A sanitation vehicle water spraying mist noise removal method based on a laser radar point cloud

Abstract

The application provides a kind of sanitation vehicle water spraying mist noise elimination method based on laser radar point cloud, belong to automatic driving / auxiliary driving field.The method of the application comprises the following steps:S1, carries out multi-source data acquisition and pretreatment operation;S2, carries out dynamic demarcation to water mist suspected area, and calibrates confidence;S3, extracts point cloud feature;S4, carries out multi-feature fusion judgment and noise elimination operation;S5, output elimination result and carry out feedback.Affirmative effect: associate water spraying angle, vehicle speed, water spraying amount with water mist area, only analyze features to suspected area, can reduce calculation amount, adapt to sanitation vehicle embedded controller;Combined with density, intensity, interframe continuity, avoid the one-sidedness of existing scheme " determine noise by intensity", especially solve the misjudgment problem of water mist intensity rising when water spraying amount is large;Threshold value is dynamically adjusted with water spraying amount and vehicle speed, adapts to different working conditions, and the stability of denoising effect is significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving / assisted driving, and specifically relates to a method for removing noise from lidar point clouds in sanitation vehicle operation scenarios. Background Technology

[0002] Existing technologies for noise reduction in water mist from sanitation vehicle sprinklers have the following core shortcomings, failing to meet practical application requirements: The existing solution suffers from poor scene adaptability and low accuracy in removing water mist noise. Water mist differs significantly from natural fog: natural fog is distributed across the entire area, while water mist is only distributed in a fan-shaped area to the side / front of the sprinkler device and dynamically changes with the spraying angle and vehicle speed. The current "full-area threshold" cannot accurately locate the water mist area, leading to either missed water mist removal (e.g., the natural fog threshold misses side water mist) or incorrect deletion of valid targets (e.g., pedestrian point clouds are removed), with a water mist removal accuracy of less than 65%.

[0003] High-performance computing solutions cannot be deployed on sanitation vehicles; the deep neural network of CN114359089A requires GPU hardware support, while the embedded controller on the sanitation vehicle takes more than 120ms to process, far exceeding the sanitation vehicle's perception latency requirement of no more than 100ms, and real-time performance cannot be guaranteed.

[0004] Threshold strategies are rigid and cannot adapt to changes in operational status. Existing solutions have fixed thresholds or only adjust them according to the weather, without considering the operational status of sanitation vehicles. For example, when the water spraying rate increases from 10L / min to 30L / min, the water mist intensity will increase from 20 to 45, and a fixed intensity threshold will retain a large amount of water mist. When the vehicle speed increases from 5km / h to 15km / h, the water mist diffusion range expands, and a fixed area threshold will miss the water mist behind, resulting in large fluctuations in noise reduction effect. Summary of the Invention

[0005] The main objective of this invention is to provide a method for noise removal from water mist generated by sanitation vehicles based on lidar point clouds, thereby solving the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for noise removal of water mist from sanitation vehicles based on lidar point clouds, comprising the following steps: S1. Perform multi-source data acquisition and preprocessing operations; S2. Dynamically delineate the suspected water mist area and assign confidence levels; S3. Extract point cloud features; S4. Perform multi-feature fusion judgment and noise removal operations; S5. Output the rejection results and provide feedback.

[0007] Furthermore, the multi-source data includes point cloud data and job status data; Point cloud data includes: point cloud 3D coordinates, reflection intensity, and timestamp; Operational status data includes: vehicle speed, water volume, water spray switch status, and spray angle.

[0008] Furthermore, the preprocessing operations include point cloud deduplication and coordinate transformation.

[0009] Furthermore, the detailed process of step S2 is as follows: S201. Determine the initial distribution area of ​​the water mist generated by the spray, with the nozzle as the center; The initial distribution area is a circle centered on the nozzle with a central angle of . °, radius is The fan-shaped area; S202. Correct the distribution area of ​​water mist based on vehicle speed to determine the final distribution area of ​​water mist; S203. Confidence level is calibrated based on water volume, and the confidence level is related to the water volume.

[0010] Furthermore, the correction process for the distribution area of ​​the water mist is as follows: Determine the vehicle speed correction factor related to vehicle speed; Set expansion speed ,when When the initial distribution area is taken as the final distribution area of ​​the water mist, the initial distribution area is used as the final distribution area of ​​the water mist. when When the radius of the rear end of the sector is increased, the increased radius value is related to the original radius value; The distance between the front end of the fan-shaped nozzle and the nozzle Let's denote the point as point A, and then denote the distance between the rear end of the fan-shaped section and the nozzle. Let point B be the point of origin. Connect points A and B with a straight line. The resulting closed region is the final distribution region at this point.

[0011] Furthermore, point cloud features include: density features, reflection intensity features, and inter-frame continuity features.

[0012] Furthermore, the density feature is specifically the point cloud density, and its extraction process is as follows: the final distribution area is divided into several voxels, the number of points in each voxel is obtained, and the point cloud density of each voxel is calculated. Reflectance intensity characteristics include: mean intensity and standard deviation of intensity; The inter-frame continuity feature is specifically the inter-frame displacement deviation.

[0013] Furthermore, the detailed process of step S4 is as follows: S401. Perform initial intensity screening based on confidence level and mark suspected noise points; S402. Mark the suspected noise points under the multi-feature judgment conditions as water mist noise points and remove the water mist noise points. If we set a threshold for point cloud density, a threshold for reflection intensity, and a threshold for inter-frame displacement, then the multi-feature judgment condition is: the point cloud density is less than the threshold for point cloud density, the reflection intensity is less than the threshold for reflection intensity, and the inter-frame displacement deviation is greater than the threshold for inter-frame displacement. S403. Set the effective area and retain the point cloud within the effective area to avoid accidentally deleting effective targets due to deviations in area delineation.

[0014] Furthermore, the effective area is divided as follows: [Setting...] The vehicle extends outwards on all sides. Boundary 1 is then obtained, extending outwards around the vehicle. After determining the distance, we obtain boundary 2. The area between boundary 1 and boundary 2 is the effective area.

[0015] Furthermore, the criteria for judging suspected noise points are: the reflection intensity is less than the suspected noise point threshold; the suspected noise point threshold is related to the confidence level.

[0016] The present invention provides a method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds, and its advantages over existing technologies are as follows: (1) By associating the sprinkling angle, vehicle speed, and sprinkling volume with the water mist area, and performing feature analysis only on suspected areas, the computational load can be reduced, which is compatible with the computing power of the sanitation vehicle embedded controller. At the same time, targeted regional feature extraction can reduce the cost of data collection compared to the whole-domain algorithm. (2) Introduce a dynamic suspected area and adjust the range of the dynamic suspected area according to the vehicle speed so that the suspected area is closer to the distribution pattern of water mist; (3) Combining density, intensity, and inter-frame continuity, avoids the one-sidedness of existing solutions that "determine noise based on intensity", especially solving the problem of misjudgment when the water mist intensity increases when the water volume is large; (4) The threshold is dynamically adjusted according to the amount of water sprinkled and the vehicle speed to adapt to different working conditions, and the stability of the noise reduction effect is significantly improved. Attached Figure Description

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is an overall flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the effective area of ​​the present invention; Figure 3 This is a schematic diagram of the water mist point of the present invention; Figure 4 This is a schematic diagram of the non-water mist point of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the embodiments of this application.

[0019] In this application, the terms "exemplarily" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplarily" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplarily" or "for example" is intended to present the relevant concepts in a specific manner.

[0020] Unless otherwise stated, the term "multiple" means two or more.

[0021] Example 1 like Figure 1 As shown, a method for noise reduction of water mist from sanitation vehicle sprinklers based on lidar point clouds includes the following steps: S1. Perform multi-source data acquisition and preprocessing operations; This step is the data processing link of the entire water mist noise removal method. The core purpose is to obtain the point cloud data of the lidar and the vehicle operation status data during the operation of the sanitation vehicle. At the same time, the redundancy and deviation in the original data are eliminated through preprocessing to ensure the accuracy of subsequent area delineation, feature extraction and noise judgment, and to avoid subsequent algorithm judgment errors due to data errors. Multi-source data includes point cloud data and job status data; The data sources are: point cloud data collected using LiDAR and operational status data collected via the chassis CAN bus; Point cloud data includes: point cloud three-dimensional coordinates, reflection intensity, and timestamp; the reflection intensity ranges from 0 to 255; the reflection intensity reflects the strength of the echo signal after the laser beam emitted by the lidar illuminates the target, and is the core basic feature for distinguishing water mist from solid targets. The operational status data includes: vehicle speed, water volume, sprinkler switch status, and sprinkler angle; the vehicle speed range is 0~30km / h, with an accuracy of ±0.5km / h; the water volume range is 0~50L / min, with an accuracy of ±1L / min; the sprinkler switch status includes: on and off; the sprinkler angle refers to the angle between the sprinkler head and the vehicle's forward direction, with the vehicle's forward direction defined as 0°, rightward spray as a positive angle, and leftward spray as a negative angle, with an accuracy of ±1°. In daily operation, the sprinkler angle is generally 0~90°; Preprocessing operations include: point cloud deduplication and coordinate transformation; Point cloud deduplication refers to removing duplicate points at the same spatial location to avoid redundant data caused by hardware errors. Specifically, for two points, if the coordinate deviation between the two points is less than 0.05m, the two points are considered duplicate points. Coordinate transformation is used to calibrate the position of the nozzle within the lidar coordinate system; the origin of the lidar coordinate system is the vertical projection of the lidar's physical geometric center onto the ground. The square axis represents the direction of travel for sanitation vehicles. The positive direction of the axis is within the horizontal driving plane of the sanitation vehicle, perpendicular to... The axis points to the right of the vehicle. The positive direction of the axis is vertically upward; after determining the lidar coordinate system, the coordinates of the nozzle in the lidar coordinate system can be obtained based on the relative position of the nozzle and the lidar.

[0022] S2. Dynamically delineate suspected water mist areas and calibrate their confidence levels. The core purpose of this step is to overcome the shortcomings of existing technologies with "full-domain thresholds" and accurately delineate suspected areas where water mist may actually exist based on the real-time operating status of the sanitation vehicle. Subsequent feature analysis and noise judgment are only performed on these areas, significantly reducing the amount of computation and adapting to the computing power limitations of the sanitation vehicle's embedded controller. At the same time, the confidence level of the area is calibrated based on the water spray volume, providing a basis for the dynamic threshold adjustment of subsequent noise judgment and realizing "linkage between operating status and threshold". The detailed process is as follows: S201. Determine the initial distribution area of ​​the water mist generated by the spray, with the nozzle as the center; The initial distribution area is a circle centered on the nozzle with a central angle of . °, radius is The fan-shaped area; since the spray range of the nozzle is usually 0~90°, the left side of the fan-shaped area is usually in front of the right side of the fan-shaped area. Therefore, the left side of the fan-shaped area is defined as the front end of the fan-shaped area, and the right side of the fan-shaped area is defined as the rear end of the fan-shaped area. S202. Correct the water mist distribution area based on vehicle speed to determine the final water mist distribution area. The correction process is as follows: If the vehicle speed influence coefficient is set to 0.02, then the expression for the vehicle speed correction coefficient is as follows: (1); in, This is the vehicle speed correction factor. For vehicle speed; Set expansion speed If it is 10 m / s, then when When the initial distribution area is taken as the final distribution area of ​​the water mist, the initial distribution area is used as the final distribution area of ​​the water mist. when At that time, because the water mist diffuses backward, the radius of the rear end of the fan shape is expanded, as expressed by: (2); in, To increase the rear radius; the distance between the front end of the fan-shaped nozzle and the nozzle... Let's denote the point as point A, and then denote the distance between the rear end of the fan-shaped section and the nozzle. Let point B be the point of origin. Connect points A and B with a straight line. The resulting closed region is the final distribution region at this point. S203. Based on the water spraying volume, the confidence level is calibrated, and the calibration coefficient is taken as 30. Then the expression for the area confidence level is: (3); in, For confidence level, This represents the amount of water sprayed; the confidence level here serves as the basis for subsequent threshold adjustments.

[0023] S3. Extract point cloud features; The core purpose of this step is to extract core features from the point cloud of the final distribution area of ​​water mist defined in step S2, which can effectively distinguish water mist noise from effective targets, and break through the one-sidedness of the existing technology of "determining noise solely based on reflection intensity", providing quantitative feature indicators for subsequent multi-feature fusion judgment. Point cloud features include: density features, reflection intensity features, and inter-frame continuity features; The density feature is specifically the point cloud density, and its extraction process is as follows: the final distribution area is divided into several voxels, the number of points in each voxel is obtained, and the point cloud density of each voxel can be calculated, as shown in the following expression: (4); in, For the first Point cloud density of individual elements, For the first The number of point clouds in a single unit; Reflectance intensity characteristics include: mean intensity and standard deviation of intensity; The expression for the mean intensity is as follows: (5); in, The average intensity The total number of point clouds, For the first The reflection intensity of a point cloud; The expression for the standard deviation of strength is as follows: (6); in, The standard deviation of strength; The inter-frame continuity feature is specifically the inter-frame displacement deviation, and its expression is as follows: (7); in, This refers to the inter-frame displacement deviation. This represents the point cloud displacement between the current frame and the previous frame. This represents the theoretical displacement between the current frame and the previous frame.

[0024] S4. Perform multi-feature fusion judgment and noise removal operations; The core purpose of this step is to accurately identify and remove water mist noise points based on the three types of quantitative features extracted in step S3, through a three-layer logic of intensity screening, multi-feature fusion judgment, and effective area protection, while absolutely ensuring that the effective target point cloud is not mistakenly deleted; Compared with the single judgment standard of the existing technology, multi-feature fusion judgment greatly improves the accuracy of noise removal and effectively solves the problem of misjudgment when the water mist intensity increases when the water volume is large; The detailed process is as follows: S401. Perform initial intensity screening based on confidence level and mark suspected noise points; The criteria for determining suspected noise points are: (8); in, The point cloud reflection intensity, This is the confidence level correction factor. This is the base value for the intensity threshold; the determination of suspected noise points here is equivalent to using a dynamic threshold. S402. Mark the suspected noise points under the multi-feature judgment conditions as water mist noise points and remove the water mist noise points. The point cloud density threshold is set to 10 points / m. 3 If the reflection intensity threshold is 20 and the inter-frame displacement threshold is 0.3 m / s, then the multi-feature judgment conditions are: the point cloud density is less than the point cloud density judgment threshold, the reflection intensity is less than the reflection intensity judgment threshold, and the inter-frame displacement deviation is greater than the inter-frame displacement judgment threshold. The mathematical expression for this is: (9); in, The point cloud density of the voxel where the suspected noise point is located; S403. Set the effective area and retain the point cloud within the effective area to avoid accidentally deleting effective targets due to deviations in area delineation; like Figure 2 As shown, the effective area is divided as follows: [Setting] The vehicle extends outwards on all sides. Boundary 1 is then obtained, extending outwards around the vehicle. After determining the distance, we obtain boundary 2. The region between boundary 1 and boundary 2 is the effective region. Even if the effective region is within the final distribution area, the point cloud within the effective region will not be processed.

[0025] S5. Output the noise removal results and provide feedback; output the point cloud after noise removal to the backend environmental perception module, and record the operation status data and noise removal effect of each noise removal operation for subsequent fine-tuning of the intensity threshold base value. The core objective of this step is to output the effective point cloud after noise removal to the backend system of the sanitation vehicle's autonomous driving system, so as to realize the practical application of perception data; at the same time, by recording the operation data and the noise removal effect, a closed loop of algorithm self-optimization is formed, so that the water mist noise removal method can be continuously iterated in actual operation, further improving its adaptability and accuracy. Preferably, this embodiment provides a fine-tuning method, the detailed process of which is as follows: First, the operating conditions are divided into multiple categories based on vehicle speed and water volume range; For a single working condition, if the noise removal rate is less than 95% for three consecutive times, the base value of the intensity threshold will be increased. The increase value is determined by the false positive rate. When the false positive rate is 0, it will be increased by 5, and when the false positive rate is greater than 0, it will be increased by 2. Here, the false positive rate refers to the ratio of the number of valid target point clouds that are mistakenly removed to the total number of original valid target point clouds in the suspected water mist area. If the false positive rate is greater than 0 for three consecutive times, the base value of the intensity threshold will be reduced by 1. After fine-tuning, the noise reduction effect of the working condition needs to be continuously tracked. If the noise removal rate is not less than 95% and the retention rate is 100% for 5 consecutive times, then the basic value of the intensity threshold at this time is fixed and the fine-tuning operation ends.

[0026] Example 2 like Figures 3-4 As shown, to illustrate the present invention more in detail, this embodiment uses the example of "a Dongfeng EQ5180TXS sanitation vehicle performing water spraying operations on the right side of a city's Yingbin Avenue (two-way six lanes, dry road surface, no natural precipitation) and identifying pedestrians on the right-side sidewalk" to further explain the implementation process in detail: The LiDAR selected is the RoboSense RS-LIDAR-M1, installed at the front of the vehicle roof. Parameters are set as follows: horizontal resolution 0.9°, vertical resolution 2.0°, and sampling frequency 10Hz. The CAN bus acquisition module is connected to the sanitation vehicle chassis to obtain the following operational status data: vehicle speed 12km / h, water spraying rate 25L / min, spraying angle 60°, central angle 30°, spraying switch status "on", voxel size 0.2m³, intensity threshold base value 30, and confidence correction coefficient 20.

[0027] The lidar outputs one frame of point cloud, containing 82,350 points, of which the right region contains 21,560 points, including approximately 3,200 water mist points, 80 pedestrian points, and 120 guardrail points. First, 128 duplicate points are removed, and then the coordinates of the sprinkler heads are determined in the lidar coordinate system. .

[0028] Since the spraying angle is 60°, the initial distribution area has a fan-shaped front angle of 45° and a fan-shaped rear angle of 75°, with a radius of 5m. Because the vehicle speed is 12km / h, exceeding 10km / h, the rear radius needs to be increased. Therefore: At this point, point A is 5m away from the nozzle at the front end, and point B is 6.2m away from the nozzle at the rear end. Connecting points A and B with a straight line will result in a closed area, which is the final distribution area at this point. The confidence level was calculated as follows: .

[0029] The threshold for judging suspected noise points was calculated to be: 0.83×20+30=46.6; The criteria for identifying suspected noise points were used to identify 3180 suspected noise points. Based on multi-feature criteria, 3118 water mist noise points were filtered out and removed. Valid area determination: [The following is a separate, unrelated section:] It is 0.5m. The sanitation vehicle has dimensions of 8m in length, 2.5m in width, and 2.5m in height. It extends 0.5m outwards to obtain boundary 1, and extends 1m outwards to obtain boundary 2. The area between boundary 1 and boundary 2 is the effective area. 80 pedestrian points and 120 guardrail points are retained within the effective area, as well as 60,770 other point clouds outside the suspected area. After removal, there were a total of 79,224 point clouds, with no obvious water mist points remaining, and the point clouds of pedestrians and guardrails were intact. Performance metrics: Water mist removal rate was 3116 / 3200≈97.4%, effective point retention rate was 100%, and processing time was 52ms, which fully meets the real-time requirements.

[0030] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for noise removal from water mist emitted by sanitation vehicles based on lidar point clouds, characterized in that, Includes the following steps: S1. Perform multi-source data acquisition and preprocessing operations; S2. Dynamically delineate the suspected water mist area and assign confidence levels; S3. Extract point cloud features; S4. Perform multi-feature fusion judgment and noise removal operations; S5. Output the rejection results and provide feedback.

2. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 1, characterized in that, Multi-source data includes point cloud data and job status data; Point cloud data includes: point cloud 3D coordinates, reflection intensity, and timestamp; Operational status data includes: vehicle speed, water volume, water spray switch status, and spray angle.

3. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 1, characterized in that, Preprocessing operations include point cloud deduplication and coordinate transformation.

4. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 1, characterized in that, The detailed process of step S2 is as follows: S201. Determine the initial distribution area of ​​the water mist generated by the spray, with the nozzle as the center; The initial distribution area is a circle centered on the nozzle with a central angle of . °, radius is The fan-shaped area; S202. Correct the distribution area of ​​water mist based on vehicle speed to determine the final distribution area of ​​water mist; S203. Confidence level is calibrated based on water volume, and the confidence level is related to the water volume.

5. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 4, characterized in that, The process for correcting the distribution area of ​​water mist is as follows: Determine the vehicle speed correction factor related to vehicle speed; Set expansion speed ,when When the initial distribution area is taken as the final distribution area of ​​the water mist, the initial distribution area is used as the final distribution area of ​​the water mist. when When the radius of the rear end of the sector is increased, the increased radius value is related to the original radius value; The distance between the front end of the fan-shaped nozzle and the nozzle Let's denote the point as point A, and then denote the distance between the rear end of the fan-shaped section and the nozzle. Let point B be the point of origin. Connect points A and B with a straight line. The resulting closed region is the final distribution region at this point.

6. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 1, characterized in that, Point cloud features include: density features, reflection intensity features, and inter-frame continuity features.

7. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 6, characterized in that, The density feature is specifically the point cloud density, and its extraction process is as follows: the final distribution area is divided into several voxels, the number of points in each voxel is obtained, and the point cloud density of each voxel is calculated. Reflectance intensity characteristics include: mean intensity and standard deviation of intensity; The inter-frame continuity feature is specifically the inter-frame displacement deviation.

8. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 1, characterized in that, The detailed process of step S4 is as follows: S401. Perform initial intensity screening based on confidence level and mark suspected noise points; S402. Mark the suspected noise points under the multi-feature judgment conditions as water mist noise points and remove the water mist noise points. If we set a threshold for point cloud density, a threshold for reflection intensity, and a threshold for inter-frame displacement, then the multi-feature judgment condition is: the point cloud density is less than the threshold for point cloud density, the reflection intensity is less than the threshold for reflection intensity, and the inter-frame displacement deviation is greater than the threshold for inter-frame displacement. S403. Set the effective area and retain the point cloud within the effective area to avoid accidentally deleting effective targets due to deviations in area delineation.

9. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 8, characterized in that, The effective area is divided as follows: set The vehicle extends outwards on all sides. Boundary 1 is then obtained, extending outwards around the vehicle. After determining the distance, we obtain boundary 2. The area between boundary 1 and boundary 2 is the effective area.

10. The method for noise removal from water mist spraying by sanitation vehicles based on lidar point clouds according to claim 8, characterized in that, The criteria for judging suspected noise points are: the reflection intensity is less than the suspected noise point threshold; the suspected noise point threshold is related to the confidence level.

Images