Dynamic water mist noise filtering method, device, unmanned vehicle and storage medium

Abstract

The present invention discloses a method, device, unmanned vehicle, and storage medium for filtering dynamic water mist noise. The method comprises: constructing a correlation model between water vapor splash distance and vehicle speed; obtaining the vehicle speed of the current frame and point cloud data from the BEV perspective, and delineating a splash domain in the point cloud data; obtaining a coarse noise domain based on the vehicle speed and the correlation model, and obtaining a coarse noise point set contained in the coarse noise domain; extracting noise points between the coarse noise domain boundary and the splash domain boundary based on Euclidean clustering to obtain a clustered noise point set; obtaining a fine noise point set based on the coarse noise point set and the clustered noise point set, and performing distance screening on the fine noise point set based on the center point set of surrounding intelligent bodies to obtain a final filtered noise point set and filter it from the point cloud data. The present invention is applied to the field of point cloud data processing, and can not only perform coarse and fine noise screening on dynamic vehicles from the BEV perspective, but also reasonably filter water mist thrown out by dynamic vehicles from the BEV perspective.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing, and in particular to a method and device for filtering dynamic water mist noise, an unmanned vehicle, and a storage medium. Background Art

[0002] As an emerging 3D sensor, lidar can obtain object information from multiple directions, thereby achieving better perception results. Although BEV (Bird's Eye View), as a radar's bird's-eye view, reduces the dimensional information in the Z-axis direction, the current development trend is multi-sensor fusion. Under the BEV perspective, feature information from different perspectives, different time series (frames), and different modalities (sensors) can be integrated to output more accurate BEV spatial features. Based on the obtained BEV features, different perception tasks such as 3D target detection, lane detection, and map semantic segmentation can be implemented. The perception results of different tasks can be naturally applied to autonomous driving planning and control.

[0003] Currently, there are many noise filtering solutions for LiDAR on the market, such as:

[0004] Patent publication number CN114581331A proposes using voxelization to segment the point cloud and then using a smoothing algorithm to correct outliers. However, this solution only works for 3D point clouds and does not significantly improve the BEV viewing angle. It also cannot effectively smooth large, dynamically changing areas of water mist.

[0005] Patent publication number CN108876744A proposes using a region segmentation approach to perform BEV calibration on LiDAR and binocular cameras. After acquiring a large amount of information, point clouds are stitched together and normalized. Compared to the three aforementioned approaches, this solution is better suited for complex environments such as point cloud noise reduction. However, it requires multi-sensor fusion, including LiDAR and binocular cameras, requires more equipment, has high calibration requirements, and has a complex algorithm, making it difficult to deploy.

[0006] In summary, the current noise filtering technology has the following shortcomings: there are few noise filtering solutions for BEVs, many devices use multi-sensor fusion methods, and the algorithm is time-consuming, which makes it unsuitable for vehicle-side application. There is no good solution for dynamically changing large-area water mist. Summary of the Invention

[0007] In response to the above-mentioned deficiencies in the prior art, the present invention provides a dynamic water mist noise filtering method, device, unmanned vehicle and storage medium, which can perform coarse and fine noise screening on dynamic vehicles from the perspective of BEV, and realize reasonable filtering of water mist thrown out by dynamic vehicles from the perspective of BEV.

[0008] To achieve the above object, the present invention provides a method for filtering dynamic water mist noise, comprising the following steps:

[0009] Step 1: Construct a correlation model between water vapor splash distance and vehicle speed;

[0010] Step 2: obtaining the vehicle speed and point cloud data from the BEV perspective collected in the current frame, and delineating a splash area in the point cloud data based on preset data;

[0011] Step 3: obtaining a coarse noise point domain in the current frame point cloud data based on the vehicle speed and the association model, and obtaining a coarse noise point set contained in the coarse noise point domain, wherein the coarse noise point domain is located in the splash domain;

[0012] Step 4: extracting noise points between the boundary of the coarse noise point domain and the boundary of the splash domain based on Euclidean clustering to obtain a clustered noise point set;

[0013] Step 5: obtain a fine noise point set based on the coarse noise point set and the clustered noise point set, and perform distance screening on the fine noise point set based on the center point set of the surrounding intelligent bodies to obtain the final filtered noise point set and filter it out from the point cloud data.

[0014] In one embodiment, in step 1, the association model is:

[0015] X max =P xmax ·(k xmax ·V BW +b xmax )+C x

[0016] X min =P xmin ·(k xmin ·V BW +b xmin )-C x

[0017] Y max =P ymax ·(k ymax ·V BW +b ymax )+C y

[0018] Y min =P ymin (k ymin ·V BW +b ymin )-C y

[0019] Among them, X max 、X minThey are the maximum and minimum values ​​of the coarse noise domain in the X-axis direction, and the Y max 、Y min are the maximum and minimum values ​​of the coarse noise domain in the Y-axis direction, P xmax 、P xmin 、P ymax 、P ymin X max 、X min 、Y max 、Y min The predicted probability, k xmax 、k xmin 、k ymax 、k ymin X max 、X min 、Y max 、Y min The slope of the fitting function, b xmax 、b xmin 、b ymax 、b ymin X max 、X min 、Y max 、Y min The intercept of the fitting function, V BW is the vehicle speed, C x is the floating constant in the X-axis direction, C y is a floating constant in the Y-axis direction, and the Y-axis direction is the vehicle speed V BW direction.

[0020] In one embodiment, the floating constant C x for:

[0021] C x =|α1·d 风向 +β1·d 车重 |

[0022] Among them, d 风向 is the wind force floating constant, d 车重 is the vehicle weight floating constant, α1 and β1 are weight coefficients, and α1>β1.

[0023] The floating constant C y for:

[0024] C y =|β2·d 风向 +α2·d 车重 |

[0025] Among them, α2 and β2 are weight coefficients, and α2>β2.

[0026] In one embodiment, in the preset data, the maximum and minimum values ​​of the splash domain in the X-axis direction are X max0 、X min0 The maximum and minimum values ​​of the splash domain in the Y-axis direction are Y max0 、Y min0 ;

[0027] Among them, X max0 ≥X max , Y max0 ≥Y max , X min0 ≤X min , Y min0 ≤Y min0 .

[0028] In one embodiment, in step 5, the set of fine noise points is:

[0029] N 细 =N 粗 +N 欧

[0030] Among them, N 细 is a set of fine noise points, N 粗 is a coarse noise point set, N 欧 is the clustered noise set;

[0031] The filtered noise point set is:

[0032] N 滤 =N 细 -N agent

[0033] Among them, N 滤 To filter out the noise point set, N agent is the center point set of the surrounding intelligent agents.

[0034] In one embodiment, in step 2, when the vehicle speed V of the current frame is acquired, BW , judge the vehicle speed V BW Is it greater than the first vehicle speed threshold?

[0035] If yes, proceed to step 3 to step 5;

[0036] Otherwise, after obtaining the vehicle speed and point cloud data from the BEV perspective collected in the next frame, proceed to step 2 again.

[0037] In one embodiment, in step 2, when the vehicle speed V of the current frame is acquired, BW , judge the vehicle speed V BW Is it greater than the second vehicle speed threshold?

[0038] If so, in step 3, the second vehicle speed threshold is brought into the association model to obtain a coarse noise point domain in the current frame point cloud data;

[0039] Otherwise, in step 3, the vehicle speed V BW The association model is brought in to obtain the coarse noise point domain in the current frame point cloud data.

[0040] To achieve the above object, the present invention further provides a dynamic water mist noise filtering device, which uses the above method to filter out noise. The dynamic water mist noise filtering device includes:

[0041] A model building module is used to build a correlation model between water vapor splash distance and vehicle speed;

[0042] a data acquisition module, configured to acquire preset data and the vehicle speed and point cloud data from the BEV perspective acquired in the current frame, and to delineate a splash domain in the point cloud data based on the preset data;

[0043] a coarse noise point extraction module, configured to obtain a coarse noise point domain in the point cloud data of the current frame according to the vehicle speed and the correlation model, and obtain a coarse noise point set contained in the coarse noise point domain, wherein the coarse noise point domain is located in the splash domain;

[0044] A Euclidean clustering module, configured to extract noise points between the boundary of the coarse noise point domain and the boundary of the splash domain according to Euclidean clustering to obtain a clustered noise point set;

[0045] The noise point filtering module is used to obtain a fine noise point set based on the coarse noise point set and the clustered noise point set, and to perform distance screening on the fine noise point set based on the center point set of the surrounding intelligent bodies to obtain a final filtered noise point set and filter it out from the point cloud data.

[0046] To achieve the above object, the present invention further provides an unmanned vehicle, wherein the unmanned vehicle is provided with:

[0047] Memory, used to store programs;

[0048] A processor is used to execute the program stored in the memory. When the program is executed, the processor is used to perform part or all of the steps of the above method.

[0049] To achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions; when the computer-executable instructions are executed by a processor, they are used to implement some or all of the steps of the above method.

[0050] Compared with the prior art, the present invention has the following beneficial technical effects:

[0051] 1. This invention uses a single-sensor LiDAR for imaging to filter out dynamic water mist noise. Compared to traditional multi-sensor fusion methods, it eliminates the time-consuming problem of multi-sensor matching.

[0052] 2. Based on a pre-built correlation model between water vapor splash distance and vehicle speed, this method can quickly delineate the coarse noise domain in point cloud data from the perspective of a BEV, significantly reducing noise filtering time.

[0053] 3. By constructing a correlation model between the water vapor splash distance and the vehicle speed, the present invention method becomes a dynamic solution applicable to different speed measurements, can be applied to multiple complex environments and scenarios, and has high filtering accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0055] Figure 1 Schematic diagram of the Cartesian 2D self-vehicle coordinate system in Example 1 of the present invention;

[0056] Figure 2 This is a flow chart of the method for filtering dynamic water mist noise in Example 1 of the present invention;

[0057] Figure 3 is the boundary X of the coarse noise region R in embodiment 1 of the present invention max Scatter plot of

[0058] Figure 4 is the boundary X of the coarse noise region R in embodiment 1 of the present invention min Scatter plot of

[0059] Figure 5 The boundary Y of the coarse noise region R in embodiment 1 of the present invention is max Scatter plot of

[0060] Figure 6 The boundary Y of the coarse noise region R in embodiment 1 of the present invention is min Scatter plot of

[0061] Figure 7 In the first embodiment of the present invention, a cubic function is used to fit the boundary X of the coarse noise domain R. max Schematic diagram of the results;

[0062] Figure 8In the first embodiment of the present invention, a linear function is used to fit the boundary X of the coarse noise domain R. max Schematic diagram of the results;

[0063] Figure 9 Schematic diagram of point cloud data collected in real time in Example 1 of the present invention;

[0064] Figure 10 Schematic diagram of point cloud data after noise is filtered out in Example 1 of the present invention;

[0065] Figure 11 This is a structural block diagram of a dynamic water mist noise filtering device in Example 2 of the present invention;

[0066] Figure 12 This is a structural block diagram of an unmanned vehicle in Example 3 of the present invention.

[0067] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0069] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0070] In addition, the terms "first," "second," and so on, used in this disclosure are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referenced. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this disclosure, "plurality" means at least two, such as two or three, unless otherwise specifically defined.

[0071] In the present invention, unless otherwise specified or limited, the terms "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can mean fixed connection, detachable connection, or integration; it can mean mechanical connection, electrical connection, physical connection, or wireless communication connection; it can mean direct connection or indirect connection through an intermediate medium; it can mean internal communication between two elements or interaction between two elements, unless otherwise specified. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0072] 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 this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0073] Example 1

[0074] When a vehicle travels on a flooded road and exceeds a certain speed, the front and rear wheels of the vehicle will inevitably produce splashes of water mist / water droplets, which in turn creates a large amount of dynamic water mist noise at the rear and sides of the vehicle in the point cloud data from the LiDAR BEV perspective. This causes the LiDAR to be unable to accurately output vehicle information or information about real objects at the rear and sides of the vehicle. Based on this, this embodiment discloses a dynamic water mist noise filtering method, which is mainly suitable for noise reduction from the BEV perspective of a LiDAR with accumulated water on the ground. This method uses a single 16-line LiDAR sensor to perform coarse and fine noise screening on dynamic vehicles from the BEV perspective, achieving reasonable filtering of water mist thrown out by dynamic vehicles from the BEV perspective, and thus accurately outputting vehicle information and information about real objects around the vehicle.

[0075] The BEV perspective is a bird's-eye view of the LiDAR centered on the Cartesian 2D vehicle coordinate system. Figure 1 The Cartesian 2D ego-vehicle coordinate system is centered on the ego-vehicle's drive wheels and includes X- and Y-axis directions. The Y-axis direction represents the vehicle's speed, and the ego-vehicle's drive wheels are the wheels connected to the power shaft driven by the vehicle's engine connecting rod. The BEV perspective is the visible range within the Cartesian 2D ego-vehicle coordinate system, with the Y-axis ranging from +Y0 meters to -Y0 meters and the X-axis ranging from +X0 to -X0 meters. For example, the BEV perspective can be set to a visible range within the Cartesian 2D ego-vehicle coordinate system, with the Y-axis ranging from 5 meters to -5 meters and the X-axis ranging from 10 to -10 meters.

[0076] refer to Figure 2 The dynamic water mist noise filtering method in this embodiment specifically includes the following steps:

[0077] Step 1: Construct a correlation model between water vapor splash distance and vehicle speed;

[0078] Step 2: Obtain the vehicle speed and point cloud data from the BEV perspective collected in the current frame, and delineate the splash area in the point cloud data based on preset data;

[0079] Step 3: obtaining a coarse noise point domain in the current frame point cloud data based on the vehicle speed and the association model, and obtaining a coarse noise point set contained in the coarse noise point domain, wherein the coarse noise point domain is located in the splash domain;

[0080] Step 4: extract the noise points between the boundary of the coarse noise domain and the boundary of the splash domain based on Euclidean clustering to obtain a clustered noise point set;

[0081] In step 5, a fine noise point set is obtained based on the coarse noise point set and the clustered noise point set, and the fine noise point set is distance-filtered based on the center point set of the surrounding intelligent bodies to obtain the final filtered noise point set and filter it out from the point cloud data.

[0082] In this embodiment, the splash region R0 is set so that the point cloud splash noise generated by the vehicle at any speed can be covered by the splash region R0. The splash region R0 is defined as (X max0 , X min0 , Y max0 , Y min0 ), where X max0 、X min0 is the maximum and minimum value of the splash domain R0 in the X direction, Y max0 、Y min0 The maximum and minimum values ​​of the splash area R0 in the Y direction. That is, the preset data in step 2 above contains the parameter X max0 、X min0 、Y max0 、Y min0 .

[0083] The coarse noise domain in step 3 above refers to the range of a part of the point cloud splash noise that needs to be filtered out in the current frame point cloud data. The coarse noise domain R is defined as (X max , X min , Y max , Y min ), where X max 、X min is the maximum and minimum value of the coarse noise domain R in the X direction, Y max 、Y min The maximum and minimum values ​​of the coarse noise domain R in the Y direction. The maximum and minimum values ​​of the coarse noise domain R in the X and Y axes are both point cloud splash noise, so the parameter relationship between the splash domain R0 and the coarse noise domain R should satisfy X max0 ≥X max , Y max0 ≥Ymax , X min0 ≤X min , Y min0 ≤Y min .

[0084] Because the speed of a vehicle changes dynamically during driving, the range of the coarse noise domain R in each frame of point cloud data also changes dynamically. Based on this, this embodiment builds a correlation model between water vapor splash distance and vehicle speed. This updates the coarse noise domain R in each frame of point cloud data based on the vehicle's real-time speed, effectively filtering out water mist thrown out by a dynamic vehicle.

[0085] As the vehicle speed increases, the distance of water vapor splashing increases, so there is a linear relationship between the distance of water vapor splashing and vehicle speed. Therefore, the four boundaries X of the coarse noise domain R can be obtained by fitting the historical point cloud data at several different speeds. max 、X min With Y max 、Y min The specific implementation process is as follows:

[0086] First, collect N frames of point cloud data at different speeds;

[0087] Secondly, for the i-th frame point cloud data, extract the distance X max0 Nearest point X maxi , distance X min0 Nearest point X mini , distance Y max0 The nearest point Y maxi and distance Y min0 The nearest point Y mini , where i = 1 to N;

[0088] Then, take vehicle speed as the independent variable and X maxi As the dependent variable, we can get the coarse noise domain R boundary X max The scatter plot is obtained by using Matlab to calculate the boundary X of the coarse noise domain R max The scatter plot is fitted with a linear curve to obtain the coarse noise domain R boundary X max The correlation model with speed, i.e. Figure 3 As shown;

[0089] Similarly, take vehicle speed as the independent variable and X mini As the dependent variable, we can get the coarse noise domain R boundary X min The scatter plot is obtained by using Matlab to calculate the boundary X of the coarse noise domain R min The scatter plot is fitted with a linear curve to obtain the coarse noise domain R boundary X min The correlation model with speed, i.e. Figure 4 As shown;

[0090] Similarly, take vehicle speed as the independent variable and Y maxi As the dependent variable, the boundary Y of the coarse noise domain R can be obtained max The scatter plot is obtained by using Matlab to calculate the boundary Y of the coarse noise domain R max The scatter plot is fitted with a linear curve to obtain the coarse noise domain R boundary Y max The correlation model with speed, i.e. Figure 5 As shown;

[0091] Similarly, take vehicle speed as the independent variable and Y min As the dependent variable, the boundary Y of the coarse noise domain R can be obtained min The scatter plot is obtained by using Matlab to calculate the boundary Y of the coarse noise domain R min The scatter plot is fitted with a linear curve to obtain the coarse noise domain R boundary Y min The correlation model with speed, i.e. Figure 6 shown.

[0092] In the specific implementation process, the traditional linear function can be used to fit the correlation model between the boundary of the coarse noise domain R and the velocity, that is, the correlation model is:

[0093] X max =P xmax ·(k xmax ·V BW +b xmax )+C x

[0094] X min =P xmin ·(k xmin ·V BW +b xmin )-C x

[0095] Y max =P ymax ·(k ymax ·V BW +b ymax )+C y

[0096] Y min =P ymin ·(k ymin ·V BW +b ymin )-C y

[0097] Among them, P xmax 、P xmin 、P ymax 、P ymin Xmax 、X min 、Y max 、Y min The predicted probability, k xmax 、k xmin 、k ymax 、k ymin X max 、X min 、Y max 、Y min The slope of the fitting function, b xmax 、b xmin 、b ymax 、b ymin X max 、X min 、Y max 、Y min The intercept of the fitting function, C x is the floating constant in the X-axis direction, C y A floating constant in the Y-axis direction.

[0098] In the specific implementation process, the prediction probability P xmax With P ymax Since it is used to fit the maximum value boundary, it is preferred to set P xmax With P ymax Set to greater than 1, for example, xmax With P ymax Set to 1.05. And the predicted probability P xmin With P ymin is the boundary for fitting the minimum value, so it is preferred to set P xmax With P ymax Set to less than 1, for example, P xmax With P ymax Set to 0.95.

[0099] It is worth noting that in specific applications, there may be a problem that the linear function or the multi-level function cannot completely fit a certain boundary of the coarse noise domain R. In this case, multiple segmented linear functions can be used to fit the boundary together. Figure 7 、 Figure 8 The example shown in this example is the boundary X of the coarse noise domain R max When fitting, Figure 7 is the result of fitting with a cubic function, Figure 8 is the result of fitting with a linear function. Figure 7 、 Figure 8 , using cubic function or linear function alone cannot produce good fitting results. Since the maximum and minimum values ​​are segmented, the segmented function is used for fitting. First, analyze the segmentation, such as Figure 8As shown, since the maximum value needs to be included, the maximum value is used to find the segment boundary, and the speed corresponding to the segment boundary is recorded as V BW1 , then the boundary X of the coarse noise region R in this example max The correlation model with speed is:

[0100] X max =P xmax ·(k xmax-1 ·V BW +b xmax-1 )+C x (V BW ≥V BW1 )

[0101] X max =P xmax ·(k xmax-2 ·V BW +b xmax-2 )+C x (V BW <V BW1 )

[0102] Among them, k xmax-1 with b xmax-1 At speed V BW ≥V BW1 The boundary X of the coarse noise domain R max The slope and intercept of the first segment of the fitting function, k xmax-2 with b xmax-2 At speed V BW <V BW1 The boundary X of the coarse noise domain R max The slope and intercept of the second segment of the fitted function.

[0103] In the real-time calculation process, the vehicle speed V measured in real time can be used to calculate the vehicle speed. BW , k xmax-1 or k xmax-2 Bringing k into the above association model xmax , and b xmax-1 or b xmax-2 Bringing k into the above association model xmax Similarly, the same approach can be used when the same problem occurs at other boundaries of the coarse noise region R. It is worth noting that the use of piecewise functions for fitting in specific applications is not limited to having only one segment boundary; the specific number of segments can be determined based on the actual situation.

[0104] In this embodiment, the floating constant C in the X-axis direction x Adjusted by the current vehicle weight and wind force. Since the horizontal direction is greatly affected by the splash caused by the crosswind, the wind force floating constant d is set. 风向The weight coefficient is set larger, and the vehicle weight floating constant d 车重 The weight coefficient is set to be small, that is:

[0105] C x =|α1·d 风向 +β1·d 车重 |

[0106] Among them, α1 and β1 are weight coefficients, and α1>β1.

[0107] In this embodiment, the floating constant C in the Y-axis direction y The weight of the car in the current state is also adjusted by the wind force. Since the longitudinal direction is greatly affected by the splash caused by the car's own gravity, the car weight floating constant d is set. 车重 The weight coefficient is set larger, and the wind force floating constant d 风向 The weight coefficient is set to be small, that is:

[0108] C y =|β2·d 风向 +α2·d 车重 |

[0109] Among them, α2 and β2 are weight coefficients, and α2>β2.

[0110] In the specific implementation process, it can be seen that the above weight coefficients are set to α1 = α2 = 0.8, β1 = β2 = 0.2. 车重 and wind force constant d 风向 , which can be determined by the vehicle's own weight and the current direction and size of the wind vane. Taking the city as the unit, the wind force in the X-axis and Y-axis directions is calculated by velocity decomposition based on the direction and size of the wind vane in the current city and environment. The wind force is obtained according to Newton's second law as:

[0111]

[0112] The vehicle weight is based on the current experimental vehicle bus basic parameter of 800kg. After obtaining the wind force and vehicle weight, the vehicle weight floating constant d can be obtained by geometric conversion according to the calibration or preset ratio. 车重 and wind force constant d 风向 .

[0113] After determining the coarse noise domain R in the current frame point cloud data through the above association model, although a large area of ​​point cloud splash noise can be filtered out through the coarse noise domain R, due to the disorder of water mist noise, some point cloud splash noise may still be outside the coarse noise domain R, resulting in incomplete filtering. Based on this, this embodiment extracts the noise between the coarse noise domain boundary and the splash domain boundary based on Euclidean clustering in step 4 to obtain a clustered noise set. Among them, the constraint of Euclidean clustering is d 聚类 >d 物体 , where d 聚类 is the distance between point sets set in Euclidean clustering, d 物体 is the average distance between points in the obstacle and pedestrian point clusters. The specific implementation of the Euclidean clustering filter is a conventional technical means in this field, so it will not be described in detail in this embodiment.

[0114] After Euclidean clustering, we get the cluster noise point set N 欧 , we can combine the coarse noise point set N in the coarse noise point domain R 粗 Get the fine noise point set N 细 =N 粗 +N 欧 Considering that there may be other intelligent entities such as vehicles and pedestrians around the vehicle during its driving process, in order to accurately output vehicle information and information about real objects around the vehicle, it is also necessary to 细 The central point set N of the surrounding intelligent agents agent That is, in the historical frame point cloud data, the position of each object in the current frame point cloud data can be accurately obtained according to the 3D target. First, the center points of the surrounding intelligent bodies within a certain range from the vehicle (for example, within 50 cm from the vehicle) are filtered out, and the point set within a certain radius (for example, within 10 cm) of the center points of the filtered surrounding intelligent bodies is used as the center point set N of the surrounding intelligent bodies. agent , then the final filtered noise point set N 滤 =N 细 -N agent .

[0115] refer to Figure 9 Schematic diagram of point cloud data collected in real time. Figure 10 This is a schematic diagram of point cloud data after noise is filtered out by the filtering method of this embodiment. Figure 9 、 Figure 10 It can be seen that the filtering method in this embodiment can perform coarse and fine noise screening on dynamic vehicles from the perspective of BEV, and reasonably filter the water mist thrown out by dynamic vehicles from the perspective of BEV, thereby accurately outputting vehicle information and information about real objects around the vehicle.

[0116] As a preferred embodiment, since there is basically no splash noise when the vehicle is driving at low speed on a road with water accumulation, and considering the low-speed safety issues of pedestrians in the surrounding area, the filtering method in this embodiment is implemented in the specific process. When it reaches step 2, the vehicle speed V of the current frame is obtained. BW Then, the vehicle speed V is determined in real time. BW Is it greater than the first vehicle speed threshold (for example, 5 km / h): If so, perform the noise filtering operations in subsequent steps 3 to 5; otherwise, no noise filtering operations in subsequent steps 3 to 5 are performed in the current frame, and the vehicle speed collected in the next frame and the point cloud data from the BEV perspective are obtained before processing the next frame of data.

[0117] As a preferred embodiment, when the speed of a vehicle on a flooded road exceeds a certain value, the splash noise generated will no longer change significantly. At the same time, in order to avoid the problem of excessively large calculation range of the coarse noise domain R due to excessive speed, the filtering method in this embodiment is implemented in the specific process. When it reaches step 2, after obtaining the vehicle speed V of the current frame, BW After that, in addition to judging the vehicle speed V BW In addition to determining whether the vehicle speed V is greater than a first vehicle speed threshold (e.g., 5 km / h), it is also necessary to determine whether the vehicle speed V BW Is it greater than the second vehicle speed threshold (e.g. 17 km / h): When the vehicle speed V BW When the vehicle speed V is greater than the second vehicle speed threshold, the second vehicle speed threshold is brought into the correlation model in the subsequent step 3 to obtain the coarse noise point domain in the current frame point cloud data; BW When the vehicle speed V is less than the second vehicle speed threshold and greater than the first vehicle speed threshold, the vehicle speed V is set to BW The association model is brought in to obtain the coarse noise domain in the current frame point cloud data.

[0118] Example 2

[0119] Based on the dynamic water mist noise filtering method in Example 1, this embodiment discloses a dynamic water mist noise filtering device. Figure 11 The dynamic water mist noise filtering device includes a model building module, a data acquisition module, a coarse noise extraction module, a Euclidean clustering module, and a noise filtering module. The dynamic water mist noise filtering device is used to perform some or all of the steps of the dynamic water mist noise filtering method in Example 1, thereby achieving dynamic water mist noise filtering. Specifically:

[0120] The model building module is used to build a correlation model between water vapor splash distance and vehicle speed;

[0121] The data acquisition module is used to obtain preset data and the vehicle speed and point cloud data from the BEV perspective collected in the current frame, and to delineate the splash domain in the point cloud data based on the preset data;

[0122] The coarse noise extraction module is used to obtain the coarse noise domain in the current frame point cloud data according to the vehicle speed and the correlation model, and obtain the coarse noise point set contained in the coarse noise domain, wherein the coarse noise domain is located in the splash domain;

[0123] The Euclidean clustering module is used to extract the noise points between the boundary of the coarse noise domain and the boundary of the splash domain according to Euclidean clustering to obtain a clustered noise point set;

[0124] The noise filtering module is used to obtain a fine noise point set based on the coarse noise point set and the clustered noise point set, and to perform distance screening on the fine noise point set based on the center point set of the surrounding intelligent bodies to obtain the final filtered noise point set and filter it out from the point cloud data.

[0125] In this embodiment, the specific working process and working principle of the model construction module, data acquisition module, coarse noise extraction module, Euclidean clustering module and noise filtering module are the same as the method in Example 1, so they will not be repeated in this embodiment.

[0126] Example 3

[0127] This embodiment discloses an unmanned vehicle, which is provided with a transmitter, a receiver, a memory and a processor. Figure 12 The transmitter is used to send instructions and data, the receiver is used to receive instructions and data, the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions stored in the memory to implement some or all of the steps performed by the dynamic water mist noise filtering method in Example 1. The specific implementation process is the same as the dynamic water mist noise filtering method in Example 1.

[0128] It should be noted that the above-mentioned memory can be independent or integrated with the processor. When the memory is set independently, the unmanned vehicle is also provided with a bus for connecting the memory and the processor.

[0129] Example 4

[0130] This embodiment discloses a computer-readable storage medium, which stores computer-executable instructions. When a processor executes the computer-executable instructions, some or all of the steps performed by the dynamic water mist noise filtering method in the above-mentioned embodiment 1 are implemented.

[0131] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A dynamic water mist noise filtering method, characterized in that: The steps include: Step 1: Construct a correlation model between water vapor splash distance and vehicle speed, which is: X max =P xmax ·(k xmax ·V BW +b xmax )+C x X min =P xmin ·(k xmin ·V BW +b xmin )-C x Y max =P ymax ·(k ymax ·V BW +b ymax )+C y Y min =P ymin ·(k ymin ·V BW +b ymin )-C y Among them, X max 、X min They are the maximum and minimum values ​​of the coarse noise domain in the X-axis direction, and the Y max 、Y min are the maximum and minimum values ​​of the coarse noise domain in the Y-axis direction, P xmax 、P xmin 、P ymax 、P ymin X max 、X min 、Y max 、Y min The predicted probability, k xmax 、k xmin 、k ymax 、k ymin X max 、X min 、Y max 、Y min The slope of the fitting function, b xmax 、b xmin 、b ymax 、b ymin X max 、X min 、Y max 、Y min The intercept of the fitting function, V BW is the vehicle speed, C x is the floating constant in the X-axis direction, C y is a floating constant in the Y-axis direction, and the Y-axis direction is the vehicle speed V BW direction; Step 2: Obtain the vehicle speed of the current frame and the point cloud data from the BEV perspective, and delineate the splash area in the point cloud data based on the preset data to determine the vehicle speed V of the current frame. BW Is it greater than the first vehicle speed threshold? If yes, proceed to step 3 to step 5; Otherwise, after obtaining the vehicle speed and point cloud data from the BEV perspective for the next frame, proceed to step 2 again; Step 3: obtaining a coarse noise point domain in the current frame point cloud data based on the vehicle speed and the association model, and obtaining a coarse noise point set contained in the coarse noise point domain, wherein the coarse noise point domain is located in the splash domain; Step 4: extracting noise points between the boundary of the coarse noise domain and the boundary of the splash domain to obtain a clustered noise point set; Step 5: obtain a fine noise point set based on the coarse noise point set and the clustered noise point set, and perform distance screening on the fine noise point set based on the center point set of the surrounding intelligent bodies to obtain the final filtered noise point set and filter it out from the point cloud data.

2. The dynamic water mist noise filtering method according to claim 1, characterized in that: The floating constant C x for: C x =|α1·d 风向 +β1·d 车重 | Among them, d 风向 is the wind force floating constant, d 车重 is the vehicle weight floating constant, α1 and β1 are weight coefficients, and α1>β1; The floating constant C y for: C y =|β2·d 风向 +α2·d 车重 | Among them, α2 and β2 are weight coefficients, and α2>β2.

3. The dynamic water mist noise filtering method according to claim 1 or 2, characterized in that: In the preset data, the maximum and minimum values ​​of the splash domain in the X-axis direction are X max0 、X min0 The maximum and minimum values ​​of the splash domain in the Y-axis direction are Y max0 、Y min0 ; Among them, X max0 ≥X max , Y max0 ≥Y max , X min0 ≤X min , Y min0 ≤Y min .

4. The dynamic water mist noise filtering method according to claim 1 or 2, characterized in that: In step 5, the fine noise point set is: N 细 =N 粗 +N 欧 Among them, N 细 is a set of fine noise points, N 粗 is a coarse noise point set, N 欧 is the clustered noise set; The filtered noise point set is: N 滤 =N 细 -N agent Among them, N 滤 To filter out the noise point set, N agent is the center point set of the surrounding intelligent agents.

5. The method for filtering dynamic water mist noise according to any one of claims 1 to 2, characterized in that: In step 2, when the vehicle speed V collected in the current frame is obtained BW , judge the vehicle speed V BW Is it greater than the second vehicle speed threshold? If so, in step 3, the second vehicle speed threshold is brought into the association model to obtain a coarse noise point domain in the current frame point cloud data; Otherwise, in step 3, the vehicle speed V BW The association model is brought in to obtain the coarse noise point domain in the current frame point cloud data.

6. A dynamic water mist noise filtering device, characterized in that: Noise filtering is performed using the method described in any one of claims 1 to 5, wherein the dynamic water mist noise filtering device comprises: A model building module is used to build a correlation model between water vapor splash distance and vehicle speed; a data acquisition module, configured to acquire preset data and the vehicle speed and point cloud data from the BEV perspective acquired in the current frame, and to delineate a splash domain in the point cloud data based on the preset data; a coarse noise point extraction module, configured to obtain a coarse noise point domain in the point cloud data of the current frame according to the vehicle speed and the correlation model, and obtain a coarse noise point set contained in the coarse noise point domain, wherein the coarse noise point domain is located in the splash domain; A Euclidean clustering module is used to extract noise points between the boundary of the coarse noise domain and the boundary of the splash domain to obtain a clustered noise point set; The noise point filtering module is used to obtain a fine noise point set based on the coarse noise point set and the clustered noise point set, and to perform distance screening on the fine noise point set based on the center point set of the surrounding intelligent bodies to obtain a final filtered noise point set and filter it out from the point cloud data.

7. An unmanned vehicle, characterized in that: The unmanned vehicle is provided with: Memory, used to store programs; A processor is configured to execute the program stored in the memory. When the program is executed, the processor is configured to execute part or all of the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions; when the computer-executable instructions are executed by a processor, they are used to implement part or all of the steps of the method according to any one of claims 1 to 5.

Images