Array point cloud real-time outlier noise removal method and system based on sliding completion

Abstract

The invention provides an array point cloud real-time outlier noise removal method and system based on sliding completion, and relates to the field of point cloud outlier noise processing. The method comprises the following steps: preprocessing initially input multi-frame array point clouds to generate incomplete point clouds and storing the incomplete point clouds in a sliding window; obtaining subsequent input incomplete point clouds of each frame, taking removed points as feature points of the current frame, and taking a union set of the point clouds in the sliding window as to-be-filled point clouds; designing a complementation strategy based on an array discrete probability model to screen the feature points; filling the screened feature points into the incomplete point cloud of the current frame, namely the denoised point cloud of the current frame, and completing the denoising of the point cloud of one frame; after denoising of the current frame point cloud is completed, the corresponding incomplete point cloud is added to the tail of the sliding window queue, and the point cloud at the head of the sliding window queue is removed at the same time, so that dynamic updating of the sliding window is realized. According to the invention, the problems of weak adaptability and low denoising precision during array point cloud outlier noise processing in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of array point cloud outlier noise processing, and in particular to a method and system for removing real-time outlier noise from array point clouds based on sliding completion. Background Art

[0002] Point cloud is a three-dimensional space point data set obtained by laser radar. It is a set of three-dimensional space coordinates of target surface points obtained by laser radar emitting laser to the surface of the target to be measured and receiving reflected light. Single-Photon Avalanche Photodiode (SPAD) array laser radar is a type of non-scanning laser radar with single-photon detection capability, that is, only one photon in the echo signal can trigger the detector. A single laser pulse can form a complete three-dimensional image of a frame of the target in real time, which is superior to the mechanical scanning architecture of traditional laser radar in terms of detection efficiency. SPAD array laser radar has single-photon sensitivity, but this also means that the system is greatly affected by noise. When it is in a strong background light detection environment, the laser echo signal will be submerged in the noise signal, resulting in a high proportion of outlier noise in the acquired point cloud; in addition, the single-frame measured point cloud obtained by SPAD array laser radar is extremely sparse, and the target characteristics are not obvious.

[0003] Existing algorithms are mainly designed based on dense synthetic or measured point clouds, and their adaptability to outlier noise removal in sparse array point clouds is weak. The characteristic differences between noise and real points in sparse point clouds are also difficult to capture. Therefore, there is an urgent need for an outlier noise removal technology for SPAD array lidar sparse point clouds on the market, which can achieve high-precision denoising of array sparse point clouds while ensuring real-time performance. Summary of the invention

[0004] Purpose of the invention: To propose a method for real-time outlier noise removal of array point cloud based on sliding completion, and further propose a system for implementing the above method, which can overcome the defect that the prior art cannot effectively remove the outlier noise of sparse point cloud of SPAD array lidar, and realize the accurate extraction of valid points in the sparse point cloud of the array.

[0005] In order to solve the above technical problems, a first aspect of the present invention provides a method for removing outlier noise from an array point cloud in real time based on sliding completion, comprising the following steps:

[0006] For the initial input multi-frame array point cloud, the layered outlier noise removal method is used for preprocessing to generate incomplete point clouds and store them in the sliding window. When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established.

[0007] After the sliding window is established, for each frame of point cloud input subsequently, the incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method. The removed points are regarded as the feature points of the current frame. The union of the point clouds in the sliding window is used as the point cloud to be filled, and a completion strategy based on the array discrete probability model is designed to screen the feature points.

[0008] Fill the filtered feature points into the incomplete point cloud of the current frame, which is the denoised point cloud of the current frame, completing the denoising of one frame of point cloud;

[0009] After the current frame point cloud denoising is completed, its corresponding incomplete point cloud will be added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue will be removed to achieve dynamic update of the sliding window.

[0010] As a preferred solution of the first aspect, the multi-frame array point cloud initially input is preprocessed by using a layered outlier noise removal method, specifically:

[0011] A bounding box is established for the input point cloud. The boundaries of the bounding box are the upper and lower bounds of the point cloud's three-dimensional coordinates. The point cloud in the bounding box is layered. The calculation formula for the thickness of a single layer is as follows:

[0012]

[0013] Where n1 is the total number of points in the input point cloud, S xoy , S yoz , S xoz They are the projection areas of the bounding boxes on the xoy, yoz, and xoz planes, respectively, and the calculation formula is as follows:

[0014]

[0015] In the formula, x max is the upper limit of the X-axis coordinate of the input point cloud, x min is the lower bound of the X-axis coordinate of the input point cloud, y max is the upper bound of the Y-axis coordinate of the input point cloud, y min is the lower bound of the Y-axis coordinate of the input point cloud, z max is the upper bound of the Z-axis coordinate of the input point cloud, z min It is the lower bound of the Z-axis coordinate of the input point cloud;

[0016] The input point cloud is layered along the Z axis according to the single layer thickness, and the total number of layers is:

[0017] l=|z max -z min | / δ

[0018] For all points in each layer of the input point cloud, the influence of the Z axis is ignored and a bounding box is established point by point. If there is only the point itself in the bounding box, the point is regarded as an outlier and removed. The side length of the established bounding box is:

[0019]

[0020] Where α is the bounding box threshold;

[0021] After performing outlier noise judgment operations on the input point cloud layer by layer and point by point, the point cloud preprocessing operation is completed.

[0022] As a preferred solution for the first aspect, after completing the point cloud preprocessing operation, LSOR with strict parameter configuration is used to perform secondary processing on it. Under the parameter configuration, the outlier noise of the point cloud after the preprocessed point cloud is secondary processed by LSOR is minimized, but some detailed features will be lost, so the final result is an incomplete point cloud.

[0023] As a preferred solution of the first aspect, when the number of incomplete point clouds generated is equal to the capacity of the sliding window, the initial sliding window is established, specifically:

[0024] Assume that the initial input multi-frame point cloud is The capacity of the sliding window is n; for each frame of the input point cloud P i , use the layered outlier noise removal method to preprocess P′ and generate an incomplete point cloud P i ′, and stored in the sliding window;

[0025] When the number of incomplete point clouds generated is equal to the capacity of the sliding window, the initial sliding window is established.

[0026] As a preferred solution of the first aspect, for each frame of point cloud subsequently input, an incomplete point cloud of the current frame is obtained based on a layered outlier noise removal method, the removed points are regarded as feature points of the current frame, and the union of the point clouds in the sliding window is used as the point cloud to be filled, specifically:

[0027] Assume that the subsequent input multi-frame point cloud is The union of the point clouds in the sliding window is P′; for the point cloud P of the current frame i The incomplete point cloud generated after preprocessing using the layered outlier noise removal method is P i ′, feature point set F = P i -P i ′;

[0028] For any point f in the feature point set F i , get its neighbor point set in P′

[0029] Neighbor point pj The relative average photoelectron number is defined as:

[0030]

[0031] Where E is the energy of a single laser pulse, λ is the laser wavelength, h is Planck's constant, c is the speed of light, ρ is the reflectivity, and k is R is the lens loss, k M is the atmospheric transmission loss, k L is the laser emission loss, A is the receiving objective lens area, d ij f i With p j The distance between the detector and the detector is γ, γ is the detector filling factor, and η is the detector quantum efficiency;

[0032] Neighbor point p j The corresponding trigger interval is:

[0033]

[0034] Where c is the speed of light, T a The length of a single time interval.

[0035] As a preferred solution of the first aspect, the design screens the feature points based on a completion strategy of an array discrete probability model, specifically:

[0036] If the laser echo photon is j The probability of the event occurring is:

[0037]

[0038] In the formula, is the average noise photoelectron number, which is calculated as follows:

[0039]

[0040] In the formula, I sun is the solar radiation intensity after bandpass filtering, k′ M is the atmospheric transmission loss between the sun and the target, It represents the diffuse reflection angle of the target surface to the incident sunlight, and θ is the field of view angle of the detector;

[0041] In the point set After calculating the trigger probability of all points, the mean function is used to convert the trigger probability to point f i The noise fraction S i :

[0042]

[0043] If point fi If the noise score of is less than the set threshold ψ, the point is considered as a noise point and removed;

[0044] After noise point detection and removal are completed for all feature points in the feature point set F, the feature point screening step is completed.

[0045] As a preferred solution of the first aspect, the filtered feature points are filled into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, to complete the denoising of a frame of point cloud, specifically:

[0046] Assume that the filtered feature point set is F′, then the point cloud after denoising of the current frame is P i ″=F′∪P i ′.

[0047] As a preferred solution of the first aspect, after the denoising of the current frame point cloud is completed, the corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to realize the dynamic update of the sliding window, specifically:

[0048] Let the incomplete point cloud of the current frame be P i ′, then the sliding window is {P′ i-n ,P′ i-n+1 ,...,P′ i-1};

[0049] The denoised point cloud P i ″Add to the end of the sliding window queue, and get {P′ i-n ,P′ i-n+1 ,...,P′ i-1 ,P′ i}; Remove the point cloud at the head of the sliding window team, and get {P′ i-n+1 ,...,P′ i-1 ,P′ i}, which is used to denoise the next frame of point cloud and realize dynamic update of the sliding window.

[0050] A second aspect of the present invention discloses a real-time outlier noise removal system for array point cloud based on sliding completion, which includes five modules.

[0051] The first module is used to pre-process the initial input multi-frame array point cloud using a layered outlier noise removal method to generate an incomplete point cloud and store it in a sliding window; when the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established;

[0052] The second module is started after the sliding window is established. The incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method for each frame of the subsequent input point cloud. The removed points are regarded as the feature points of the current frame. The union of the point clouds in the sliding window is used as the point cloud to be filled. A completion strategy based on the array discrete probability model is designed to screen the feature points.

[0053] The third module is used to fill the filtered feature points into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, and complete the denoising of one frame of point cloud;

[0054] The fourth module is started after the point cloud denoising of the current frame is completed. The corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to realize the dynamic update of the sliding window.

[0055] In addition, the present invention also discloses an electronic device, which includes a processor and a memory storing computer program instructions. When the processor executes the computer program instructions, the method for removing real-time outlier noise from array point clouds based on sliding completion as disclosed in the first aspect and its preferred solution is implemented.

[0056] In addition, the present invention also discloses a computer-readable storage medium, in which at least one executable instruction is stored. When the executable instruction is executed on an electronic device, the electronic device executes the real-time outlier noise removal method for array point cloud based on sliding completion as disclosed in the first aspect and its preferred embodiment.

[0057] Beneficial effect: The present invention proposes an outlier noise removal technology for SPAD array lidar sparse point cloud, which can achieve high-precision denoising of the array sparse point cloud while ensuring real-time performance, overcoming the problems of weak adaptability and low denoising accuracy of the prior art when processing outlier noise in array point cloud. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Flow chart of a method for real-time outlier noise removal of array point cloud based on sliding completion in an embodiment.

[0059] Figure 2 Schematic diagram of the measured point cloud of SPAD array lidar.

[0060] Figure 3 This is a schematic diagram of the visualization results of the denoising of the measured tank point cloud using six algorithms.

[0061] Figure 4 Schematic diagram of the structure of a system for removing real-time outlier noise from an array point cloud based on sliding completion in an embodiment.

[0062] Figure 5 Schematic diagram of the structure of the electronic device in the embodiment. DETAILED DESCRIPTION

[0063] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described.

[0064] Embodiment 1:

[0065] Figure 1 What is shown is a flow chart of the method for real-time outlier noise removal of array point cloud based on sliding completion in this embodiment:

[0066] Please refer to Figure 1 , is a flowchart of a method and system for removing outlier noise from an array point cloud in real time based on sliding completion provided in Embodiment 1 of the present invention. The method includes steps 101 to 104, and each step is specifically as follows:

[0067] Step 101, for the initial input multi-frame array point cloud, a layered outlier noise removal method is used for preprocessing, an incomplete point cloud is generated and stored in a sliding window. When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established.

[0068] Specifically, let the initial input multi-frame point cloud be The capacity of the sliding window is n; for each frame of the input point cloud P i , using the layered outlier noise removal method for preprocessing to generate an incomplete point cloud P i ′, and stored in the sliding window. The steps of the layered outlier noise removal method are:

[0069] A bounding box is established for the input point cloud. The boundaries of the bounding box are the upper and lower bounds of the three-dimensional coordinates of the point cloud. The point cloud in the bounding box is layered. The calculation process of the single layer thickness is:

[0070] Among them, n1 is the total number of points in the input point cloud, S xoy , S yoz and S xoz The projection area of ​​the bounding box on the xoy, yoz and xoz planes is calculated as follows:

[0071]

[0072] Among them, x max is the upper limit of the X-axis coordinate of the input point cloud, x min is the lower bound of the X-axis coordinate of the input point cloud, y max is the upper bound of the Y-axis coordinate of the input point cloud, ymin is the lower bound of the Y-axis coordinate of the input point cloud, z max is the upper bound of the Z-axis coordinate of the input point cloud, z min is the lower bound of the Z-axis coordinate of the input point cloud. The input point cloud is layered along the Z axis according to the single layer thickness, and the total number of layers is:

[0073] l=|z max -z min | / δ

[0074] For all points in each layer of the input point cloud, the influence of the Z axis is ignored and a bounding box is established point by point. If there is only the point itself in the bounding box, the point is considered as an outlier and removed. The side length of the established bounding box is:

[0075] Among them, α is the bounding box threshold.

[0076] After performing outlier noise judgment operations on the input point cloud layer by layer and point by point, the point cloud preprocessing operation is completed to generate an incomplete point cloud. When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established.

[0077] Step 102, after the sliding window is established, for each frame of point cloud input subsequently, an incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method, and the removed points are regarded as feature points of the current frame. The union of the point clouds in the sliding window is used as the point cloud to be filled, and a completion strategy based on the array discrete probability model is designed to screen the feature points.

[0078] Specifically, let the subsequent input multi-frame point cloud be The union of the point clouds in the sliding window is P′. i The incomplete point cloud generated after preprocessing using the layered outlier noise removal method is P i ′, feature point set F = P i -P i ′. For any point f in the feature point set F i , get its neighbor point set in P′ Then the neighbor point p j The relative average photoelectron number is defined as:

[0079] Where E is the energy of a single laser pulse, λ is the laser wavelength, h is Planck's constant, c is the speed of light, ρ is the reflectivity, and k is R is the lens loss, k M is the atmospheric transmission loss, k L is the laser emission loss, A is the receiving objective lens area, d ij f i With p j, γ is the detector filling factor, and η is the detector quantum efficiency.

[0080] Neighbor point p j The corresponding trigger interval is:

[0081] Where c is the speed of light, T a The length of a single time interval.

[0082] If the laser echo photon is j The probability of the event occurring is:

[0083] in, is the average noise photoelectron number, which can be calculated as:

[0084] Among them, I sun is the solar radiation intensity after bandpass filtering, k′ M is the atmospheric transmission loss between the sun and the target, It represents the diffuse reflection angle of the target surface to the incident sunlight, and θ is the field of view angle of the detector.

[0085] In the point set After calculating the trigger probability of all points, the mean function is used to convert the trigger probability to point f i Noise fraction:

[0086]

[0087] If point f i If the noise score of is less than the set threshold ψ, the point is considered as a noise point and removed. After noise point detection and removal are completed for all feature points in the feature point set F, the feature point screening step is completed.

[0088] Step 103, fill the filtered feature points into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, and complete the denoising of one frame of point cloud.

[0089] Specifically, let the filtered feature point set be F′, then the point cloud after denoising of the current frame is P i ″=F′∪P i ′.

[0090] Step 104, after the point cloud denoising of the current frame is completed, its corresponding incomplete point cloud will be added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue will be removed to achieve dynamic update of the sliding window.

[0091] Specifically, let the incomplete point cloud of the current frame be P i ′, then the sliding window is {P′ i-n ,P′ i-n+1,…,P′ i-1}. i ″Add to the end of the sliding window queue, and get {P′ i-n ,P′ i-n+1 ,...,P′ i-1 ,P′ i}. Remove the point cloud at the head of the sliding window team and get {P′ i-n+1 ,...,P′ i-1 ,P′ i}, used to denoise the next frame of point cloud. This method can realize the dynamic update of the sliding window.

[0092] At this point, the process of the method for removing real-time outlier noise from an array point cloud based on sliding completion disclosed in an embodiment of the present invention ends.

[0093] The effect of the technical solution of the present invention is further verified and explained below in conjunction with specific experiments.

[0094] In order to verify the accuracy of the method of the present invention, a test experiment with a public data set is performed to prove it. The simulation results of the present invention are shown as follows: Figure 2 and Figure 3 shown. Figure 2 The visualization results of the SPAD array lidar measured point cloud are shown. Figure 3 The figure shows the superposition of 10 consecutive frames of point cloud processing results of SPAD array lidar using six outlier noise processing methods. Figure 3 (a), (b), (c), (d) and (e) are the processing results of the SOR algorithm, the Local density algorithm, the Noniterative algorithm, the MCMD algorithm and the CIMD algorithm respectively. It can be seen that there are many non-isolated noise points remaining in the result of the SOR algorithm, and there is a problem of feature loss in the point cloud of the tank body. The denoising results of the Local density algorithm, the Noniterative algorithm and the MCMD algorithm all retain isolated noise points to varying degrees. The background point cloud features are seriously lost in the results of the Local density algorithm and the MCMD algorithm. A small amount of non-isolated noise points remain in the denoising result of the CIMD algorithm, but the features are relatively intact. Figure 3 (f) is the processing result of the method of the present invention. It can be seen that the denoising result contains almost no noise points and the features are most completely retained.

[0095] In summary, the simulation experiments have verified the correctness, effectiveness and reliability of the present invention.

[0096] Embodiment 2:

[0097] Please refer to Figure 4, another embodiment of the present invention provides a real-time outlier noise removal system for array point clouds based on sliding completion, which consists of four modules. The first module is used to pre-process the multi-frame array point clouds initially input by using a layered outlier noise removal method to generate an incomplete point cloud and store it in a sliding window; when the number of incomplete point clouds generated is equal to the capacity of the sliding window, the initial sliding window is established. The second module is started after the sliding window is established, and the incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method for each frame of point cloud subsequently input. The removed points are regarded as feature points of the current frame, and the union of the point clouds in the sliding window is used as the point cloud to be filled, and a completion strategy based on the array discrete probability model is designed to filter the feature points. The third module is used to fill the filtered feature points into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, and complete the denoising of a frame of point cloud. The fourth module is started after the denoising of the point cloud of the current frame is completed, and the corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to realize the dynamic update of the sliding window.

[0098] In the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0099] In practical applications, the operation logic of the array point cloud real-time outlier noise removal method based on sliding completion disclosed in the above embodiment can be written into a set of executable instructions, which are written into a storage medium and run on an electronic device. Figure 5 . More specific examples of the computer-readable storage medium mentioned in this embodiment may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or device.

[0100] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0101] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for removing outlier noise from array point cloud in real time based on sliding completion, characterized in that: The steps include: For the initial input multi-frame array point cloud, a layered outlier noise removal method is used for preprocessing to generate an incomplete point cloud and store it in a sliding window. When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established. After the sliding window is established, for each frame of point cloud input subsequently, the incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method. The removed points are regarded as the feature points of the current frame. The union of the point clouds in the sliding window is used as the point cloud to be filled, and a completion strategy based on the array discrete probability model is designed to screen the feature points. Fill the filtered feature points into the incomplete point cloud of the current frame, which is the denoised point cloud of the current frame, completing the denoising of one frame of point cloud; After the current frame point cloud denoising is completed, its corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to achieve dynamic update of the sliding window.

2. The method for removing outlier noise from array point cloud in real time based on sliding completion according to claim 1, characterized in that: The initial input multi-frame array point cloud is preprocessed using a layered outlier noise removal method, specifically including: A bounding box is established for the input point cloud. The boundaries of the bounding box are the upper and lower bounds of the point cloud's three-dimensional coordinates. The point cloud in the bounding box is layered. The calculation formula for the thickness of a single layer is as follows: Where n1 is the total number of points in the input point cloud, S xoy , S yoz , S xoz They are the projection areas of the bounding boxes on the xoy, yoz, and xoz planes, respectively, and the calculation formula is as follows: In the formula, x max is the upper limit of the X-axis coordinate of the input point cloud, x min is the lower bound of the X-axis coordinate of the input point cloud, y max is the upper bound of the Y-axis coordinate of the input point cloud, y min is the lower bound of the Y-axis coordinate of the input point cloud, z max is the upper bound of the Z-axis coordinate of the input point cloud, z min It is the lower bound of the Z-axis coordinate of the input point cloud; The input point cloud is layered along the Z axis according to the single layer thickness, and the total number of layers is: l=|z max -With min | / δ For all points in each layer of the input point cloud, the influence of the Z axis is ignored and a bounding box is established point by point. If there is only the point itself in the bounding box, the point is regarded as an outlier and removed. The side length of the established bounding box is: Where α is the bounding box threshold; After performing outlier noise judgment operations on the input point cloud layer by layer and point by point, the point cloud preprocessing operation is completed.

3. The method for removing real-time outlier noise from array point cloud based on sliding completion according to claim 2, characterized in that: When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established, which specifically includes: Assume that the initial input multi-frame point cloud is The capacity of the sliding window is n; for each frame of the input point cloud P i , use the layered outlier noise removal method to preprocess P′ and generate an incomplete point cloud P i ′, and stored in the sliding window; When the number of incomplete point clouds generated is equal to the capacity of the sliding window, the initial sliding window is established.

4. The method for removing real-time outlier noise from array point cloud based on sliding completion according to claim 2, characterized in that: For each frame of point cloud subsequently input, an incomplete point cloud of the current frame is obtained based on a layered outlier noise removal method, the removed points are regarded as feature points of the current frame, and the union of the point clouds in the sliding window is used as the point cloud to be filled, specifically including: Assume that the subsequent input multi-frame point cloud is The union of the point clouds in the sliding window is P′; for the current frame point cloud P i The incomplete point cloud generated after preprocessing using the layered outlier noise removal method is P i ′, feature point set F = P i -P i ′; For any point f in the feature point set F i , get its neighbor point set in P′ Neighbor point p j The relative average photoelectron number is defined as: Where E is the energy of a single laser pulse, λ is the laser wavelength, h is Planck's constant, c is the speed of light, ρ is the reflectivity, and k is R is the lens loss, k M is the atmospheric transmission loss, k L is the laser emission loss, A is the receiving objective lens area, d ij f i With p j The distance between the detector and the detector is γ, γ is the detector filling factor, and η is the detector quantum efficiency; Neighbor point p j The corresponding trigger interval is: Where c is the speed of light, T a is the length of a single time interval.

5. The method for removing real-time outlier noise from array point cloud based on sliding completion according to claim 4, characterized in that: The design screens feature points based on a completion strategy of an array discrete probability model, specifically including: If the laser echo photon is j The probability of the event occurring is: In the formula, is the average noise photoelectron number, which is calculated as follows: In the formula, I sun is the solar radiation intensity after bandpass filtering, k′ M is the atmospheric transmission loss between the sun and the target, It represents the diffuse reflection angle of the target surface to the incident sunlight, and θ is the field of view angle of the detector; In the point set After calculating the trigger probability of all points, the mean function is used to convert the trigger probability to point f i The noise fraction S i : If point f i If the noise score of is less than the set threshold ψ, the point is considered as a noise point and removed; After noise point detection and removal are completed for all feature points in the feature point set F, the feature point screening step is completed.

6. The method for removing real-time outlier noise from array point cloud based on sliding completion according to claim 4, characterized in that: The filtered feature points are filled into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, and the denoising of a frame of point cloud is completed, specifically: Assume that the filtered feature point set is F′, then the point cloud after denoising of the current frame is P i ″=F′∪P i ′.

7. The method for removing real-time outlier noise from array point cloud based on sliding completion according to claim 4, characterized in that: After the current frame point cloud denoising is completed, the corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to achieve dynamic update of the sliding window, specifically including: Let the incomplete point cloud of the current frame be P i ′, then the sliding window is {P′ i-n ,P′ i-n+1 ,…,P′ i-1 }; The denoised point cloud P i ″Add to the end of the sliding window queue, and get {P′ i-n ,P′ i-n+1 ,...,P′ i-1 ,P i ′}; remove the point cloud at the head of the sliding window team, and get {P′ i-n+1 ,...,P′ i-1 ,P i ′}, which is used to denoise the next frame of point cloud and realize the dynamic update of the sliding window.

8. A real-time outlier noise removal system for array point cloud based on sliding completion, characterized in that: include: Module 1; The first module is used to pre-process the initially input multi-frame array point cloud using a layered outlier noise removal method to generate an incomplete point cloud and store it in a sliding window; When the number of generated incomplete point clouds is equal to the capacity of the sliding window, the initial sliding window is established; The second module is started after the sliding window is established. The incomplete point cloud of the current frame is obtained based on the layered outlier noise removal method for each frame of point cloud input subsequently. The removed points are regarded as feature points of the current frame. The union of the point clouds in the sliding window is used as the point cloud to be filled. A completion strategy based on the array discrete probability model is designed to screen the feature points. Module 3; The third module is used to fill the filtered feature points into the incomplete point cloud of the current frame, that is, the denoised point cloud of the current frame, and complete the denoising of one frame of point cloud; The fourth module: The fourth module is started after the point cloud denoising of the current frame is completed, and the corresponding incomplete point cloud is added to the end of the sliding window queue, and the point cloud at the head of the sliding window queue is removed to realize the dynamic update of the sliding window.

9. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for removing real-time outlier noise from array point cloud based on sliding completion as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one executable instruction. When the executable instruction is executed on the electronic device, the electronic device executes the array point cloud real-time outlier noise removal method based on sliding completion as described in any one of claims 1 to 7.

Images