between the indices or positions of the target raster cells in the current frame and the target raster cell in the previous frame is within a set raster cell size, it means that the target movement distance between adjacent frames is no greater than the set raster cell size. The specific value of the set raster cell size is set according to the actual situation.
[0081] T11: Retain the corresponding raster cell in the target raster map of the current frame, and set the detection count of the target raster cell in the target raster map of the current frame to the detection count of the corresponding target raster cell in the target raster map of the previous frame + 1, then execute T13.
[0082] Specifically, the number of probes is assigned in step T12 and updated by incrementing the number of probes by 1 in step T11.
[0083] T12: Sets the number of probes for this target raster cell in the target raster map of the current frame to 1.
[0084] T13: Determine whether the cumulative number of detected frames has reached the specified cumulative threshold; if yes, execute T14; otherwise, execute T15. Specifically, the specific value of the specified cumulative threshold is set according to the actual situation of the trade-off between detection alarm delay requirements and detection stability, and is preferably 3 frames in this application.
[0085] T14: Mark the raster cells in the target raster map of the current frame that have been detected more than once as valid targets, and execute T16; or mark the raster cells in the target raster map of the current frame that have been detected more than once as valid targets, and form a target raster map based on the valid targets, and execute T17.
[0086] Specifically, the final target raster can be generated directly based on the preferred target (i.e., directly execute T17). Alternatively, the final target raster can be generated by reducing the number of output targets (i.e., directly execute T16).
[0087] T15: Input the new laser point cloud and execute T2.
[0088] T16: Aggregate adjacent raster cells marked as valid targets into a convex polygon, and use this convex polygon as the target convex polygon, then execute T17.
[0089] The aggregation of convex polygons requires additional computational resources, but it can simplify subsequent target data processing.
[0090] Furthermore, the Monotone Chain Convex Hull (MCCH) algorithm is used to aggregate adjacent grid cells marked as valid targets into a convex polygon, but it is not limited to the Monotone Chain Convex Hull algorithm. In this application, the Monotone Chain Convex Hull algorithm is preferred.
[0091] T17: Convert the target raster or target polygon into detection output information and output the detection output information.
[0092] The beneficial effects achieved by this application are as follows: (1) The multi-frame fusion intrusion target detection method based on elevation grid map of this application simplifies the synchronous localization mapping algorithm in the field of robotics, and proposes an automatic and fast two-dimensional elevation grid map construction algorithm based on Kalman filtering, which does not require manual editing of map data in the later stage. Compared with the traditional point cloud synchronous localization mapping algorithm in the field of robotics (such as three-dimensional NDT (Normal Distribution Transform), the complexity of the mapping algorithm is significantly reduced, the amount of map data and storage space are reduced, and it can quickly adapt to the perimeter environment of complex and varied terrain.
[0093] (2) The multi-frame fusion intrusion target detection method based on elevation grid map of this application adopts a multi-frame fusion mechanism that combines the threshold of the number of grid unit point clouds with the threshold of the target movement distance between adjacent frames on the basis of two-dimensional elevation grid map. This can effectively filter out point cloud noise generated by severe weather such as rain, snow and fog with random characteristics, improve the detection performance of the perimeter intrusion target detection system under severe weather conditions, and reduce the false alarm rate.
[0094] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the scope of protection of this application is intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application. Obviously, those skilled in the art can make various alterations and variations to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of protection of this application and its equivalents, this application also intends to include these modifications and variations.
Claims
1. A multi-frame fusion intrusion target detection method based on elevation raster maps, characterized in that, Includes the following steps: S1: Construct a perimeter environmental elevation raster map to obtain a static elevation raster map; S2: Perform multi-frame fusion intrusion target detection on a static elevation raster map, obtain detection output information, and output it. The detection output information includes at least the number of targets and the actual location of the targets. The sub-steps for constructing a perimeter environmental elevation raster map to obtain a static elevation raster map are as follows: S11: Initialize the height, variance, number of observations, cumulative point cloud data falling into the grid cell, and number of received laser point cloud frames for each grid cell in the elevation grid map being constructed; S12: After incrementing the number of received laser point cloud frames by one, filter the laser point cloud according to the pre-set perimeter defense zone, thereby filtering out laser point clouds located outside the pre-set perimeter defense zone, and using the laser point clouds located within the pre-set perimeter defense zone as the filtered point cloud data; wherein, the pre-set perimeter defense zone includes at least: length, width and height. S13: Discretize the filtered point cloud data according to the size of the grid cells preset in the elevation grid map being constructed, and obtain the height of each grid cell and the number of point clouds falling into each grid cell to form a temporary elevation grid map. S14: Count the number of point clouds in each grid cell of the temporary elevation raster. If the number of point clouds in the grid cell of the temporary elevation raster is greater than the threshold N, determine whether the observation count of the grid cell in the elevation raster being constructed is 0. If not, proceed to S15; if yes, proceed to S16. If the number of point clouds in each grid cell of the temporary elevation raster is less than the threshold N, ignore the grid cell and jump to step S19. S15: Determine whether the height of the grid cell in the temporary elevation raster map is within the height variance range of the corresponding grid cell in the elevation raster map being constructed, where the height variance range = [height - variance, height + variance]; if yes, proceed to S16; if no, proceed to S17. S16: Based on the height and variance of the raster cell in the temporary elevation raster map and the height and variance of the raster cell in the elevation raster map being constructed, the Kalman filter (KF) algorithm is used to update the height and variance of the raster cell in the elevation raster map being constructed. After the update is completed, S18 is executed. S17: Do not update the height of the raster cell in the elevation raster map being built, and increase the variance of the raster cell in the elevation raster map being built by outlier_var, where outlier_var is used to represent the change in variance when outliers occur; S18: Update the cumulative point cloud data of the grid cell in the currently constructed elevation raster based on the point cloud count of the grid cell in the temporary elevation raster, and increment the observation count of the grid cell in the currently constructed elevation raster by one, then execute S19. S19: Determine whether each grid cell in the temporary elevation grid has been traversed. If it has, execute S20; otherwise, execute S14 again. S20: Traverse the raster cells in the elevation raster map that have an observation count greater than 0 and have not been updated by the data of the temporary elevation raster map. If the height of the raster cell in the elevation raster map is not updated, the variance of the raster cell in the elevation raster map is increased by time_var, and S21 is executed. Here, time_var is used to represent the change in variance when no point cloud appears. S21: Determine whether the number of received laser point cloud frames is greater than the specified threshold M. If not, receive the next frame of laser point cloud and execute S12. If yes, traverse the grid cells of the elevation raster map being constructed, record the average number of points in each grid cell whose cumulative point cloud count exceeds the specified threshold M, and calculate the average number of points, minimum height, and maximum height of each grid cell whose cumulative point cloud count exceeds the specified threshold M. Store the maximum height, minimum height, and average number of points in each grid cell as a static elevation raster map, and the mapping is completed.
2. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1, characterized in that, The following are the sub-steps for performing multi-frame fusion intrusion target detection on a static elevation raster map to obtain detection output information: T1: Load a static elevation raster map, which includes: the maximum height of each raster cell, the minimum height of each raster cell, and the average number of point clouds; T2: Perform range filtering on each received laser point cloud according to the pre-set three-dimensional range of the perimeter defense zone, thereby filtering out laser point clouds located outside the pre-set three-dimensional range of the perimeter defense zone, and using laser point clouds located within the pre-set three-dimensional range of the perimeter defense zone as the filtered point cloud data; wherein, the pre-set three-dimensional range of the perimeter defense zone includes at least: length, width and height; T3: Discretize the filtered point cloud data according to the size of the preset grid cells of the static elevation raster map to obtain the height of each grid cell and the number of point clouds falling into each grid cell, thus forming a temporary elevation raster map. T4: Determine the number of point clouds in each grid cell of the temporary elevation raster map. If the number of point clouds in the grid cell is greater than the threshold K, determine whether the number of point clouds in the corresponding grid cell in the static elevation raster map is 0. If not, proceed to T5; if yes, proceed to T6. If the number of point clouds in the grid cell is less than the threshold K, ignore the grid cell and jump directly to step T8. T5: Determine whether the grid cell in the temporary elevation grid is within the static height range of the corresponding grid cell in the static elevation grid, where the static height range = [height - 2 times variance, height + 2 times variance]; if not, it means that a new target may have appeared in the grid cell, and execute T6; if yes, execute T7. T6: Determine whether the height of the grid cell in the temporary elevation grid is greater than the target height threshold. If yes, set the grid cell in the temporary elevation grid as the target grid cell, mark the grid cell in the temporary elevation grid as the target grid cell, obtain the target grid cell of the current frame based on the target grid cell, and execute T9; otherwise, execute T8. T7: Determine whether the number of point clouds in the temporary elevation raster is less than the average number of point clouds in the corresponding raster in the static elevation raster plus the threshold K. If yes, execute T8; otherwise, execute T6. T8: This grid cell in the temporary elevation grid is ignored; T9: Determine whether the temporary elevation raster map has been traversed. If yes, execute T10; otherwise, re-execute T4. T10: Compare the target raster image of the current frame with the target raster image of the previous frame. If the difference between the index or position of each target raster unit in the target raster image of the current frame and the target raster image of the previous frame is within a set raster unit size, then execute T11; if the difference between the index or position of each target raster unit in the target raster image of the current frame and the target raster image of the previous frame is not within a set raster unit size, it means that the corresponding raster unit in the target raster image of the current frame may correspond to a newly appearing target, then execute T12. T11: Retain the corresponding grid cell in the target grid map of the current frame, and set the detection count of the target grid cell in the target grid map of the current frame to the detection count of the corresponding target grid cell in the target grid map of the previous frame + 1, and execute T13; T12: Sets the detection count of the target raster cell in the target raster map of the current frame to 1; T13: Determine whether the cumulative number of detected frames has reached the specified cumulative threshold; if yes, execute T14; otherwise, execute T15. T14: Mark the grid cells in the target grid map of the current frame that have a detection count greater than 1 as valid targets and execute T16; or mark the grid cells in the target grid map of the current frame that have a detection count greater than 1 as valid targets and form a target grid map based on the valid targets and execute T17. T15: Input the new laser point cloud and execute T2; T16: Aggregate adjacent grid cells marked as valid targets into a convex polygon, and use this convex polygon as the target convex polygon, then execute T17; T17: Convert the target raster or target polygon into detection output information and output the detection output information.
3. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1, characterized in that, Initialize the height of each grid cell to 0, the variance to a preset positive value, the number of observations to 0, the cumulative point cloud data falling into the grid cell to 0, and the number of received laser point cloud frames to 0.
4. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 3, characterized in that, The default positive value is 1000.
0.
5. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1 or 2, characterized in that, The maximum height of the raster cell in the elevation raster map being constructed is calculated as the height of the raster cell being constructed plus twice the variance of the raster cells being constructed.
6. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1 or 2, characterized in that, The minimum height of the raster cell in the elevation raster map being constructed is calculated as the height of the raster cell minus twice the variance of the raster cells in the elevation raster map being constructed.
7. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1 or 2, characterized in that, The average number of point clouds is equal to the cumulative number of point clouds divided by the number of observations.
8. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 1, characterized in that, The elevation raster map being constructed has a preset raster cell size of 0.5m.
9. The multi-frame fusion intrusion target detection method based on elevation grid maps according to claim 2, characterized in that, Pre-set the target height threshold based on the characteristics of the target of interest.
Citation Information
Patent Citations
Target motion state detection method and device, mobile device and storage medium
CN116309693A
Elevation estimation method, device and system for foot-type robot
CN116540206A
