Point cloud noisy point identification method and device, laser radar and computer readable storage medium
By screening and analyzing point cloud points with the middle distance and reflectivity of lidar that meet preset conditions, combining the characteristics of noise clustering, identifying and processing noise, the problem of the impact of lidar in the disturbing environment is solved, and the accuracy of the application is improved reliability.
Patent Information
- Application Number
- CN202310547676.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-15
- Publication Date
- 2024-11-15
AI Technical Summary
Lidar generates noise when encountering interference characteristics such as rain, fog, dust, etc., affecting its application effect in application scenarios such as autonomous driving.
By obtaining the current point cloud in the initial point cloud, filtering the target point cloud with distance and reflectivity that meets the preset conditions, using the characteristics of noise clustering and attribute value distribution, we determine whether the current point cloud is noise, and thus reduce the noise to the laser The impact of radar detection.
Effectively identify and remove or mark noise, improve the accuracy and reliability of lidar in application scenarios such as autonomous driving, and reduce interference to detection results.
Smart Images

Figure CN118962639A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of laser detection, and in particular to a point cloud noise recognition method, device, laser radar and computer-readable storage medium. Background Art
[0002] Lidar is an active remote sensing device that uses photoelectric technology for detection. It combines photoelectric detection technology with laser technology and is an advanced detection method that uses laser as a detection light source. Lidar includes a transmitting module, a scanning module, a receiving module, and a data processing module. The transmitting module is used to transmit a detection signal to the target, the scanning module is used to change the transmission line of the detection signal, the receiving module is used to receive the echo signal of the detection signal, and the data processing module is used to process the echo signal to obtain the distance, reflectivity, speed, size and other information of the detection target. Lidar has been widely used in the fields of autonomous driving, vehicle-road collaboration, logistics vehicles, robots, public smart transportation, etc.
[0003] However, in the actual detection process, the transmitted detection signal may also generate an echo signal when encountering interference features such as rain, fog, and dust. This echo signal is received by the receiving module and detected by the data processing module, and noise points due to rain, fog, and dust will be generated at the corresponding position. If this noise point feature is not processed, it will affect the application of lidar in application scenarios such as autonomous driving. Summary of the invention
[0004] The present application provides a point cloud noise recognition method, device, laser radar and computer-readable storage medium, which can reduce the impact of noise on laser radar detection.
[0005] In a first aspect, the point cloud noise recognition method comprises:
[0006] Acquire a current point cloud point in an initial point cloud, wherein the initial point cloud includes a plurality of point cloud points detected within a preset time period;
[0007] When the distance and reflectivity of the current point cloud point meet the preset conditions, a target point cloud is screened from the initial point cloud, the target point cloud including the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point;
[0008] Screening a valid point cloud from the target point cloud, wherein the valid point cloud includes valid point cloud points whose distance and reflectivity meet the preset conditions;
[0009] Determine whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
[0010] Optionally, when the distance of the current point cloud point is not greater than a first distance threshold, and the reflectivity of the current point cloud point is not greater than a first reflectivity threshold, it is determined that the distance and the reflectivity of the current point cloud point meet a preset condition.
[0011] Optionally, the at least one attribute value of the valid point cloud includes at least one of the following:
[0012] The number distribution of the valid point cloud,
[0013] The distance distribution of the valid point cloud,
[0014] The height distribution of the valid point cloud.
[0015] Optionally, screening the target point cloud from the initial point cloud includes:
[0016] At least one point cloud set is determined from the initial point cloud, and the target point cloud includes the at least one point cloud set, wherein the point cloud points in the same point cloud set are located in the same row or the same column, wherein the difference between the maximum pitch angle and the minimum pitch angle of the point cloud points in the same row is less than a first preset angle, and the difference between the maximum azimuth angle and the minimum azimuth angle of the point cloud points in the same column is less than a second preset angle.
[0017] Optionally, the at least one point cloud set includes:
[0018] A point cloud set including a current point cloud point and a first preset number of point cloud points adjacent to the current point cloud point in a row where the current point cloud point is located; and
[0019] A point cloud set including a current point cloud point and a second preset number of point cloud points adjacent to the current point cloud point in the column where the current point cloud point is located.
[0020] Optionally, the initial point cloud is obtained by scanning a laser radar in a first scanning mode, and the at least one point cloud set is determined from the initial point cloud according to the first scanning mode.
[0021] Optionally, when the first scanning mode makes the arrangement density of the point cloud points in the same row in the initial point cloud greater than the arrangement density of the point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the rows, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the rows is greater than the number of point cloud sets corresponding to the columns; or,
[0022] When the first scanning mode causes the arrangement density of point cloud points in the same row of the initial point cloud to be smaller than the arrangement density of point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the columns, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the columns is greater than the number of point cloud sets corresponding to the rows.
[0023] Optionally, determining whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud includes:
[0024] Determine whether the current point cloud point is a noise point according to at least one of the following:
[0025] The number of valid point cloud points in each point cloud set,
[0026] the continuity of at least one attribute value of the valid point cloud points in each point cloud set,
[0027] The similarity of at least one attribute value between the valid point cloud points in each point cloud set.
[0028] Optionally, determining whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud includes:
[0029] Determine whether the current point cloud point is a noise point according to at least one of the following four items:
[0030] Whether the number of valid point cloud points in each point cloud set is greater than a preset number threshold,
[0031] Whether the continuity of the distance distribution of the valid point cloud points in each point cloud set is less than a preset continuity threshold,
[0032] Whether the similarity of the distance distribution of the valid point cloud points between the point cloud sets is greater than a first preset similarity threshold,
[0033] Whether the similarity of the height distribution of the valid point cloud points between the point cloud sets is greater than a second preset similarity threshold.
[0034] Optionally, whether the current point cloud point is a noise point is determined based on a weighted sum of at least two of the four items.
[0035] Optionally, the continuity of the distance distribution of the valid point cloud points in each point cloud set is calculated according to at least one of the following two items:
[0036] The fitting error of the distance of the valid point cloud points in each point cloud set, wherein the fitting error is calculated based on the difference between the distance of the valid point cloud points in each point cloud set and the corresponding fitting distance, the fitting distance corresponding to the valid point cloud point is the fitting distance obtained based on the position of the valid point cloud point and the fitting function, and the fitting function is the function between the distance and the position obtained by fitting the distance and the position of the valid point cloud points in the point cloud set;
[0037] and / or,
[0038] In each point cloud set, the similarity of the number of positive changes, the number of negative changes, and the number of slight changes in the attribute values of two adjacent valid point cloud points,
[0039] Wherein, when the fitting error is larger, the continuity of the distance distribution of the valid point cloud points in the point cloud set is smaller;
[0040] Among them, when the number of times of positive change, reverse change, and weak change is more similar, the continuity of the distance distribution of the valid point cloud in the point cloud set is smaller.
[0041] Optionally, the similarity of the distance distribution of the valid point cloud points between the point cloud sets is determined according to at least one of the following two items:
[0042] The difference between the distance of the valid point cloud points of each point cloud set and the average distance of the point cloud set;
[0043] The similarity between the average distances of each point cloud.
[0044] Optionally, the similarity of the height distribution of the valid point cloud points between the point cloud sets is determined according to at least one of the following two items:
[0045] The difference between the height of the valid point cloud point in each point cloud set and the average height of the point cloud set;
[0046] The similarity between the average heights of each point cloud.
[0047] Optionally, the method further comprises:
[0048] Determine candidate point cloud points, wherein the candidate point cloud points include point cloud points whose distance and reflectivity meet the preset conditions in the initial point cloud and are determined not to be noise points,
[0049] According to the number of point cloud points determined to be the noise points in the neighborhood point cloud of the candidate point cloud point, the candidate point cloud point is checked to see whether it is the noise point.
[0050] Optionally, the checking whether the candidate point cloud point is the specific noise point according to the number of point cloud points that are the noise points in the neighborhood point cloud of the candidate point cloud point includes:
[0051] When the number of point cloud points that are noise points and are located in the neighborhood point cloud of the candidate point cloud point reaches a second number threshold, the candidate point cloud point is re-checked as the noise point.
[0052] Optionally, the neighborhood point cloud of the candidate point cloud point is an L1*L2 point cloud array including the candidate point cloud point, where L1 and L2 are positive integers respectively.
[0053] Optionally, the method further comprises:
[0054] Eliminate the noise points from the initial point cloud; or,
[0055] The noise points are marked in the initial point cloud.
[0056] In a second aspect, the present application provides a point cloud noise recognition device, comprising:
[0057] An acquisition module, used to acquire a current point cloud point in an initial point cloud, wherein the initial point cloud includes a plurality of point cloud points detected within a preset time period;
[0058] A first screening module, configured to screen a target point cloud from the initial point cloud when the distance and reflectivity of the current point cloud point meet a preset condition, the target point cloud including the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point;
[0059] A second screening module is used to screen a valid point cloud from the target point cloud, wherein the valid point cloud includes valid point cloud points whose distance and reflectivity meet the preset conditions;
[0060] The first determination module is used to determine whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
[0061] In a third aspect, the present application provides a laser radar, comprising:
[0062] Processor; and
[0063] A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute any one of the methods described above.
[0064] In a fourth aspect, the present application provides a computer-readable storage medium storing an executable code, which, when executed by a processor of an electronic device, enables the processor to execute any one of the methods described.
[0065] In an embodiment of the present application, the distance and reflectivity characteristics of noise points corresponding to tiny particle objects (such as rain, fog, dust, etc.) in the current detection environment are utilized, and a point cloud point whose distance and reflectivity meet preset conditions is used as the current point cloud point, and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point is screened out as a valid point cloud. The distribution characteristics of noise points clustering and the attribute values of each noise point are utilized to determine whether the current point cloud point is a noise point based on at least one attribute value of each valid point cloud point in the valid point cloud, so as to reduce the impact of such noise points on the detection of the laser radar. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] Figure 1 is a schematic diagram of an embodiment of a point cloud noise recognition method of the present application;
[0067] Figure 2 is a schematic diagram of an embodiment of a point cloud noise recognition device in the present application;
[0068] Figure 3 It is a schematic diagram of an embodiment of the laser radar in the present application. DETAILED DESCRIPTION
[0069] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.
[0070] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.
[0071] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.
[0072] Lidar is a radar system that emits laser beams to detect the position, speed and other characteristic quantities of a target. Its working principle is to use a laser transmitter to emit multiple laser beams in sequence. If the laser beam encounters an object, it will be reflected, and a receiver will receive the reflected echo. Lidar can be a mechanical Lidar, a semi-solid Lidar or a pure solid-state Lidar. Lidar calculates the time difference between the time of receiving the reflected echo and the time of emitting the laser beam, and the time difference is the flight time of the laser beam. Lidar calculates the distance, direction and other parameters of the object that reflects the echo based on the flight time, and calculates the reflectivity based on the energy of the echo and the energy of the emitted laser beam, thereby detecting the external environment.
[0073] Laser radar generally includes a transceiver module and a processing module. The transceiver module converts the electrical signal into an optical signal and transmits it, and then restores the optical signal reflected by the object into an electrical signal. The processing module calculates relevant information based on the electrical signal. Optionally, the laser radar may also include a scanning module (such as a MEMS two-dimensional galvanometer, a one-dimensional galvanometer, a rotating mirror, a rotating platform, etc., and the present application does not limit the specific form of the scanning module), which is used to change the emission direction of the emitted laser beam to achieve scanning of the laser beam within a certain field of view. The optical path of the reflected echo and the emitted laser can be coaxial, and the reflected echo returns through the scanning module-transceiver module path and is received by the receiver in the transceiver module. The laser radar can be a single-channel scan, or it can realize multi-channel simultaneous or interlaced scanning. In the case of multi-channel scanning, it can be realized by multiple groups of transceiver modules, and the light between each channel (the emitted laser and the reflected echo) is isolated from each other.
[0074] After the laser radar calculates the distance, direction, reflectivity and other parameters of the reflected echo, it uses a point cloud point to represent the object that reflects the echo. The point cloud points collected over a continuous period of time can be used to obtain a frame of point cloud. Generally speaking, the duration of each point cloud frame output by the laser radar is the same, that is, the point cloud points collected in the period are output as a frame of point cloud at fixed intervals. Of course, in some cases, point cloud frames can also be output at a variable frequency, which is not limited here. At least one frame of point cloud acquired by the laser radar can be used for further identification and perception of the external environment.
[0075] However, each echo received by the LiDAR is not necessarily reflected by an object of a certain volume in the external environment. It may be reflected by tiny particles such as dust, rain, fog, hail, etc. These tiny particles are not the detection targets of the LiDAR. The existence of point cloud points corresponding to these tiny particles in the point cloud may affect the decisions made based on point cloud recognition, such as making decisions on the path planning of moving objects (such as cars, robots, etc.) based on obstacles identified by the point cloud. Therefore, the point cloud points corresponding to these tiny particles are considered to be noise points and need to be identified or removed.
[0076] Combine the following Figure 1 The point cloud noise recognition method of this application is introduced by example. Figure 1 As shown, Figure 1 is a schematic diagram of an embodiment of a point cloud noise recognition method of the present application. The point cloud noise recognition method comprises:
[0077] Step S101, obtaining a current point cloud point in an initial point cloud, wherein the initial point cloud includes a plurality of point cloud points detected within a preset time period.
[0078] The initial point cloud in the present application includes multiple point cloud points detected within a preset time length, for example, it may refer to a frame of point cloud, it may refer to a point cloud in a partial area of a frame of point cloud, or it may be a collection of continuous multiple frames of point cloud, without limitation here. The initial point cloud may be acquired by a laser radar. The executor of the point cloud noise recognition method may be the laser radar. For example, after the laser radar acquires a point cloud in a partial area of a frame of point cloud, or acquires a frame of point cloud, or acquires multiple frames of point cloud, it determines which points in the initial point cloud are noise points according to the point cloud noise recognition method in this embodiment.
[0079] For another example, the execution subject of the point cloud noise identification method can be a processing device other than a laser radar, such as a processing device in a control system of a car or a robot. The processing device obtains the initial point cloud scanned by the laser radar, and then determines which points in the initial point cloud are noise points according to the point cloud noise identification method in this embodiment.
[0080] Step S102, when the distance and reflectivity of the current point cloud point meet the preset conditions, a target point cloud is screened from the initial point cloud, wherein the target point cloud includes the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point.
[0081] Optionally, all point cloud points in the initial point cloud may be sequentially used as current point cloud points to perform step S102. Alternatively, only point cloud points in a portion of the field of view of the initial point cloud may be sequentially used as current point cloud points to perform step S102. For example, in an example where point cloud is used for path planning of a moving object, obstacles above the height of the moving object do not affect the path planning of the moving object. When identifying noise points, only point cloud points in the field of view below the height of the moving object in the initial point cloud may be used as current point cloud points for noise point identification.
[0082] The distance and reflectivity of the point cloud points as noise points have certain characteristics. In this application, the distance and reflectivity of the current point cloud points are used as the preliminary screening conditions for noise points by confirming whether the distance and reflectivity of the current point cloud points meet the preset conditions. In one example, when the distance of the current point cloud point is not greater than the first distance threshold, and the reflectivity of the current point cloud point is not greater than the first reflectivity threshold, it is determined that the distance and reflectivity of the current point cloud point meet the preset conditions. Due to the weak ability of small particles such as rain, fog, and dust to reflect laser beams, the reflectivity of the point cloud points corresponding to these small particles is low, and also because of their weak ability to reflect laser beams, the echoes reflected by the farther small particles are too weak to be detected. Therefore, the distance and reflectivity of the point cloud points belonging to noise points in the detected echoes will be less than a certain value respectively. The first distance threshold and the first reflectivity threshold can be determined by experimental values or empirical values. For example, the first distance threshold can be a value between 20 cm and 30 cm.
[0083] Because the noise points corresponding to tiny particulate objects such as rain, fog, and dust generally appear in clusters, when it is determined that the distance and reflectivity of the current point cloud point meet the preset conditions, the target point cloud is determined from the point cloud around the current point cloud point based on the positional relationship with the current point cloud point, and the target point cloud is used to determine whether the current point cloud point is a noise point.
[0084] Step S103: Filter out valid point clouds whose distance and reflectivity meet the preset conditions from the target point cloud.
[0085] Because the noise points corresponding to tiny particles such as rain, fog, and dust generally appear in clusters, their characteristic changes are uneven and relatively random. Therefore, after determining the target point cloud, the valid point clouds whose distance and reflectivity meet the preset conditions are also screened out from the target point cloud. The probability that these valid point clouds are noise points is greater than the probability that the point cloud points in the target point cloud that do not meet the preset conditions are noise points. By analyzing the attribute values of the valid point cloud, it is possible to reduce complexity and improve accuracy compared to analyzing the attribute values of the target point cloud to determine whether the current point cloud point is a noise point.
[0086] Step S104: determining whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
[0087] Optionally, at least one attribute value of the valid point cloud includes at least one of the following: the number distribution of the valid point cloud, the distance distribution of the valid point cloud, and the height distribution of the valid point cloud. For example, the number distribution of the valid point cloud can be whether the number of valid point cloud points is greater than a preset threshold. For example, the distance distribution of the valid point cloud can be the continuity and / or similarity of the change in the distance of each point cloud point in the valid point cloud. For another example, the continuity and / or similarity of the change in the height of each point cloud point in the valid point cloud. Generally, the local distribution of the distance and / or height of the noise points generated by small particles (such as rain, fog and dust) is likely to be random, and there is a small probability that the distribution will be continuous or regular; and overall it has the characteristic of clustering, so the overall distance or height has a certain similarity. By analyzing these attribute values of the valid point cloud points, it can be judged whether the current point cloud point is likely to be a noise point.
[0088] In an embodiment of the present application, the distance and reflectivity characteristics of the noise points corresponding to the tiny particle objects in the current detection environment are utilized, and the point cloud points whose distance and reflectivity meet the preset conditions are taken as the current point cloud points, and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point are screened out as valid point clouds. The noise points are clustered and the distribution characteristics between the attribute values of each noise point are utilized to determine whether the current point cloud point is a noise point based on at least one attribute value of each valid point cloud point in the valid point cloud, so as to reduce the impact of such noise points on the detection of the laser radar.
[0089] In one example, because the noise points corresponding to tiny particles such as rain, fog, and dust generally appear in clusters, and their horizontal and vertical feature changes are uneven and relatively random, the target attribute values of the point cloud points in the target point cloud can be analyzed horizontally and / or vertically. The following describes how to determine the point cloud set for horizontal analysis and / or vertical analysis, and how to determine whether the current point cloud point is a noise point based on the point cloud set.
[0090] In step S102, when selecting a target point cloud from the initial point cloud, at least one point cloud set is determined as a target point cloud according to the spatial arrangement of the point cloud points or according to the scanning order of the laser radar. The point cloud points in the same point cloud set are located in the same row or the same column, and the maximum pitch angle and the minimum pitch angle of the point cloud points in the same row differ by less than a first preset angle, and the maximum azimuth angle and the minimum azimuth angle of the point cloud points in the same column differ by less than a second preset angle.
[0091] For example, taking the point cloud set as a row of point clouds as an example, taking the current point cloud point as the starting point, or as the end point, or as the middle point, determine the preset number of point cloud points adjacent to the current point cloud point and in the same row as the current row. The current row can be taken as the first row, or as the last row, or as the middle row, and at least one row of point cloud points adjacent to the current row is determined to form a point cloud array with the current row as the target point cloud. The method of selecting the target point cloud with the point cloud set as a column of point clouds is similar to the method of selecting the target point cloud with the point cloud set as a row of point clouds.
[0092] Further optionally, the target point cloud includes at least one point cloud set of the corresponding row and at least one point cloud set of the corresponding column. Specifically, the target point cloud includes: a point cloud set of a first preset value of point cloud points adjacent to the current point cloud point in the row where the current point cloud point is located; and a point cloud set of a second preset value of point cloud points adjacent to the current point cloud point in the column where the current point cloud point is located. The first preset value and the second preset value may be the same or different.
[0093] Different scanning methods of the laser radar will result in different distributions of the point cloud. In some examples, the scanning method of the laser radar is to scan along a row and then move to the next row to continue scanning, or to scan along a column and then move to the next column to continue scanning, or to scan along circles or ellipses of different diameters, or to scan along spirals, etc. In this example, a row of point clouds is defined by the pitch angles of each point cloud point, wherein the pitch angles of each point cloud point in a row of point clouds are similar, and the difference between the maximum pitch angle and the minimum pitch angle is less than a first preset angle; and a column of point clouds is defined by the azimuth angles of each point cloud point, wherein the azimuth angles of each point cloud point in a row of point clouds are similar, and the difference between the maximum azimuth angle and the minimum azimuth angle is less than a second preset angle. The first preset angle and the second preset angle may be the same or different.
[0094] Optionally, the initial point cloud is obtained by scanning the laser radar in a first scanning mode, and the at least one point cloud set is determined from the initial point cloud according to the first scanning mode. In one example, when the first scanning mode makes the arrangement density of the point cloud points in the same row of the initial point cloud greater than the arrangement density of the point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the rows, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the rows is greater than the number of point cloud sets corresponding to the columns. In one example, when the first scanning mode makes the arrangement density of the point cloud points in the same row of the initial point cloud less than the arrangement density of the point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the columns, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the columns is greater than the number of point cloud sets corresponding to the rows.
[0095] For example, in practical applications, when the laser radar scans one row at a time and then moves to the next row to continue scanning in sequence, the arrangement density of the point cloud points in the same row is greater than the arrangement density of the point cloud points in the same column. In this way, when selecting the target point cloud from the initial point cloud, the point cloud points obtained by N continuous scans of the current point cloud point in the current row point cloud where the current point cloud point is located can be added to the target point cloud as a point cloud set, or the M point cloud points adjacent to the N point cloud points in each row of at least one row of point cloud adjacent to the current row can be added to the target point cloud as a point cloud set. Wherein, N and M are positive integers, and N and M can be the same or different. Optionally, the point cloud points obtained by X continuous scans of the current point cloud point in the current column point cloud where the current point is located can be added to the target point cloud as a point cloud set, or the Y point cloud points adjacent to the X point cloud points in each column of at least one column of point cloud adjacent to the current column can be added to the target point cloud as a point cloud set. Wherein, X and Y are positive integers, X and Y may be the same or different, and X and M may be the same or different.
[0096] In a specific example, the N point cloud points containing the current point cloud point in the current row, the N point cloud points that are closest to the N point cloud points in the two rows of point clouds scanned after the current row, and the N point cloud points containing the current point cloud point in the current column can be added to the target point cloud as four point cloud sets, a total of a 3*N point cloud array and an X*1 point cloud array.
[0097] In one example, when the first scanning mode causes the arrangement density of point cloud points in the same row of the initial point cloud to be less than the arrangement density of point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the columns, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the columns is greater than the number of point cloud sets corresponding to the rows.
[0098] When the scanning method of the laser radar is to scan along a column in sequence and then move to the next column to continue scanning in sequence, the arrangement density of the point cloud points in the same column is greater than the arrangement density of the point cloud points in the same row. In this way, when selecting the target point cloud from the initial point cloud, the point cloud points obtained by N continuous scans of the current point cloud point in the current column point cloud where the current point cloud point is located can be added to the target point cloud as a point cloud set, or the M point cloud points adjacent to the N point cloud points in each column of at least one column of point cloud adjacent to the current column can be added to the target point cloud as a point cloud set. Optionally, the point cloud points obtained by X continuous scans of the current point cloud point in the current row point cloud where the current point is located can be added to the target point cloud as a point cloud set, or the Y point cloud points adjacent to the X point cloud points in each row of at least one row of point cloud adjacent to the current row can be added to the target point cloud as a point cloud set.
[0099] In a specific example, the X point cloud points in the current column that contain the current point cloud point, the X point cloud points in the two columns of point clouds scanned after the current column that are respectively closest to the X point cloud points, and the N point cloud points in the current row that contain the current point cloud point can be added to the target point cloud as four point cloud sets, a total of a 3*X point cloud array and an N*1 point cloud array.
[0100] In the example where the target point cloud is a plurality of point cloud sets, after all valid point clouds are screened out in each point cloud set, the analysis of the valid point clouds can be performed in units of point cloud sets. In some examples, when determining whether the current point cloud point is a noise point based on at least one attribute value of the valid point cloud, whether the current point cloud point is a noise point can be determined based on at least one of the following four items: (1) whether the number of valid point cloud points in each point cloud set is greater than a preset number threshold, (2) whether the continuity of the distance distribution of the valid point cloud in each point cloud set is less than a preset continuity threshold, (3) whether the similarity of the distance distribution of the valid point cloud between each point cloud set is greater than a first preset similarity threshold, (4) whether the similarity of the height distribution of the valid point cloud between each point cloud set is greater than a second preset similarity threshold. Further optionally, whether the current point cloud point is a noise point can also be determined based on the weighted sum of at least two of the above four items.
[0101] In one example, the continuity of the distance distribution of the valid point cloud in each point cloud set in item (2) can be obtained based on at least one of the following two items: (a) whether the fitting error of the distance of the valid point cloud points in each point cloud set is greater than a preset fitting error threshold, and (b) the similarity of the number of positive changes, the number of reverse changes, and the number of slight changes in the attribute values of two adjacent valid point cloud points in each point cloud set. For example, when the fitting error of the distance of the valid point cloud points in a point cloud set is greater than a preset fitting error threshold, and / or the similarity of the number of positive changes, the number of reverse changes, and the number of slight changes in the attribute values of two adjacent valid point cloud points in the point cloud set is greater than a third preset similarity threshold, the continuity of the distance distribution of the valid point cloud in the point cloud set is less than the preset continuity threshold.
[0102] Among them, in (a), the fitting error is calculated based on the difference between the distance of each point cloud point of the valid point cloud in the current point cloud set and the corresponding fitting distance, and the fitting distance corresponding to the point cloud point is the fitting distance obtained based on the position of the point cloud point and the fitting function, and the fitting function is the function between the distance and the position obtained by fitting the distance and position of each point cloud point of the valid point cloud in the current point cloud set. When the fitting error is larger, the continuity of the distance distribution of the valid point cloud in the point cloud set is smaller.
[0103] In a specific example, the target point cloud is taken as three point cloud sets corresponding to three rows of point clouds respectively, and the valid point clouds in each point cloud set form an array, and the position of each valid point cloud point in the array corresponds to the position in the row. In this way, according to the position of each point cloud point in the array and the distance of the point cloud point, the function y=f(x) between distance and position can be fitted, where x represents the position of the point cloud point in the array, and y represents the distance of the point cloud point. The fitting function can be a quadratic function, a linear function, an inverse proportional function, a multi-time function, etc., which is not limited here. Then calculate the difference between the actual distance of each point cloud point and the fitting distance calculated according to the function y=f(x). The fitting error is calculated according to the gap corresponding to each valid point cloud point. The larger the fitting error, the more irregular the distance arrangement of each point cloud point, the smaller the continuity of the distance distribution of the valid point cloud in the point cloud set, and the smaller the probability that each point cloud point corresponds to the surface of an obstacle object, and the greater the probability that it is a noise point with random distance.
[0104] Among them, in (b), the positive change, negative change, and weak change in a point cloud set refer to the arrangement order of each valid point cloud point in the point cloud set in a row or a column of point clouds. When the difference between the distance of the next point cloud point and the distance of the previous point cloud point is greater than the preset distance value, a positive change occurs; when the difference between the distance of the previous point cloud point and the distance of the next point cloud point is greater than the preset distance value, a negative change occurs; when the absolute value of the difference between the distance of the previous point cloud point and the distance of the next point cloud point is not greater than the preset distance value, a weak change occurs. The specific value of the preset distance value can be determined based on experience or experiments.
[0105] In one example, the similarity of the distance distribution of the valid point cloud between each point cloud set in item (3) can be determined based on at least one of the following two items: (a) the difference between the distance of the valid point cloud point of each point cloud set and the average distance of the point cloud set; (b) the similarity between the average distances of each point cloud set. Specifically, when the distance error calculated by the difference between the distance of each valid point cloud point in each point cloud set and the average distance of the point cloud set is less than the preset distance error, and the similarity of the average distances between each point cloud set is greater than the preset similarity, it is determined that the similarity of the distance distribution of the valid point cloud points between each point cloud set is greater than the first preset similarity threshold.
[0106] In one example, the similarity of the height distribution of the valid point cloud between each point cloud set in item (4) can be determined based on at least one of the following two items: (a) the difference between the height of the valid point cloud point in each point cloud set and the average height of the point cloud set; (b) the similarity between the average heights of each point cloud set. Specifically, when the height error calculated by the difference between the height of each valid point cloud point in each point cloud set and the average height of the point cloud set is less than the preset height error, and the similarity of the average heights between each point cloud set is greater than the preset similarity, it is determined that the similarity of the height distribution of the valid point cloud points between each point cloud set is greater than the second preset similarity threshold.
[0107] Since the noise points corresponding to tiny particles generally have the characteristic of being clustered, the above items (3) and (4) are used to analyze whether the distance or height of the effective point cloud points in general has a certain similarity to determine whether the current point cloud point is likely to be a noise point.
[0108] Optionally, after taking each point cloud point in the initial point cloud as the current point cloud point in turn and determining all the point cloud points that are noise points according to the above-mentioned point cloud noise point identification method, the point cloud points that are confirmed not to be noise points in the valid point cloud can also be reviewed to see if they are noise points by utilizing the overall clustering characteristics of the noise points corresponding to the tiny particle objects. Specifically, the point cloud noise point identification method in the present application also includes: determining candidate point cloud points, the candidate point cloud points including the point cloud points whose distance and reflectivity meet the preset conditions in the initial point cloud and are determined not to be noise points; and reviewing whether the candidate point cloud points are noise points according to the number of point cloud points that are determined to be noise points and are located in the neighborhood point cloud of the candidate point cloud points.
[0109] There are multiple ways to determine the neighborhood point cloud of the candidate point cloud point. For example, the point cloud array L1*L2 including the candidate point cloud point can be used as the neighborhood point cloud, where L1 and L2 are positive integers respectively. The candidate point cloud point can be a point cloud at any position of the point cloud array (for example, at the center or close to the center). There may be multiple conditions for verifying whether it is the noise point. For example, when the number of point cloud points that are noise points in the neighborhood point cloud of the candidate point cloud point reaches a second quantity threshold, the candidate point cloud point is verified as the noise point. Optionally, the point cloud points in the initial point cloud that meet the preset conditions in terms of distance and reflectivity and are determined not to be noise points are reviewed in turn as the candidate point cloud points.
[0110] After the noise points are determined from the initial point cloud, the point cloud points determined as noise points can be optionally marked, or the point cloud points determined as noise points can be directly removed from the initial point cloud. Optionally, when marking the noise points, the point cloud points determined as noise points in the initial judgment (that is, determined as noise points according to steps S101 to S104) and the point cloud points determined as noise points during the recheck judgment can be marked in the same way, or different ways can be used for marking, for example, the invalid value of the point cloud points determined as noise points in the initial judgment can be marked as 0, and the invalid value of the point cloud points determined as noise points during the recheck judgment can be marked as 1.
[0111] The present application also provides a point cloud noise recognition device. Figure 2 As shown, Figure 2 2 is a schematic diagram of an embodiment of a point cloud noise recognition device in an embodiment of the present application. The laser ranging device 200 includes:
[0112] An acquisition module 201 is used to acquire a current point cloud point in an initial point cloud, where the initial point cloud includes a plurality of point cloud points detected within a preset time period;
[0113] A first screening module 202 is used to screen a target point cloud from the initial point cloud when the distance and reflectivity of the current point cloud point meet a preset condition, the target point cloud including the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point;
[0114] A second screening module 203 is used to screen valid point clouds from the target point cloud, wherein the valid point clouds include valid point cloud points whose distance and reflectivity meet the preset conditions;
[0115] The first determination module 204 is used to determine whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
[0116] Optionally, when the distance of the current point cloud point is not greater than a first distance threshold, and the reflectivity of the current point cloud point is not greater than a first reflectivity threshold, it is determined that the distance and the reflectivity of the current point cloud point meet a preset condition.
[0117] Optionally, the at least one attribute value of the valid point cloud includes at least one of the following:
[0118] The number distribution of the valid point clouds,
[0119] The distance distribution of the valid point cloud,
[0120] The height distribution of the valid point cloud.
[0121] Optionally, when screening the target point cloud from the initial point cloud, the first screening module 202 is specifically used to:
[0122] At least one point cloud set is determined from the initial point cloud, and the target point cloud includes the at least one point cloud set, wherein the point cloud points in the same point cloud set are located in the same row or the same column, wherein the difference between the maximum pitch angle and the minimum pitch angle of the point cloud points in the same row is less than a first preset angle, and the difference between the maximum azimuth angle and the minimum azimuth angle of the point cloud points in the same column is less than a second preset angle.
[0123] Optionally, the at least one point cloud set includes:
[0124] A point cloud set including a current point cloud point and a first preset number of point cloud points adjacent to the current point cloud point in a row where the current point cloud point is located; and
[0125] A point cloud set including a current point cloud point and a second preset number of point cloud points adjacent to the current point cloud point in the column where the current point cloud point is located.
[0126] Optionally, the initial point cloud is obtained by scanning a laser radar in a first scanning mode, and the at least one point cloud set is determined from the initial point cloud according to the first scanning mode.
[0127] Optionally, when the first scanning mode makes the arrangement density of the point cloud points in the same row in the initial point cloud greater than the arrangement density of the point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the rows, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the rows is greater than the number of point cloud sets corresponding to the columns; or,
[0128] When the first scanning mode causes the arrangement density of point cloud points in the same row of the initial point cloud to be smaller than the arrangement density of point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the columns, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the columns is greater than the number of point cloud sets corresponding to the rows.
[0129] Optionally, when determining whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud, the first determining module 204 is specifically configured to:
[0130] Determine whether the current point cloud point is a noise point according to at least one of the following:
[0131] The number of valid point cloud points in each point cloud set,
[0132] the continuity of at least one attribute value of the valid point cloud points in each point cloud set,
[0133] The similarity of at least one attribute value between the valid point cloud points in each point cloud set.
[0134] Optionally, when determining whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud, the first determining module 204 is specifically configured to:
[0135] Determine whether the current point cloud point is a noise point according to at least one of the following four items:
[0136] Whether the number of valid point cloud points in each point cloud set is greater than a preset number threshold,
[0137] Whether the continuity of the distance distribution of the valid point cloud points in each point cloud set is less than a preset continuity threshold,
[0138] Whether the similarity of the distance distribution of the valid point cloud points between the point cloud sets is greater than a first preset similarity threshold,
[0139] Whether the similarity of the height distribution of the valid point cloud points between the point cloud sets is greater than a second preset similarity threshold.
[0140] Optionally, whether the current point cloud point is a noise point is determined based on a weighted sum of at least two of the four items.
[0141] Optionally, the continuity of the distance distribution of the valid point cloud points in each point cloud set is calculated according to at least one of the following two items:
[0142] The fitting error of the distance of the valid point cloud points in each point cloud set, wherein the fitting error is calculated based on the difference between the distance of the valid point cloud points in each point cloud set and the corresponding fitting distance, the fitting distance corresponding to the valid point cloud point is the fitting distance obtained based on the position of the valid point cloud point and the fitting function, and the fitting function is the function between the distance and the position obtained by fitting the distance and the position of the valid point cloud points in the point cloud set;
[0143] and / or,
[0144] In each point cloud set, the similarity of the number of positive changes, the number of negative changes, and the number of slight changes in the attribute values of two adjacent valid point cloud points,
[0145] Wherein, when the fitting error is larger, the continuity of the distance distribution of the valid point cloud points in the point cloud set is smaller;
[0146] Among them, when the number of times of positive change, reverse change, and weak change is more similar, the continuity of the distance distribution of the valid point cloud in the point cloud set is smaller.
[0147] Optionally, the similarity of the distance distribution of the valid point cloud points between the point cloud sets is determined according to at least one of the following two items:
[0148] The difference between the distance of the valid point cloud points of each point cloud set and the average distance of the point cloud set;
[0149] The similarity between the average distances of each point cloud.
[0150] Optionally, the similarity of the height distribution of the valid point cloud points between the point cloud sets is determined according to at least one of the following two items:
[0151] The difference between the height of the valid point cloud point in each point cloud set and the average height of the point cloud set;
[0152] The similarity between the average heights of each point cloud.
[0153] Optionally, the device further comprises:
[0154] The second determination module is used to determine candidate point cloud points, wherein the candidate point cloud points include point cloud points whose distance and reflectivity meet the preset conditions in the initial point cloud and are determined not to be noise points.
[0155] The review module is used to review whether the candidate point cloud point is the noise point according to the number of point cloud points determined to be the noise points in the neighborhood point cloud of the candidate point cloud point.
[0156] Optionally, when the review module reviews whether the candidate point cloud point is the specific noise point according to the number of point cloud points that are the noise points in the neighborhood point cloud of the candidate point cloud point, it is specifically used to:
[0157] When the number of point cloud points that are noise points and are located in the neighborhood point cloud of the candidate point cloud point reaches a second number threshold, the candidate point cloud point is re-checked as the noise point.
[0158] Optionally, the neighborhood point cloud of the candidate point cloud point is an L1*L2 point cloud array including the candidate point cloud point, where L1 and L2 are positive integers respectively.
[0159] Optionally, the device further comprises:
[0160] A removal module, used to remove the noise points from the initial point cloud; or,
[0161] A marking module is used to mark the noise points in the initial point cloud.
[0162] The point cloud noise identification device in the embodiment of the present application utilizes the distance and reflectivity characteristics of the noise points corresponding to the tiny particle objects (such as rain, fog, dust, etc.) in the current detection environment. The point cloud points whose distance and reflectivity meet the preset conditions are used as the current point cloud points, and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud points are screened out as valid point clouds. The noise points are clustered and the distribution characteristics between the attribute values of each noise point are utilized to determine whether the current point cloud point is a noise point based on at least one attribute value of each valid point cloud point in the valid point cloud, so as to reduce the influence of such noise points on the detection of the laser radar.
[0163] The present application also provides a laser radar, such as Figure 3 As shown, Figure 3 FIG. 3 is a schematic diagram of an embodiment of a laser radar in an embodiment of the present application. The laser radar 30 includes a memory 31 and a processor 32 .
[0164] The processor 32 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.
[0165] The memory 31 may include various types of storage units, such as system memory, read-only memory (ROM) and permanent storage devices. Among them, ROM can store static data or instructions required by the processor 32 or other modules of the computer. The permanent storage device can be a readable and writable storage device. The permanent storage device can be a non-volatile storage device that will not lose the stored instructions and data even if the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, an optical drive). The system memory can be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory can store some or all instructions and data required by the processor at run time. In addition, the memory 61 can include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks can also be used. In some embodiments, the memory 510 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and transient electronic signals transmitted wirelessly or wired.
[0166] The memory 31 stores executable codes, and when the executable codes are processed by the processor 32 , the processor 32 can execute part or all of the methods described above.
[0167] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.
[0168] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.
[0169] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.
Claims
1. A point cloud noise recognition method, It is characterized in that include: Acquire a current point cloud point in an initial point cloud, wherein the initial point cloud includes a plurality of point cloud points detected within a preset time period; When the distance and reflectivity of the current point cloud point meet the preset conditions, a target point cloud is screened from the initial point cloud, the target point cloud including the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point; Screening a valid point cloud from the target point cloud, wherein the valid point cloud includes valid point cloud points whose distance and reflectivity meet the preset conditions; Determine whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
2. The method according to claim 1, It is characterized in that When the distance of the current point cloud point is not greater than a first distance threshold, and the reflectivity of the current point cloud point is not greater than a first reflectivity threshold, it is determined that the distance and the reflectivity of the current point cloud point meet a preset condition.
3. The method according to claim 1, It is characterized in that The at least one attribute value of the valid point cloud includes at least one of the following: The number distribution of the valid point clouds, The distance distribution of the valid point cloud, The height distribution of the valid point cloud.
4. The method according to any one of claims 1 to 3, It is characterized in that The step of screening the target point cloud from the initial point cloud comprises: At least one point cloud set is determined from the initial point cloud, and the target point cloud includes the at least one point cloud set, wherein the point cloud points in the same point cloud set are located in the same row or the same column, wherein the difference between the maximum pitch angle and the minimum pitch angle of the point cloud points in the same row is less than a first preset angle, and the difference between the maximum azimuth angle and the minimum azimuth angle of the point cloud points in the same column is less than a second preset angle.
5. The method according to claim 4, It is characterized in that The at least one point cloud set comprises: A point cloud set including a current point cloud point and a first preset number of point cloud points adjacent to the current point cloud point in a row where the current point cloud point is located; and A point cloud set including a current point cloud point and a second preset number of point cloud points adjacent to the current point cloud point in the column where the current point cloud point is located.
6. The method according to claim 4, It is characterized in that The initial point cloud is obtained by scanning the laser radar in a first scanning mode, and the at least one point cloud set is determined from the initial point cloud according to the first scanning mode.
7. The method according to claim 6, It is characterized in that When the first scanning mode causes the arrangement density of the point cloud points in the same row in the initial point cloud to be greater than the arrangement density of the point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the rows, or includes at least two point cloud sets corresponding to the rows and columns, and the number of the point cloud sets corresponding to the rows is greater than the number of the point cloud sets corresponding to the columns; or, When the first scanning mode causes the arrangement density of point cloud points in the same row of the initial point cloud to be smaller than the arrangement density of point cloud points in the same column, the at least one point cloud set includes at least two point cloud sets corresponding to the columns, or includes at least two point cloud sets corresponding to the rows and columns, and the number of point cloud sets corresponding to the columns is greater than the number of point cloud sets corresponding to the rows.
8. A point cloud noise recognition device, It is characterized in that include: An acquisition module, used to acquire a current point cloud point in an initial point cloud, wherein the initial point cloud includes a plurality of point cloud points detected within a preset time period; A first screening module, configured to screen a target point cloud from the initial point cloud when the distance and reflectivity of the current point cloud point meet a preset condition, the target point cloud including the current point cloud point and at least part of the point cloud points detected within a preset time range before and / or after the current point cloud point; A second screening module is used to screen a valid point cloud from the target point cloud, wherein the valid point cloud includes valid point cloud points whose distance and reflectivity meet the preset conditions; The first determination module is used to determine whether the current point cloud point is a noise point according to at least one attribute value of the valid point cloud.
9. A laser radar, It is characterized in that include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, It is characterized in that An executable code is stored, and when the executable code is executed by a processor of an electronic device, the processor is caused to execute the method according to any one of claims 1 to 7.
