CN-116823668-B - Three-dimensional point cloud data processing method and system

CN116823668BCN 116823668 BCN116823668 BCN 116823668BCN-116823668-B

Abstract

The invention discloses a three-dimensional point cloud data processing method and system, which are used for acquiring three-dimensional point cloud data acquired at each moment of a vehicle-mounted laser radar, constructing a to-be-processed point cloud frame list according to the acquired data, constructing an initial starting point Yun Zhen list according to the to-be-processed point cloud frame list, constructing an ending point cloud frame list according to the initial to-be-processed point cloud frame list, updating the initial starting point cloud frame list, taking out one-to-one corresponding ending point cloud frames and starting points Yun Zhen from the ending point cloud frame list and the updated starting point cloud frame list, taking out all point cloud frames with acquisition time between the starting point cloud frames and the ending point cloud frames from the to-be-processed point cloud frame list, merging all point cloud frames, the starting point cloud frames and the ending point cloud frames between the starting point cloud frames to obtain original three-dimensional point cloud image segments, denoising each original three-dimensional point cloud image segment to obtain processed three-dimensional point cloud image segments, and constructing a data index table.

Inventors

CHEN GUIYOU
LIU CHENGXIN

Assignees

山东大学

Dates

Publication Date: 20260505
Application Date: 20230707

Claims (8)

1. The three-dimensional point cloud data processing method is characterized by comprising the following steps of: acquiring three-dimensional point cloud data acquired at each moment of the vehicle-mounted laser radar, and acquiring real-time pose of the vehicle-mounted laser radar and real-time coordinates of a vehicle; constructing an initial starting point Yun Zhen list with equidistant intervals or time intervals according to the point cloud frame list to be processed; Constructing an ending point cloud frame list according to the initial starting point cloud frame list and the point cloud frame list to be processed, and simultaneously updating the initial starting point cloud frame list to obtain an updated starting point Yun Zhen list, and obtaining a starting point cloud frame list and an ending point cloud frame list which are equal in corresponding distance or time interval; according to the initial starting point cloud frame list and the to-be-processed point cloud frame list, an ending point cloud frame list is constructed, and meanwhile, the initial starting point cloud frame list is updated to obtain an updated starting point Yun Zhen list, and the starting point cloud frame list and the ending point cloud frame list with equal corresponding distance intervals are obtained, specifically comprising the following steps: 3-a1 setting a distance interval between a start point Yun Zhen and a corresponding end point cloud frame ; 3-A2, creating an empty list named as an end point cloud frame list; 3-a3, traversing the initial starting point Yun Zhen list according to the time stamp ascending order, for each element in the initial starting point cloud frame list Traversing the to-be-processed point cloud frame list, and assuming that a group of point cloud frames taken out from the to-be-processed point cloud frame list are The acquisition time is within The first group thereafter satisfies A kind of electronic device Placing an end point cloud frame list, wherein, Is that And (3) with Is provided with a coordinate spacing of (c) in the plane of the optical axis, Is greater than Is a time stamp of (2); if the satisfaction is not found after traversing the to-be-processed point cloud frame list A kind of electronic device Then consider from Can not be generated to meet the set distance interval Point cloud image segments of (2) will Deleting from the starting point cloud frame list; 3-a4, repeating the step 3-a3 until all elements in the initial starting point cloud frame list are traversed, and further obtaining an ending point cloud frame list and an updated starting point Yun Zhen list; Or, constructing an end point cloud frame list according to the initial start point cloud frame list and the to-be-processed point cloud frame list, and simultaneously updating the initial start point cloud frame list to obtain an updated start point Yun Zhen list, and obtaining a start point cloud frame list and an end point cloud frame list with equal corresponding time intervals, wherein the method specifically comprises the following steps: 3-b1 setting a time interval between a start point Yun Zhen and a corresponding end point cloud frame ; 3-B2, creating an empty list named as an end point cloud frame list; 3-b3 traversing the list of initial starting points Yun Zhen in ascending order according to the time stamp, for each element in the list of initial starting point cloud frames Traversing the to-be-processed point cloud frame list, and assuming that a group of point cloud frames taken out from the to-be-processed point cloud frame list are The acquisition time is within The first group thereafter satisfies A kind of electronic device Placing an end point cloud frame list, wherein, Is that And (3) with Is used to determine the time stamp difference value of (c), Is greater than Is a time stamp of (2); If the satisfaction is not found after traversing the to-be-processed point cloud frame list A kind of electronic device Then consider from Failure to generate a signal meeting the set time interval Point cloud image segments of (2) will Deleting from the starting point cloud frame list; 3-b4, repeating 3-b3 until all elements in the initial starting point cloud frame list are traversed, further obtaining an ending point cloud frame list and an updated starting point Yun Zhen list, The method comprises the steps of sequentially taking out an end point cloud frame and an initial point Yun Zhen which are in one-to-one correspondence from an end point cloud frame list and an updated initial point cloud frame list, taking out all point cloud frames with acquisition time between the initial point cloud frame and the end point cloud frame from a to-be-processed point cloud frame list, merging the corresponding initial point cloud frame and the end point cloud frame and all point cloud frames between the corresponding initial point cloud frame and the end point cloud frame to obtain an original three-dimensional point cloud image segment; Denoising each original three-dimensional point cloud image segment to obtain a processed three-dimensional point cloud image segment, and removing unusable three-dimensional point cloud image segments; And constructing a data index table according to the processed three-dimensional point cloud image segments.
2. The method for processing three-dimensional point cloud data according to claim 1, wherein constructing a list of equidistant initial starting points Yun Zhen according to the list of point cloud frames to be processed specifically comprises: 2a-1 designating a distance interval between two adjacent starting point cloud frames ; 2A-2, creating an empty list named "initial starting point Yun Zhen list"; 2a-3, putting the first group of point cloud frame data in the point cloud frame list to be processed into an initial starting point Yun Zhen list; 2a-4, starting traversing elements of the point cloud frame list to be processed, and calculating coordinate distances between each traversed group of point cloud frame data and the latest group of point cloud frame data in the initial starting point cloud frame list Will first satisfy the coordinate spacing Greater than or equal to Adding the point cloud frame data of (1) into an initial starting point cloud frame list; 2a-5, repeating the step 2a-4 until all elements of the point cloud frame list to be processed are traversed, and further obtaining an initial starting point Yun Zhen list with equidistant intervals.
3. The method for processing three-dimensional point cloud data according to claim 1, wherein constructing a list of initial starting points Yun Zhen of equal time intervals according to the list of point cloud frames to be processed specifically comprises: 2b-1 designating a time interval between two adjacent starting point cloud frames ; 2B-2, creating an empty list named "initial starting point Yun Zhen list"; 2b-3, putting the first group of point cloud frame data in the point cloud frame list to be processed into an initial starting point Yun Zhen list; 2b-4, starting traversing elements of the point cloud frame list to be processed, and calculating time intervals between each traversed group of point cloud frame data and the latest group of point cloud frame data in the initial starting point cloud frame list Will first meet the time interval Greater than or equal to Adding the point cloud frame data of (1) into an initial starting point cloud frame list; 2b-5, repeating the step 2b-4 until all elements of the point cloud frame list to be processed are traversed, and further obtaining an initial starting point Yun Zhen list with equal time intervals.
4. The method of claim 1, wherein the one-to-one corresponding end point cloud frame and start point Yun Zhen are sequentially extracted from the end point cloud frame list and the updated start point cloud frame list, all point cloud frames with acquisition time between the start point cloud frame and the end point cloud frame are extracted from the to-be-processed point cloud frame list, and the corresponding end point cloud frame and all point cloud frames between the corresponding start point cloud frame and the end point cloud frame are combined to obtain the original three-dimensional point cloud image segment, and the method further comprises the steps of: Is arranged at At moment, the two-dimensional or low-density three-dimensional point cloud frame acquired by the laser radar is Wherein the coordinates of the single point are The total number of points is Euler angle posture data acquired by the inertial sensor are as follows 、 And The coordinate data acquired by the GPS sensor is that 、 And ; The rotation matrix can be calculated by the gesture data The method comprises the following steps: By passing through Can be calculated to The actual coordinates of the points in (a) are A three-dimensional point set can be generated by combining a plurality of transformed point cloud data Three-dimensional point set Referred to as a "three-dimensional point cloud image segment" or "point cloud image segment".
5. The method for processing three-dimensional point cloud data according to claim 1, wherein denoising is performed on each original three-dimensional point cloud image segment to obtain a processed three-dimensional point cloud image segment, and removing unusable three-dimensional point cloud image segments comprises: Performing histogram analysis on Z-axis coordinates of all points in each original three-dimensional point cloud image fragment, judging whether each point belongs to a discrete point or the ground, and if so, removing the point from a point set; calculating the variance of each axial coordinate of the three-dimensional point cloud image segment, and if one or more axial coordinate variances of the three-dimensional point cloud image segment are smaller than a set threshold value, determining that details of the current three-dimensional point cloud image segment are seriously lost, belonging to unavailable three-dimensional point cloud image segments, and eliminating the current three-dimensional point cloud image segment; and judging whether the total point number in the three-dimensional point cloud image segment is larger than a set threshold value, if so, performing downsampling processing on the current three-dimensional point cloud image segment, and if so, removing the current three-dimensional point cloud image segment or performing upsampling processing on the current three-dimensional point cloud image segment to obtain the three-dimensional point cloud image segment with uniform size.
6. The method for processing three-dimensional point cloud data according to claim 1, wherein constructing a data index table according to the processed three-dimensional point cloud image segments comprises: Storing the three-dimensional point cloud image segments with uniform sizes to a storage position designated by a user; Reading a storage path, a time stamp and coordinates of the three-dimensional point cloud image fragment, and generating a data packet for storing the index information of the point cloud image fragment according to the read information; The data packet comprises a point cloud image fragment list, and each element of the point cloud image fragment list comprises a storage path of a point cloud image fragment, a time stamp and coordinates of a start point cloud frame, a time stamp and coordinates of an end point cloud frame and coordinates of all point cloud frames of the current point cloud image fragment.
7. A three-dimensional point cloud data processing system, comprising: The first module is configured to acquire three-dimensional point cloud data acquired at each moment of the vehicle-mounted laser radar, acquire real-time pose of the vehicle-mounted laser radar and real-time coordinates of a vehicle, and construct a point cloud frame list to be processed according to the acquired data; The second module is configured to construct an initial starting point Yun Zhen list according to a to-be-processed point cloud frame list, construct an end point cloud frame list according to the initial starting point cloud frame list and the to-be-processed point cloud frame list, update the initial starting point cloud frame list to obtain an updated starting point Yun Zhen list, take out one-to-one corresponding end point cloud frames and starting points Yun Zhen from the end point cloud frame list and the updated starting point cloud frame list, take out all point cloud frames with acquisition time between the starting point cloud frames and the end point cloud frames from the to-be-processed point cloud frame list, merge all point cloud frames, the starting point cloud frames and the end point cloud frames between the starting point cloud frames and the end point cloud frames to obtain original three-dimensional point cloud image segments, and denoise each original three-dimensional point cloud image segment to obtain a processed three-dimensional point cloud image segment; according to the initial starting point cloud frame list and the to-be-processed point cloud frame list, an ending point cloud frame list is constructed, and meanwhile, the initial starting point cloud frame list is updated to obtain an updated starting point Yun Zhen list, and the starting point cloud frame list and the ending point cloud frame list with equal corresponding distance intervals are obtained, specifically comprising the following steps: 3-a1 setting a distance interval between a start point Yun Zhen and a corresponding end point cloud frame ; 3-A2, creating an empty list named as an end point cloud frame list; 3-a3, traversing the initial starting point Yun Zhen list according to the time stamp ascending order, for each element in the initial starting point cloud frame list Traversing the to-be-processed point cloud frame list, and assuming that a group of point cloud frames taken out from the to-be-processed point cloud frame list are The acquisition time is within The first group thereafter satisfies A kind of electronic device Placing an end point cloud frame list, wherein, Is that And (3) with Is provided with a coordinate spacing of (c) in the plane of the optical axis, Is greater than Is a time stamp of (2); if the satisfaction is not found after traversing the to-be-processed point cloud frame list A kind of electronic device Then consider from Can not be generated to meet the set distance interval Point cloud image segments of (2) will Deleting from the starting point cloud frame list; 3-a4, repeating the step 3-a3 until all elements in the initial starting point cloud frame list are traversed, and further obtaining an ending point cloud frame list and an updated starting point Yun Zhen list; Or, constructing an end point cloud frame list according to the initial start point cloud frame list and the to-be-processed point cloud frame list, and simultaneously updating the initial start point cloud frame list to obtain an updated start point Yun Zhen list, and obtaining a start point cloud frame list and an end point cloud frame list with equal corresponding time intervals, wherein the method specifically comprises the following steps: 3-b1 setting a time interval between a start point Yun Zhen and a corresponding end point cloud frame ; 3-B2, creating an empty list named as an end point cloud frame list; 3-b3 traversing the list of initial starting points Yun Zhen in ascending order according to the time stamp, for each element in the list of initial starting point cloud frames Traversing the to-be-processed point cloud frame list, and assuming that a group of point cloud frames taken out from the to-be-processed point cloud frame list are The acquisition time is within The first group thereafter satisfies A kind of electronic device Placing an end point cloud frame list, wherein, Is that And (3) with Is used to determine the time stamp difference value of (c), Is greater than Is a time stamp of (2); If the satisfaction is not found after traversing the to-be-processed point cloud frame list A kind of electronic device Then consider from Failure to generate a signal meeting the set time interval Point cloud image segments of (2) will Deleting from the starting point cloud frame list; 3-b4, repeating 3-b3 until all elements in the initial starting point cloud frame list are traversed, further obtaining an ending point cloud frame list and an updated starting point Yun Zhen list, The third module is configured to construct a data index table according to the processed three-dimensional point cloud image segments, and upload the data index table and the processed three-dimensional point cloud image segments to the fourth module in a wireless transmission mode; and a fourth module configured to implement data reception and storage.
8. The three-dimensional point cloud data processing system of claim 7, wherein each of said first module, second module, third module and fourth module are independently operable, one or more of said modules being configured to perform different data processing tasks including, but not limited to: the first module, the second module and the third module form a three-dimensional point cloud data processing system, and the three-dimensional point cloud data processing system is used for realizing automatic acquisition and processing of three-dimensional point cloud data and generating a three-dimensional point cloud image fragment and an index table; in a second sample, the first module, the second module and the third module form a three-dimensional point cloud data processing subsystem, the number of the three-dimensional point cloud data processing subsystems is one or more, each three-dimensional point cloud data processing subsystem uploads own data to the fourth module, the fourth module is responsible for centralized storage and distribution of the data, and the fourth module and all three-dimensional point cloud data processing equipment form the three-dimensional point cloud data processing system together; In a third example, the second module and the third module form a three-dimensional point cloud data processing system, and the three-dimensional point cloud data processing system directly processes data stored in a computer-readable storage medium to generate three-dimensional point cloud image fragments and an index table, wherein the data stored in the computer-readable storage medium comprises three-dimensional point cloud data, real-time pose of a vehicle-mounted laser radar and real-time coordinates of a vehicle.

Description

Three-dimensional point cloud data processing method and system Technical Field The invention relates to the technical field of three-dimensional point cloud data acquisition and processing, in particular to a three-dimensional point cloud data processing method and system. Background The statements in this section merely relate to the background of the present disclosure and may not necessarily constitute prior art. In recent years, with the continuous development of sensor technology, information processing technology, and integrated circuit manufacturing technology, the unmanned area has come to receive a great deal of attention. The three-dimensional point cloud data has the characteristics of strong stability and large information quantity, and becomes important data required by the unmanned system for realizing the functions of environment perception, path decision, emergency event processing and the like. The multi-line laser radar has the characteristics of high acquisition speed, multiple redundant data, relatively fixed visual field range and the like. The existing unmanned system processes laser point cloud data containing environment information in a data processing mode based on algorithms such as deep learning or reinforcement learning, so that operations such as environment sensing and high-precision map matching are realized. The data processing algorithm based on deep learning or reinforcement learning is very sensitive to the organization structure, quantity, contained information quantity and other factors of the input laser point cloud data, the quality of the laser point cloud data is often directly related to the training effect of the deep neural network model, and the training of the neural network model by using low-quality point cloud data can greatly reduce the operation effect of the unmanned system and even lead to the fact that the whole system cannot work normally. When training the unmanned system based on the deep learning or reinforcement learning algorithm, a large amount of high-quality three-dimensional point cloud data is needed, but the existing point cloud data acquisition work has the problems of wide acquisition range, high labor cost, large data post-processing difficulty, large workload and the like. In addition, processing and transmitting huge amounts of data in real time puts very high demands on the back-end information processing algorithm and the supporting hardware. The prices of commercial multi-line laser radars, high-speed data processing transmission platforms and other devices on the market at present are generally high, so that the technologies such as environment sensing, path decision and the like based on the multi-line laser radars are difficult to realize through lower cost, and the development and popularization of unmanned technologies are limited to a certain extent. Based on the problems, the method and the system for generating the three-dimensional point cloud data with large information quantity, low redundancy and wide perception range by low cost are researched, and have very important theoretical and practical significance for realizing the automatic and efficient construction of the three-dimensional point cloud data with high information density and low data quantity. Disclosure of Invention The invention provides a three-dimensional point cloud data processing method and system for solving the problems of large data size, high cost, large implementation difficulty and the like in the environment sensing technology based on a multi-line laser radar. The method utilizes the laser radar with single line or few lines to cooperate with the high-precision IMU (Inertia Measurement Unit, inertial sensor) and GPS data to generate the large-range three-dimensional point cloud data, and can divide the large-range point cloud data into a plurality of small-size point cloud images according to different requirements and a certain time or space interval. Meanwhile, the three-dimensional point cloud image is subjected to data cleaning according to a certain standard, invalid data are removed, data are outputted in a standardized mode, the information content of data accommodation is further improved, the utilization efficiency of the data is improved, and time cost and hardware cost are saved. In a first aspect, the present invention provides a three-dimensional point cloud data processing method; A three-dimensional point cloud data processing method, comprising: acquiring three-dimensional point cloud data acquired at each moment of the vehicle-mounted laser radar, and acquiring real-time pose of the vehicle-mounted laser radar and real-time coordinates of a vehicle; constructing an initial starting point Yun Zhen list with equidistant intervals or time intervals according to the point cloud frame list to be processed; Constructing an ending point cloud frame list according to the initial starting point cloud frame list and the point cloud frame