Search

CN-122023653-A - Rapid rainy day laser radar point cloud generation method

CN122023653ACN 122023653 ACN122023653 ACN 122023653ACN-122023653-A

Abstract

The invention discloses a rapid rainy day laser radar point cloud generation method which comprises the following steps of S1, deploying a laser radar in a real rainy day environment, collecting original point cloud data containing raindrops, S2, collecting sunny day laser radar point cloud data, S3, constructing an elongated prism area around an axis in a three-dimensional space by taking a corresponding laser beam emission direction of each effective point in the sunny day point cloud as the axis, and S4, calculating the quantity of raindrops with different diameters in unit volume of air under the current rainfall rate according to the rainfall rate specified by a user by utilizing a raindrop particle size distribution function, and normalizing the quantity to be used as the conditional probability that the raindrops with specific diameters are hit by a laser beam in the subarea. According to the invention, starting from the data acquired from the real scene of a sunny day, the data of the rainy day is generated on the basis, so that the automatic driving algorithm for generating the data training can be directly applied to the actual scene.

Inventors

  • LIU YE

Assignees

  • 上海应用技术大学

Dates

Publication Date
20260512
Application Date
20260129

Claims (9)

  1. 1. A rapid rainy day laser radar point cloud generation method is characterized by comprising the following steps: S1, deploying a laser radar in a real rainy day environment, collecting original point cloud data containing raindrops, and calculating the space distribution probability of the raindrops; S2, acquiring fine day laser radar point cloud data; S3, constructing an elongated prism area around the axis in a three-dimensional space by taking the corresponding laser beam emission direction as the axis for each effective point in the sunny point cloud, wherein the cross section side of the area is longer than the spot diameter of the laser beam at the distance; S4, calculating the quantity of raindrops with different diameters in unit volume of air under the current rainfall rate by utilizing a raindrop particle size distribution function according to any rainfall rate specified by a user, and normalizing the quantity to be used as the conditional probability that the raindrops with specific diameters are hit by a laser beam in the subarea; S5, determining upper and lower limits of effective detection distances according to technical specifications of laser radar equipment, randomly sampling a sub-region index corresponding to a potential raindrop position in a limited slender prism region according to a calculated raindrop space distribution probability model, randomly selecting a raindrop diameter in a raindrop diameter effective range according to a particle size distribution model, and calculating the prior probability of the raindrop at the position and the conditional probability of the selected diameter at the position, wherein the product of the prior probability and the conditional probability forms the joint probability of the raindrop effectively detected by the laser radar; s6, for each laser beam path, if the calculated joint probability is greater than a preset probability threshold, judging that the laser beam interacts with raindrops in the propagation process; S7, calculating the echo intensity of the newly added raindrop reflection point by adopting an intensity attenuation model; s8, for an original sunny point which is not replaced by a raindrop, performing raindrop intensity correction operation; S9, packaging the processing flow into a computing unit capable of being executed in parallel, and synchronously implementing raindrop insertion, point rejection and intensity recalculation operation on all laser beam paths in the point cloud; and S10, outputting complete point cloud data subjected to rainy day effect simulation.
  2. 2. The rapid rainy day laser radar point cloud generating method according to claim 1, wherein in the step S1, in a rainy day test scene, driving scene data are collected for a specific laser radar, and then a probability function of raindrop distribution is obtained through statistics for raindrop position and intensity information in the laser radar data.
  3. 3. The method for generating a rapid rainy day laser radar point cloud according to claim 1, wherein the step S2 specifically includes collecting fine day test scene data by using a laser radar, or collecting fine day scene data in public data sets, and obtaining a theoretical laser beam emission angle corresponding to the laser radar according to the type of the laser radar used for data collection.
  4. 4. The method of claim 1, wherein the step S3 includes traversing points in the clear sky data of the laser radar, and for each point, corresponding to a laser beam emitted by the laser radar, and constructing a cuboid with minimum length and width along the laser beam in three dimensions starting from the laser radar, wherein the laser beam is high and the laser beam is located at the center of the cuboid.
  5. 5. The method of claim 1, wherein the step S4 specifically includes calculating, for each small cuboid, a parameter value of the formula using a gamma distribution function of ullrich or an exponential distribution function of Marshall-Palmer for a given arbitrary rainfall rate.
  6. 6. A rapid rainy day lidar point cloud generation method according to claim 5, wherein for a given rainy droplet diameter, the ratio of the number of raindrops in a unit cube to the total number of raindrops is used as the conditional probability that a raindrop of a given diameter is hit by a laser beam.
  7. 7. The method for generating a rapid rainy day laser radar point cloud according to claim 5, wherein the step S5 specifically includes determining the total number of small rectangular solids as N according to the maximum and minimum detection distances given in the laser radar manual, and randomly taking a value in the interval [1, N ] according to the probability function of raindrop distribution, and defining the value as the raindrop position O.
  8. 8. A rapid rainy day laser radar point cloud generating method as claimed in claim 7, characterized in that a value is also randomly taken between the minimum and maximum values of the raindrops, which value is defined as the diameter D of the raindrops, the ratio between the position O of the raindrops and the total number N is denoted as the probability P (O) that the raindrops are hit at the position O, and the ratio of the number of raindrops with the diameter D in the unit cube to the total number of raindrops is denoted as the conditional probability P (d|o) that the raindrops have the diameter D in the presence of the position O.
  9. 9. A rapid rainy day lidar point cloud generation method according to claim 8, wherein P (d|o) P (O) is calculated for each laser beam while setting the probability threshold according to an empirical value.

Description

Rapid rainy day laser radar point cloud generation method Technical Field The invention relates to the technical field of automatic driving, in particular to a rapid laser radar point cloud generation method in rainy days. Background Autopilot technology is currently in the advanced assisted drive technology stage. To drive the development of autopilot technology, testing and verification of algorithms is essential in a wide variety of driving scenarios. At present, laser radar data of a rainy day environment only appear in a small quantity in part of public data sets, and even if a vehicle enterprise has relevant data acquisition, all driving scenes cannot be covered. During the running of the vehicle, the human driver is concerned about the surrounding of the vehicle, whether there are other vehicles, pedestrians, etc. For the sensor lidar of an automatic driving automobile, three-dimensional information of the surrounding environment is constructed by transmitting and receiving laser beams during the running of the automobile. While not as detailed as can be seen by the naked eye, the general contour of the surrounding objects is available. In addition, for the laser radar data in rainy days, the laser radar is used for directly acquiring data in corresponding driving scenes when rainfall occurs. However, in the automatic driving technology, various scenes need to be tested, and the laser radar is directly used for acquisition, and all rainy day scenes cannot be covered, so that the main technology is to construct a virtual test scene by using a simulator, generate raindrops in the scene, further emit laser beams from the simulated laser radar, and generate simulated rainy day laser radar data. The prior art mainly relies on simulators, which utilize physical characteristics such as reflection, diffraction and the like of electromagnetic waves by means of computer graphics, generate driving scenes similar to real scenes by means of a game engine such as a illusion engine 4 only by means of software, simulate the generated scenes when the computing resources are particularly sufficient, and are not easy to visually see the false of the scenes by human eyes. The scenes in the game are all realized by simulators. The closest technical scheme of the invention is PCSim issued by Shanghai artificial intelligence laboratory, the core idea is that using simulator CARLA, automatic driving test scenes of roads, buildings, vehicles and the like with textures similar to the surface textures of real objects are generated, raindrops splashed by road surfaces after wheels drive through rain are generated in the scenes, then laser radar is simulated in virtual scenes to emit laser beams, when the intensity of the reflected laser beams exceeds a given threshold, the reflection point is considered to be effective, and three-dimensional position information of the reflection point is read and written into laser radar data. On the other hand, PCSim uses neural networks to further modify the intensity information in the lidar data so that the value is closer to the value actually acquired. The prior art relies on simulator realization, is completely software-level realization, and the generated data is inconsistent with the real data distribution, thereby affecting the accuracy of an automatic driving algorithm. In particular, the test scene generated by the game engine, while realistic, is only human visual. Meanwhile, the simulation scene can not be compared with a scene in the real world, especially the actual reflectivity of surrounding vehicles, the actual texture of roads and the like, so that the data collected by the laser radar in the simulation scene is not true for an automatic driving algorithm. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a rapid rainy day laser radar point cloud generation method, which is used for generating corresponding rainy day environment data by utilizing the probability and the characteristic of actively emitting laser beams by a laser radar on the basis of acquired sunny day data and solving the problem that the rainy day laser radar data cannot be fully covered. To achieve the above objects and other advantages, the present invention provides a rapid method for generating a laser radar point cloud in rainy days, comprising: S1, deploying a laser radar in a real rainy day environment, collecting original point cloud data containing raindrops, and calculating the space distribution probability of the raindrops; S2, acquiring fine day laser radar point cloud data; S3, constructing an elongated prism area around the axis in a three-dimensional space by taking the corresponding laser beam emission direction as the axis for each effective point in the sunny point cloud, wherein the cross section side of the area is longer than the spot diameter of the laser beam at the distance; S4, calculating the quantity of raindrops with differen