CN116524472B - Obstacle detection method, device, storage medium and equipment - Google Patents
Obstacle detection method, device, storage medium and equipment Download PDFInfo
- Publication number
- CN116524472B CN116524472B CN202310789686.6A CN202310789686A CN116524472B CN 116524472 B CN116524472 B CN 116524472B CN 202310789686 A CN202310789686 A CN 202310789686A CN 116524472 B CN116524472 B CN 116524472B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- grid
- distance
- initial
- height
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 68
- 230000011218 segmentation Effects 0.000 claims abstract description 44
- 238000001914 filtration Methods 0.000 claims abstract description 24
- 238000012216 screening Methods 0.000 claims abstract description 16
- 238000004891 communication Methods 0.000 claims description 127
- 239000011159 matrix material Substances 0.000 claims description 40
- 230000008569 process Effects 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 11
- 238000009499 grossing Methods 0.000 claims description 11
- 238000012952 Resampling Methods 0.000 claims description 5
- 238000000638 solvent extraction Methods 0.000 claims 1
- 230000004927 fusion Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 20
- 239000004973 liquid crystal related substance Substances 0.000 description 12
- 230000006870 function Effects 0.000 description 7
- 230000003595 spectral effect Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 239000000463 material Substances 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000011521 glass Substances 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 239000010426 asphalt Substances 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/64—Three-dimensional objects
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The application provides an obstacle detection method, a device, a storage medium and equipment, wherein in the method, an original point cloud is intercepted within a preset height range to obtain an initial ROI point cloud, a two-dimensional grid network is constructed by utilizing the distribution condition of the initial ROI point cloud in space, plane fitting is carried out on the point cloud in each grid to obtain grid planes, screening is carried out on each grid plane according to the included angle between each grid plane and the Z axis of a laser radar, point clouds are filtered according to Euclidean distance from the point clouds to the screened grid planes to obtain initial point clouds to be detected, and clustering segmentation is carried out on the initial point clouds according to Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among fusion point clouds to obtain target point clouds. Therefore, by comprehensively utilizing the characteristics that non-ground obstacles such as low obstacles and slow-slope green belts are different from the ground, the detection precision of targets is improved, and the detection efficiency is improved through operations such as interception, screening and filtering.
Description
Technical Field
The application relates to the technical field of automatic driving, in particular to a method and a device for detecting obstacles, a storage medium and equipment.
Background
The vehicle with intelligent driving function senses the surrounding environment through the vehicle-mounted camera or the laser radar and other sensors in the autonomous driving process, and automatically plans the driving path according to the sensing result. Currently, a deep neural network sensing algorithm based on a camera and a laser radar has the capability of accurately detecting large targets with higher sizes, such as vehicles, pedestrians, trees and the like. But the detection effect of the deep learning perception algorithm is not good enough for small goods scattered on the road surface, glass bottles, tires, trucks and other obstacles with low sizes such as road pits, road edges and the like. Meanwhile, the gentle slope green belt connected with the road surface does not belong to a vehicle running area, and once the vehicle is missed, a great potential safety hazard can be brought to the vehicle.
At present, the mode of detecting low obstacle and slow slope greenbelt in the related art is mainly based on deep learning model to detect, and the detection effect is improved by optimizing the neural network structure. However, this approach relies heavily on training samples, and when the amount of training samples is small, the model cannot be guaranteed to have good detection effect, and moreover, since it is difficult to exhaust all types of short obstacles, the probability of missed detection of untrained obstacles increases.
Disclosure of Invention
The application aims to provide a method, a device, a storage medium and equipment for detecting obstacles, and aims to solve the problems of poor detection effect and easiness in missed detection in a mode of detecting short obstacles and slow-slope green belts in the related art.
In a first aspect, the present application provides a method for detecting an obstacle, including: constructing a two-dimensional grid network by utilizing the distribution condition of the point clouds of the initial region of interest in space, and carrying out plane fitting on the point clouds in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range; screening each grid plane according to an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane; dividing the initial cloud to be detected to obtain point cloud communication blocks, and carrying out clustering division on the point cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks to obtain target point cloud; the target point cloud is used to indicate a non-ground obstacle.
In the implementation process, intercepting the original point cloud in a preset height range to obtain an initial point cloud of interest, constructing a two-dimensional grid network by utilizing the distribution condition of the initial point cloud of interest in space, carrying out plane fitting on the point cloud in each grid to obtain grid planes, screening the grid planes according to the included angles between the grid planes and the Z axis of the laser radar, filtering the point cloud according to the Euclidean distance from the point cloud to the screened grid planes to obtain an initial point cloud to be detected, and clustering and segmenting the point cloud according to the Euclidean distance, the height characteristic distance, the density characteristic distance and the reflection intensity characteristic distance among the fused point clouds to obtain a target point cloud. Therefore, by comprehensively utilizing the characteristics that non-ground obstacles such as low obstacles and slow-slope green belts are different from the ground, the detection precision of targets is improved, and the detection efficiency is improved through operations such as interception, screening and filtering.
Further, in some examples, the screening the grid planes according to an included angle between each grid plane and a coordinate axis vertically upward in a coordinate system where the laser radar is located includes: calculating an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is positioned; if the included angle is smaller than or equal to a preset angle threshold value, reserving the grid plane; and discarding the grid plane if the included angle is larger than a preset angle threshold.
In the implementation process, the included angle between each grid plane and the Z axis of the laser radar is calculated, and the grid planes with included angles larger than a preset angle threshold are filtered, so that abnormal planes are effectively filtered in advance.
Further, in some examples, before filtering to obtain the initial point cloud to be detected according to euclidean distances between each point cloud point in the initial point cloud of interest and the screened grid plane, the method includes: and smoothing the screened grid plane by using a Gaussian smoothing algorithm.
In the implementation process, before the initial cloud of to-be-detected points is filtered, gaussian smoothing is performed on the grid plane, so that false extraction of interference points is reduced.
Further, in some examples, the dividing the initial point cloud to be detected to obtain a point cloud communication block includes: the method comprises the steps of constructing a rough segmentation two-dimensional grid network, distributing the initial point cloud to be detected to each grid of the rough segmentation two-dimensional grid network, and performing rough segmentation on the initial point cloud to be detected in an eight-neighborhood mode to obtain an initial point cloud communication block; the single grid width and height dimensions of the rough-divided two-dimensional grid net are the first dimensions; constructing a finely divided two-dimensional grid network, distributing the initial point cloud communication blocks to grids of the finely divided two-dimensional grid network, and finely dividing the initial point cloud communication blocks in an eight-neighborhood mode to obtain point cloud communication blocks; the single grid width and height dimensions of the finely divided two-dimensional grid net are the second dimensions; the second dimension is smaller than the first dimension.
In the implementation process, the initial point cloud to be detected is roughly segmented to obtain an initial point cloud communication block with a larger area, and then the initial point cloud communication block is finely segmented to obtain a point cloud communication block with a smaller area, so that a good foundation is laid for subsequent clustering segmentation.
Further, in some examples, the euclidean distance, the height feature distance, the density feature distance, and the reflection intensity feature distance between the point cloud communication blocks are obtained based on the following manner: constructing a target two-dimensional grid network, and distributing each point cloud communication block to the target two-dimensional grid network; the single grid width and height dimensions of the target two-dimensional grid net are the third dimensions; the third dimension is smaller than the second dimension; constructing a three-channel feature map according to the point cloud communication blocks in the target two-dimensional grid network, resampling the three-channel feature map, and extracting the height feature, the density feature and the reflection intensity feature of each point cloud point in a single grid; based on the coordinates of the cloud points, euclidean distance between the cloud communication blocks is calculated, and based on the height features, density features and reflection intensity features of the cloud points, the height feature distance, the density feature distance and the reflection intensity feature distance between the cloud communication blocks are calculated respectively.
In the implementation process, a three-channel height-density-reflection intensity characteristic diagram is constructed, and the point cloud height characteristic, the density characteristic and the reflection intensity characteristic are extracted from the characteristic diagram, so that the detection precision is improved.
Further, in some examples, the clustering and segmentation are performed on the cloud communication blocks of each point based on the euclidean distance, the height feature distance, the density feature distance and the reflection intensity feature distance between the cloud communication blocks of each point, to obtain a cloud of target points, including: obtaining weights among the cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the cloud communication blocks; constructing a similarity matrix and a degree matrix based on weights among all the point cloud connected blocks, and calculating to obtain a Laplace matrix based on the similarity matrix and the degree matrix; and normalizing the Laplace matrix, carrying out cluster segmentation on the connected blocks of each point cloud according to the normalized Laplace matrix, and obtaining a target point cloud based on the point cloud clusters obtained by the cluster segmentation.
In the implementation process, a spectral clustering algorithm is used for carrying out more accurate clustering segmentation on the point cloud communication blocks, so that point clouds of a road surface, a gentle slope green belt, a road pit, a road edge and a low obstacle are respectively aggregated into different point cloud clusters, and the target point cloud is obtained.
Further, in some examples, the weights between the cloud connected blocks of points are calculated based on the following formula:
in the method, in the process of the application,is->The point cloud communication block and +.>Weights among the point cloud communication blocks; />Is->The point cloud communication block and +.>Euclidean distance between the point cloud communication blocks; />Is->The point cloud communication block and +.>Height characteristic distance between the point cloud communication blocks; />Is->The point cloud communication block and +.>Density characteristic distance between the point cloud communicating blocks; />Is->The point cloud communication block and +.>Reflection intensity characteristic distances among the point cloud communication blocks; />、/>、/>、/>Weights of euclidean distance, height feature distance, density feature distance and reflection intensity feature distance,；/>、/>、/>、/>the gaussian kernel coefficients for euclidean distance, height feature distance, density feature distance, and reflected intensity feature distance, respectively.
In the implementation process, the weight among the point clouds is obtained rapidly through the formula, and the detection efficiency is improved.
In a second aspect, the present application provides an obstacle detection device, including: the fitting module is used for constructing a two-dimensional grid network by utilizing the distribution condition of the point clouds of the initial region of interest in space, and carrying out plane fitting on the point clouds in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range; the filtering module is used for screening each grid plane according to the included angle between each grid plane and the vertically upward coordinate axis in the coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to the Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane; the segmentation module is used for segmenting the initial cloud of points to be detected to obtain point cloud communication blocks, and clustering segmentation is carried out on the point cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks to obtain target point cloud; the target point cloud is used to indicate a non-ground obstacle.
In a third aspect, the present application provides an electronic device, including: a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method according to any one of the first aspects when the computer program is executed.
In a fourth aspect, the present application provides a computer readable storage medium having instructions stored thereon which, when run on a computer, cause the computer to perform the method according to any of the first aspects.
In a fifth aspect, the present application provides a computer program product which, when run on a computer, causes the computer to perform the method according to any of the first aspects.
Additional features and advantages of the disclosure will be set forth in the description which follows, or in part will be obvious from the description, or may be learned by practice of the disclosure.
In order to make the above objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and should not be considered as limiting the scope, and other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of an obstacle detection method according to an embodiment of the present application;
fig. 2 is a schematic diagram of a workflow of a point cloud detection scheme for a low obstacle and a slow slope green belt according to an embodiment of the present application;
FIG. 3 is a block diagram of an obstacle detecting apparatus according to an embodiment of the present application;
fig. 4 is a block diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures. Meanwhile, in the description of the present application, the terms "first", "second", and the like are used only to distinguish the description, and are not to be construed as indicating or implying relative importance.
As described in the background art, the detection effect is poor and the detection is easy to be missed in the mode of detecting the short obstacle and the slow-slope green belt in the related art. Based on the above, the embodiment of the application provides a new obstacle detection scheme to solve the above problems.
The following describes embodiments of the present application:
as shown in fig. 1, fig. 1 is a flowchart of an obstacle detection method according to an embodiment of the present application, where the method may be applied to an automatic driving assistance system on a vehicle, a controller on a vehicle, such as a whole vehicle controller (Vehicle Control Unit, VCU) or a domain controller (Domain Control Unit, DCU), or the like, or may also be applied to a server that establishes a communication connection with a vehicle. The obstacles mentioned in the method may include low obstacles on the ground, such as stones, tires, scattered small goods, etc., and may also include gentle green belts, low curbs, pits, etc. in close proximity to the ground.
The method comprises the following steps:
in step 101, constructing a two-dimensional grid network by utilizing the distribution condition of the point clouds of the initial region of interest in space, and carrying out plane fitting on the point clouds in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range;
The original point cloud mentioned in this step may refer to point cloud data acquired by the vehicle through the lidar. In order to detect low obstacles on a road surface, a gentle slope green belt connected with the road surface and the like, the scheme of the embodiment intercepts original point clouds within a certain height range, reduces the number of the point clouds, and accordingly improves the real-time performance of an algorithm to a certain extent. Taking a preset height range below-1 m as an example, taking a laser radar as a three-dimensional coordinate system origin (0, 0), and intercepting a Z-axis coordinate downwards atThe point cloud in the range can obtain the point cloud of the initial region of interest (Region Of Interest, ROI).
After the initial ROI point cloud is obtained through interception, a grid is constructed, and plane fitting of the road point cloud is conducted. Specifically, a two-dimensional (2D) grid network is constructed according to the distribution condition of the initial ROI point cloud in space, namely the initial ROI point cloud is projected to a certain plane, and then the projected point cloud is spatially divided by a uniform grid with a fixed size. Thereafter, a plane fit is performed on the point cloud falling within the ith grid, e.g. a plane equation is fitted using a random consensus algorithm (Random Sample Consensus, RANSAC)Thereby obtaining a grid plane corresponding to each grid. Of course, in other embodiments, the grid planes may be obtained using other plane fitting methods.
102, screening each grid plane according to an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane;
the method comprises the following steps: according to the included angle between each grid plane and the vertical upward coordinate axis in the coordinate system of the laser radar, namely the Z axis, filtering out grid planes with too large gradients, and filtering out non-ground points according to Euclidean distance between point clouds and the screened grid planes, so as to obtain initial point clouds to be detected, wherein the filtered initial point clouds to be detected possibly comprise targets such as gentle slope greenbelts, road pits, road edges, short obstacles and the like.
Specifically, in some embodiments, the screening of each grid plane according to the included angle between each grid plane and the vertically upward coordinate axis in the coordinate system where the laser radar is located may include: calculating an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is positioned; if the included angle is smaller than or equal to a preset angle threshold value, reserving the grid plane; and discarding the grid plane if the included angle is larger than a preset angle threshold. That is, a preset angle threshold is set, when the included angle between any one grid plane and the laser radar Z axis is smaller than or equal to the preset angle threshold, the gradient of the grid plane is indicated to meet the requirement, and the grid plane is reserved, otherwise, when the included angle between any one grid plane and the laser radar Z axis is larger than the preset angle threshold, the gradient of the grid plane is indicated to be too large, and the point cloud corresponding to the grid plane is not the ground point cloud in a large probability, so that the grid plane is discarded. Thus, some abnormal planes can be filtered out effectively in advance. The included angle between the grid plane and the Z axis of the laser radar can be calculated based on the following formula:
In the above-mentioned method, the step of,is the included angle between the grid plane and the Z axis of the laser radar; />Is the normal vector of the grid plane; />Is the Z-axis unit vector of the laser radar. In addition, the preset angle threshold value can be set according to the requirements of specific scenes.
Further, in some embodiments, before filtering to obtain the initial point cloud to be detected according to the euclidean distance between each point cloud point in the initial point cloud of interest and the screened grid plane, the method may include: and smoothing the screened grid plane by using a Gaussian smoothing algorithm. That is, before the initial point cloud to be detected is filtered out, the grid plane may be gaussian smoothed, thereby reducing false extraction of the interference points. In the implementation, the coordinates of the center points of each grid are determined to be in sequence in an 8 neighborhood mode，/>，……，/>，/>The grid coordinates of the central position areAfter that, according to Gaussian function->The weight coefficient of each grid is determined as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,，/>the method comprises the steps of carrying out a first treatment on the surface of the By calculation of the weighted sum, it can be obtainedThe plane equation to the grid plane after Gaussian smoothing is:
wherein, the liquid crystal display device comprises a liquid crystal display device,is the plane equation of the ith grid in the 8 neighborhood grids.
When the Euclidean distance from any one point cloud point to the grid plane where the Gaussian smoothing is performed is larger than a preset distance threshold, the probability that the point cloud point is a non-ground point is indicated to be large, and the probability is corresponding to trees, other vehicles, pedestrians and the like, so that the point cloud point is filtered out, otherwise, when the Euclidean distance from any one point cloud point to the grid plane where the point cloud point is positioned is smaller than or equal to the preset distance threshold, the point cloud point is reserved, and thus, after all the point cloud points are traversed, the initial point cloud point to be detected is obtained. Alternatively, the preset distance threshold may be 0.5m. Of course, in other embodiments, the preset distance threshold may be set differently according to the requirements of different scenes. Through the point cloud filtering rule, non-ground point clouds can be effectively filtered, and the detection accuracy of targets such as low obstacles and slow-slope green belts is improved.
Step 103, dividing the initial cloud of points to be detected to obtain point cloud communication blocks, and carrying out clustering division on the point cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks to obtain target point cloud; the target point cloud is used to indicate a non-ground obstacle.
The method comprises the following steps: the initial point cloud to be detected obtained through filtering may comprise targets such as a gentle slope green belt, a road pit, a road edge and a low obstacle, wherein the point cloud reflection intensity of the green belt, a rubber tire and a glass bottle is obviously different from that of a pavement made of asphalt or cement due to different materials and materials, the height characteristics and the point cloud density characteristics of the road edge and the low obstacle are obviously different from those of the pavement, and based on the points, the points cloud is subjected to clustering segmentation by comprehensively utilizing Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point clouds to obtain the target point cloud, so that the targets such as the low obstacle and the gentle slope green belt outside the pavement are detected.
In some embodiments, the segmenting the initial cloud of points to be detected mentioned in this step, to obtain a point cloud communication block may include: the method comprises the steps of constructing a rough segmentation two-dimensional grid network, distributing the initial point cloud to be detected to each grid of the rough segmentation two-dimensional grid network, and performing rough segmentation on the initial point cloud to be detected in an eight-neighborhood mode to obtain an initial point cloud communication block; the single grid width and height dimensions of the rough-divided two-dimensional grid net are the first dimensions; constructing a finely divided two-dimensional grid network, distributing the initial point cloud communication blocks to grids of the finely divided two-dimensional grid network, and finely dividing the initial point cloud communication blocks in an eight-neighborhood mode to obtain point cloud communication blocks; the single grid width and height dimensions of the finely divided two-dimensional grid net are the second dimensions; the second dimension is smaller than the first dimension. That is, for the initial point cloud to be detected, a coarse segmentation 2D grid network is firstly constructed, the initial point cloud to be detected is subjected to coarse segmentation in an 8 neighborhood mode to obtain a point cloud communication block with a larger area, namely an initial point cloud communication block, then a fine segmentation 2D grid network is constructed, and the coarse segmentation initial point cloud communication block is subjected to fine segmentation in an 8 neighborhood mode to obtain a point cloud communication block with a smaller area. Thus, through rough segmentation and fine segmentation, small-area point cloud communication blocks are obtained, and whether merging and splitting are needed or not is determined by a subsequent clustering segmentation algorithm. Alternatively, the first dimension may be 1m and the second dimension may be 0.25m. Of course, in other embodiments, the values of the individual grid width and height dimensions of the coarse-divided 2D grid mesh and the fine-divided 2D grid mesh may also be set differently according to the requirements of different scenes.
After the point cloud communication blocks are obtained, based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks, more accurate clustering segmentation is carried out on the point cloud communication blocks so as to respectively aggregate the point clouds of the road surface, the gentle slope green belt, the road pit, the road edge and the low obstacle into different point cloud clusters, and further accurate detection of the obstacles is achieved. In some embodiments, euclidean distance, height feature distance, density feature distance, and reflection intensity feature distance between each point cloud connectivity blocks may be obtained based on: constructing a target two-dimensional grid network, and distributing each point cloud communication block to the target two-dimensional grid network; the single grid width and height dimensions of the target two-dimensional grid net are the third dimensions; the third dimension is smaller than the second dimension; constructing a three-channel feature map according to the point cloud communication blocks in the target two-dimensional grid network, resampling the three-channel feature map, and extracting the height feature, the density feature and the reflection intensity feature of each point cloud point in a single grid; based on the coordinates of the cloud points, euclidean distance between the cloud communication blocks is calculated, and based on the height features, density features and reflection intensity features of the cloud points, the height feature distance, the density feature distance and the reflection intensity feature distance between the cloud communication blocks are calculated respectively. That is, a target 2D grid network with a single grid with smaller width and height is constructed, after each point cloud communication block is distributed to the target 2D grid network according to coordinates, a three-channel feature map is constructed for each point cloud communication block, wherein the pixel size of the three-channel feature map can be set based on the total number of grids, namely one pixel represents one grid, at the moment, the gray value of the pixel of the first channel of the three-channel feature map is the average value of the Z-axis height accumulation of all point clouds falling in the corresponding grid, the gray value of the pixel of the second channel is the number of the point clouds falling in the unit area of the corresponding grid, and the gray value of the pixel of the third channel is the average value of the reflection intensity accumulation of all point clouds falling in the corresponding grid. In this way, resampling the three-channel feature map can extract point cloud height features, density features, and reflected intensity features within a single grid. The euclidean distance between the cloud communication blocks of each point can be calculated based on the coordinates of the cloud points of the corresponding point, and the height feature distance, the density feature distance and the reflection intensity feature distance between the cloud communication blocks of each point can be calculated based on the height feature, the density feature and the reflection intensity feature between the cloud points of the corresponding point.
Further, the clustering segmentation of the point cloud communication blocks based on the euclidean distance, the height feature distance, the density feature distance and the reflection intensity feature distance between the point cloud communication blocks mentioned in the step may include: obtaining weights among the cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the cloud communication blocks; constructing a similarity matrix and a degree matrix based on weights among all the point cloud connected blocks, and calculating to obtain a Laplace matrix based on the similarity matrix and the degree matrix; and normalizing the Laplace matrix, carrying out cluster segmentation on the connected blocks of each point cloud according to the normalized Laplace matrix, and obtaining a target point cloud based on the point cloud clusters obtained by the cluster segmentation. That is, the point cloud communication blocks can be clustered and segmented more accurately by using a spectral clustering algorithm, so that the point clouds of the road surface, the gentle slope green belt, the road pit, the road edge and the low obstacle are respectively aggregated into different point cloud clusters, and the target point cloud is obtained.
Specifically, the above procedure may be: the Euclidean distance, the height characteristic distance, the density characteristic distance and the reflection intensity characteristic distance between the point cloud communicating blocks are fused to obtain the weight between the point cloud communicating blocks; spectral clustering is based on graph theory, for any point in the graph, its degree is defined as the sum of weights of all edges connected with it, and accordingly, by using the definition of each point degree, a similarity matrix can be formed according to weights among all point cloud connected blocks Sum matrix->The similarity matrix->The j-th value of the i-th row in the row (a) can correspond to the weight between the i-th point cloud communication block and the j-th point cloud communication block, and the degree matrix is ≡>The diagonal matrix only has values on the diagonal, and the element values on the diagonal correspond to the degrees of the ith point of the ith row; according to the similarity matrix->Sum matrix->The Laplace matrix can be obtained>After that, the Laplace matrix is +.>And (3) carrying out standardization, so as to construct a feature vector space, and finally carrying out clustering segmentation on the feature vector space according to an Ncut cut chart or a Ratiocut chart to obtain a plurality of segmented point cloud clusters, namely, realizing aggregation of target point clouds such as gentle slope greenbelts, road pits, road edges, low obstacles and the like. Therefore, the detection precision of targets such as low obstacles, slow-slope green belts and the like is effectively improved.
The weights between the cloud connected blocks of each point can be expressed based on the following formula:
in the method, in the process of the invention,is->The point cloud communication block and +.>Weights among the point cloud communication blocks; />Is->The point cloud communication block and +.>Euclidean distance between the point cloud communication blocks; />Is->The point cloud communication block and +.>Height characteristic distance between the point cloud communication blocks; / >Is->The point cloud communication block and +.>Density characteristic distance between the point cloud communicating blocks; />Is->The point cloud communication block and +.>Reflection intensity characteristic distances among the point cloud communication blocks; />、/>、/>、/>Weights of euclidean distance, height feature distance, density feature distance and reflection intensity feature distance,；/>、/>、/>、/>the gaussian kernel coefficients for euclidean distance, height feature distance, density feature distance, and reflected intensity feature distance, respectively. Optionally, ->，/>，/>，；/>. Through the formula, the weight among the point clouds can be obtained quickly, and the detection efficiency is improved.
According to the embodiment of the application, intercepting an original point cloud in a preset height range to obtain an initial region-of-interest point cloud, constructing a two-dimensional grid network by utilizing the distribution condition of the initial region-of-interest point cloud in space, carrying out plane fitting on the point cloud in each grid to obtain grid planes, screening each grid plane according to the included angle between each grid plane and the Z axis of the laser radar, filtering the point cloud according to the Euclidean distance from the point cloud to the screened grid plane to obtain an initial point cloud to be detected, and clustering and segmenting the point cloud according to the Euclidean distance, the height characteristic distance, the density characteristic distance and the reflection intensity characteristic distance among the fused point clouds to obtain a target point cloud. Therefore, by comprehensively utilizing the characteristics that non-ground obstacles such as low obstacles and slow-slope green belts are different from the ground, the detection precision of targets is improved, and the detection efficiency is improved through operations such as interception, screening and filtering.
For a more detailed description of the solution of the application, a specific embodiment is described below:
the embodiment provides a point cloud detection scheme for low obstacles and a slow-slope green belt. The workflow of this scheme is shown in fig. 2, comprising:
s201, acquiring laser radar point cloud data;
s202, intercepting an initial ROI point cloud;
specifically, the laser radar is taken as the origin (0, 0) of the three-dimensional coordinate system, and the Z-axis coordinate is intercepted downwardsObtaining an initial ROI point cloud by the point cloud in the range;
s203, constructing a grid to perform plane fitting of the road point cloud;
specifically, firstly, constructing a 2D grid network according to the distribution condition of an initial ROI point cloud in space; then, a random consistency algorithm is applied to the point cloud falling in the ith grid to fit a plane equationThe method comprises the steps of carrying out a first treatment on the surface of the Then, the normal vector of the grid plane corresponding to the ith grid is calculated as +.>Filtering the grid plane with the included angle larger than a preset angle threshold value from the included angle of the Z axis of the laser radar; finally, smoothing fitting planes of different grids by using a Gaussian smoothing algorithm to obtain a smoothed pavement point cloud plane;
s204, filtering the point cloud according to the Euclidean distance from the point cloud to the grid plane to obtain an initial point cloud to be detected;
Specifically, the Euclidean distance from each point cloud point in the initial ROI point cloud to the road surface point cloud plane obtained in S203 is traversedOnly the Euclidean distance is reserved->Point cloud points smaller than or equal to 0.5m are obtained through filtering, so that initial point cloud to be detected is obtained;
s205, performing rough segmentation on the initial cloud to be detected to obtain a large-point cloud communication block;
specifically, a coarse segmentation 2D grid network is constructed, and a laser radar is used as a coordinate origin, and grid space ranges in the x and y directions are formedThe single grid width and height dimensions were 1.0m, then the total number of grids was 40000; then, roughly dividing the initial point cloud to be detected obtained in the step S204 in an 8 neighborhood mode to obtain a point cloud communication block with a larger area;
s206, carrying out fine segmentation on the large point cloud communication blocks to obtain small point cloud communication blocks;
specifically, a finely divided 2D grid network is constructed, and a laser radar is used as a coordinate origin, and grid space ranges in the x and y directions are formedThe single grid width and height dimensions were 0.25m, then the total number of grids was 640000; then, the large point cloud communication blocks obtained in the step S205 are finely divided one by one in an 8 neighborhood mode, so that point cloud communication blocks with smaller areas are obtained>Wherein->Represents->The first division of the large point cloud communication block>The small point cloud communication blocks;
S207, constructing a three-channel height-density-reflection intensity characteristic diagram, and extracting point cloud height characteristics, density characteristics and reflection intensity characteristics;
specifically, a 2D grid network is constructed, and a laser radar is taken as a coordinate origin, and grid space ranges in the x and y directions are formedThe single grid width and height dimensions were 0.20m, then the total number of grids was 1000000; constructing a three-channel height-density-reflection intensity characteristic diagram with the pixel size of 1000 x 1000 for the finely divided small point cloud communication blocks, resampling the three-channel characteristic diagram, and extracting point cloud height characteristics, density characteristics and reflection intensity characteristics in a single grid;
the pixel gray value of the 1 st channel of the three-channel characteristic diagramThe average value of the Z-axis height accumulation of all the point clouds falling in the corresponding grid is that:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing the number of point clouds falling within the grid; />The Z-axis height representing the ith point cloud;
the pixel gray value of the 2 nd channel of the three-channel characteristic diagramThe number of point clouds in a unit area of the corresponding grid is:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing the area of a single grid;
the pixel gray value of the 3 rd channel of the three-channel characteristic diagram isThe average value of the summation of the reflection intensities of all the point clouds falling in the corresponding grids is that:
wherein, the liquid crystal display device comprises a liquid crystal display device, Representing the reflection intensity of the ith point cloud;
after the construction of the original height-density-reflection intensity characteristic diagram of the three channels is completed, each channel is normalized, and the gray level value of the image pixel is normalized to 0-255;
s208, constructing a similarity matrix, a degree matrix and a Laplacian matrix;
specifically, 4 features including Euclidean distance, height feature distance, point cloud density feature distance and point cloud reflection intensity distance between small point cloud communication blocks are fused to obtain a similarity matrix between point cloud clustersDegree matrix->And Laplace matrix>The method comprises the steps of carrying out a first treatment on the surface of the Wherein:
euclidean distance between point clouds；
Height feature distance between point clouds；
Density characteristic distance between point clouds；
Reflection intensity characteristic distance between point clouds；
Wherein, the liquid crystal display device comprises a liquid crystal display device,is the ith point coordinate; />Is the j-th point coordinate; />And->Respectively corresponding to the pixel gray values of the i point and the j point on the 1 st channel of the three-channel feature map; accordingly, the +>、、/>、/>The pixel gray values of the i point and the j point on the 2 nd channel and the 3 rd channel of the three-channel feature map are respectively corresponding;
then the weight between the point clouds；
Wherein, the weight of 4 feature distances is distributed as follows:，/>，/>，the method comprises the steps of carrying out a first treatment on the surface of the In Gaussian kernel function, < >>；
S209, obtaining an obstacle point cloud based on Ncut spectral cluster segmentation;
specifically, the laplace matrix L is normalized: The method comprises the steps of carrying out a first treatment on the surface of the Then, carrying out more accurate clustering segmentation on the small point cloud communication blocks according to the flow of an Ncut spectral clustering algorithm, wherein the maximum iteration number of Ncut is set to be 100 in order to avoid the influence on the timeliness of the algorithm caused by overlong time consumption of the large number of small point cloud communication blocks in the Ncut spectral clustering process;
after NCut spectrum clustering treatment, point clouds of a road surface, a gentle slope green belt, a road pit, a road edge and a low obstacle can be respectively polymerized into different point cloud clusters, so that detection of the low obstacle and the gentle slope green belt outside the road surface is realized.
In the scheme of the embodiment, ROI interception is carried out on the original point cloud within a certain height range, road surface fitting is carried out on the intercepted ROI point cloud, and initial point cloud to be detected, which is within a height range of 0.5m from a road surface, is selected, so that the number of point clouds is reduced, and the algorithm time efficiency is improved; the characteristics of different materials and different point cloud reflection intensities of the materials are fully utilized, meanwhile, the characteristics of the height characteristics and the point cloud density characteristics of the road edges and the short obstacles are considered to be different from the characteristics of the road surface, a three-channel height-density-reflection intensity characteristic diagram is constructed, and the point cloud height characteristics, the density characteristics and the reflection intensity characteristics are extracted from the characteristic diagram, so that the detection precision is improved; and the Euclidean distance characteristic, the point cloud height distance characteristic, the point cloud density characteristic and the point cloud reflection intensity characteristic are comprehensively utilized to construct a similarity matrix among the point cloud clusters, and then the Ncut spectral clustering algorithm is utilized to aggregate target point clouds such as the gentle slope greenbelt, the road pits, the road edges and the low obstacle, so that the detection of the low obstacle and the gentle slope greenbelt is effectively realized, and the running safety of intelligent driving vehicles is ensured.
Corresponding to the embodiments of the foregoing method, the present application further provides an embodiment of the obstacle detection device and a terminal to which the obstacle detection device is applied:
as shown in fig. 3, fig. 3 is a block diagram of an obstacle detecting apparatus according to an embodiment of the present application, where the apparatus includes:
the fitting module 31 is configured to construct a two-dimensional grid network 32 by using the distribution situation of the point cloud of the initial region of interest in space, and perform plane fitting on the point cloud in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range;
the filtering module is used for screening each grid plane according to the included angle between each grid plane and the vertically upward coordinate axis in the coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to the Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane;
the segmentation module 33 is configured to segment the initial cloud to be detected to obtain point cloud communication blocks, and cluster-segment the point cloud communication blocks based on euclidean distance, height feature distance, density feature distance and reflection intensity feature distance between the point cloud communication blocks to obtain a target point cloud; the target point cloud is used to indicate a non-ground obstacle.
The implementation process of the functions and roles of each module in the above device is specifically shown in the implementation process of the corresponding steps in the above method, and will not be described herein again.
The application further provides an electronic device, please refer to fig. 4, and fig. 4 is a block diagram of an electronic device according to an embodiment of the application. The electronic device may include a processor 410, a communication interface 420, a memory 430, and at least one communication bus 440. Wherein the communication bus 440 is used to enable direct connection communication of these components. The communication interface 420 of the electronic device in the embodiment of the present application is used for performing signaling or data communication with other node devices. The processor 410 may be an integrated circuit chip with signal processing capabilities.
The processor 410 may be a general-purpose processor, including a central processing unit (CPU, centralProcessingUnit), a network processor (NP, networkProcessor), etc.; but may also be a Digital Signal Processor (DSP), application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components. The disclosed methods, steps, and logic blocks in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor 410 may be any conventional processor or the like.
The Memory 430 may be, but is not limited to, random access Memory (RAM, randomAccessMemory), read Only Memory (ROM), programmable Read Only Memory (PROM, programmable Read-Only Memory), erasable Read Only Memory (EPROM, erasable Programmable Read-Only Memory), electrically erasable Read Only Memory (EEPROM, electric Erasable Programmable Read-Only Memory), and the like. The memory 430 has stored therein computer readable instructions which, when executed by the processor 410, can cause the electronic device to perform the steps described above in relation to the method embodiment of fig. 1.
Optionally, the electronic device may further include a storage controller, an input-output unit.
The memory 430, the memory controller, the processor 410, the peripheral interface, and the input/output unit are electrically connected directly or indirectly to each other to realize data transmission or interaction. For example, the elements may be electrically coupled to each other via one or more communication buses 440. The processor 410 is configured to execute executable modules stored in the memory 430, such as software functional modules or computer programs included in the electronic device.
The input-output unit is used for providing the user with the creation task and creating the starting selectable period or the preset execution time for the task so as to realize the interaction between the user and the server. The input/output unit may be, but is not limited to, a mouse, a keyboard, and the like.
It will be appreciated that the configuration shown in fig. 4 is merely illustrative, and that the electronic device may also include more or fewer components than shown in fig. 4, or have a different configuration than shown in fig. 4. The components shown in fig. 4 may be implemented in hardware, software, or a combination thereof.
The embodiment of the application also provides a storage medium, wherein the storage medium stores instructions, and when the instructions run on a computer, the computer program is executed by a processor to implement the method described in the method embodiment, so that repetition is avoided, and no further description is provided here.
The application also provides a computer program product which, when run on a computer, causes the computer to perform the method according to the method embodiments.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus embodiments described above are merely illustrative, for example, of the flowcharts and block diagrams in the figures that illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, functional modules in the embodiments of the present application may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and variations will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application. It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Claims (9)
1. An obstacle detection method, comprising:
constructing a two-dimensional grid network by utilizing the distribution condition of the point clouds of the initial region of interest in space, and carrying out plane fitting on the point clouds in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range;
screening each grid plane according to an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane;
dividing the initial cloud to be detected to obtain point cloud communication blocks, and carrying out clustering division on the point cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks to obtain target point cloud; the target point cloud is used for indicating a non-ground obstacle;
the method for clustering and dividing the cloud communication blocks of each point based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance between the cloud communication blocks of each point to obtain a target point cloud comprises the following steps:
Obtaining weights among the cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the cloud communication blocks;
constructing a similarity matrix and a degree matrix based on weights among all the point cloud connected blocks, and calculating to obtain a Laplace matrix based on the similarity matrix and the degree matrix;
and normalizing the Laplace matrix, carrying out cluster segmentation on the connected blocks of each point cloud according to the normalized Laplace matrix, and obtaining a target point cloud based on the point cloud clusters obtained by the cluster segmentation.
2. The method of claim 1, wherein the screening each grid plane according to an angle between each grid plane and a coordinate axis vertically upward in a coordinate system where the lidar is located comprises:
calculating an included angle between each grid plane and a coordinate axis which is vertically upwards in a coordinate system where the laser radar is positioned;
if the included angle is smaller than or equal to a preset angle threshold value, reserving the grid plane;
and discarding the grid plane if the included angle is larger than a preset angle threshold.
3. The method according to claim 1, wherein before filtering to obtain the initial point cloud to be detected according to euclidean distances from each point cloud point in the initial point cloud of interest to the grid plane after screening, the method comprises:
And smoothing the screened grid plane by using a Gaussian smoothing algorithm.
4. The method of claim 1, wherein the partitioning the initial point cloud to be detected to obtain a point cloud communication block includes:
the method comprises the steps of constructing a rough segmentation two-dimensional grid network, distributing the initial point cloud to be detected to each grid of the rough segmentation two-dimensional grid network, and performing rough segmentation on the initial point cloud to be detected in an eight-neighborhood mode to obtain an initial point cloud communication block; the single grid width and height dimensions of the rough-divided two-dimensional grid net are the first dimensions;
constructing a finely divided two-dimensional grid network, distributing the initial point cloud communication blocks to grids of the finely divided two-dimensional grid network, and finely dividing the initial point cloud communication blocks in an eight-neighborhood mode to obtain point cloud communication blocks; the single grid width and height dimensions of the finely divided two-dimensional grid net are the second dimensions; the second dimension is smaller than the first dimension.
5. The method of claim 4, wherein the euclidean distance, the height feature distance, the density feature distance, and the reflection intensity feature distance between the cloud communication blocks of each point are obtained based on:
Constructing a target two-dimensional grid network, and distributing each point cloud communication block to the target two-dimensional grid network; the single grid width and height dimensions of the target two-dimensional grid net are the third dimensions; the third dimension is smaller than the second dimension;
constructing a three-channel feature map according to the point cloud communication blocks in the target two-dimensional grid network, resampling the three-channel feature map, and extracting the height feature, the density feature and the reflection intensity feature of each point cloud point in a single grid;
based on the coordinates of the cloud points, euclidean distance between the cloud communication blocks is calculated, and based on the height features, density features and reflection intensity features of the cloud points, the height feature distance, the density feature distance and the reflection intensity feature distance between the cloud communication blocks are calculated respectively.
6. The method of claim 5, wherein the weights between the cloud connected blocks are calculated based on the following formula:
in the method, in the process of the invention,is->The point cloud communication block and +.>Weights among the point cloud communication blocks; />Is->The point cloud communication block and +.>Euclidean distance between the point cloud communication blocks; />Is->The point cloud communication block and +.>Height characteristic distance between the point cloud communication blocks; / >Is->The point cloud communication block and +.>Density features between individual point cloud connected blocksA distance; />Is->The point cloud communication block and +.>Reflection intensity characteristic distances among the point cloud communication blocks; />、/>、/>、/>Weights of Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance respectively, +.>；/>、、/>、/>The gaussian kernel coefficients for euclidean distance, height feature distance, density feature distance, and reflected intensity feature distance, respectively.
7. An obstacle detecting apparatus, comprising:
the fitting module is used for constructing a two-dimensional grid network by utilizing the distribution condition of the point clouds of the initial region of interest in space, and carrying out plane fitting on the point clouds in each grid to obtain a grid plane; the initial point cloud of the region of interest is obtained by intercepting an original point cloud within a preset height range;
the filtering module is used for screening each grid plane according to the included angle between each grid plane and the vertically upward coordinate axis in the coordinate system where the laser radar is located, and filtering to obtain an initial point cloud to be detected according to the Euclidean distance from each point cloud point in the initial point cloud of the region of interest to the screened grid plane;
the segmentation module is used for segmenting the initial cloud of points to be detected to obtain point cloud communication blocks, and clustering segmentation is carried out on the point cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the point cloud communication blocks to obtain target point cloud; the target point cloud is used for indicating a non-ground obstacle;
The segmentation module is specifically configured to:
obtaining weights among the cloud communication blocks based on Euclidean distance, height characteristic distance, density characteristic distance and reflection intensity characteristic distance among the cloud communication blocks;
constructing a similarity matrix and a degree matrix based on weights among all the point cloud connected blocks, and calculating to obtain a Laplace matrix based on the similarity matrix and the degree matrix;
and normalizing the Laplace matrix, carrying out cluster segmentation on the connected blocks of each point cloud according to the normalized Laplace matrix, and obtaining a target point cloud based on the point cloud clusters obtained by the cluster segmentation.
8. A computer readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, implements the method according to any of claims 1 to 6.
9. An electronic device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 6 when the computer program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310789686.6A CN116524472B (en) | 2023-06-30 | 2023-06-30 | Obstacle detection method, device, storage medium and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310789686.6A CN116524472B (en) | 2023-06-30 | 2023-06-30 | Obstacle detection method, device, storage medium and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116524472A CN116524472A (en) | 2023-08-01 |
| CN116524472B true CN116524472B (en) | 2023-09-22 |
Family
ID=87397935
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310789686.6A Active CN116524472B (en) | 2023-06-30 | 2023-06-30 | Obstacle detection method, device, storage medium and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116524472B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107719361A (en) * | 2017-10-10 | 2018-02-23 | 深圳市豪恩汽车电子装备股份有限公司 | Automatic parking householder method and system based on image vision |
| WO2021097618A1 (en) * | 2019-11-18 | 2021-05-27 | 深圳市大疆创新科技有限公司 | Point cloud segmentation method and system, and computer storage medium |
| CN113176585A (en) * | 2021-04-14 | 2021-07-27 | 浙江工业大学 | Three-dimensional laser radar-based road surface anomaly detection method |
| CN114495026A (en) * | 2022-01-07 | 2022-05-13 | 武汉市虎联智能科技有限公司 | Laser radar identification method and device, electronic equipment and storage medium |
| WO2022099620A1 (en) * | 2020-11-13 | 2022-05-19 | 深圳市大疆创新科技有限公司 | Three-dimensional point cloud segmentation method and apparatus, and mobile platform |
| CN116203587A (en) * | 2022-11-30 | 2023-06-02 | 吉林大学 | Pothole detection system and method for unmanned vehicle |
| CN116206286A (en) * | 2023-04-21 | 2023-06-02 | 北京四维图新科技股份有限公司 | Obstacle detection method, device, equipment and medium under high-speed road condition |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8332134B2 (en) * | 2008-04-24 | 2012-12-11 | GM Global Technology Operations LLC | Three-dimensional LIDAR-based clear path detection |
-
2023
- 2023-06-30 CN CN202310789686.6A patent/CN116524472B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107719361A (en) * | 2017-10-10 | 2018-02-23 | 深圳市豪恩汽车电子装备股份有限公司 | Automatic parking householder method and system based on image vision |
| WO2021097618A1 (en) * | 2019-11-18 | 2021-05-27 | 深圳市大疆创新科技有限公司 | Point cloud segmentation method and system, and computer storage medium |
| WO2022099620A1 (en) * | 2020-11-13 | 2022-05-19 | 深圳市大疆创新科技有限公司 | Three-dimensional point cloud segmentation method and apparatus, and mobile platform |
| CN113176585A (en) * | 2021-04-14 | 2021-07-27 | 浙江工业大学 | Three-dimensional laser radar-based road surface anomaly detection method |
| CN114495026A (en) * | 2022-01-07 | 2022-05-13 | 武汉市虎联智能科技有限公司 | Laser radar identification method and device, electronic equipment and storage medium |
| CN116203587A (en) * | 2022-11-30 | 2023-06-02 | 吉林大学 | Pothole detection system and method for unmanned vehicle |
| CN116206286A (en) * | 2023-04-21 | 2023-06-02 | 北京四维图新科技股份有限公司 | Obstacle detection method, device, equipment and medium under high-speed road condition |
Non-Patent Citations (2)
| Title |
|---|
| 基于多线激光雷达的非结构化道路感知技术研究;李宁等;《车辆与动力技术》;第8页至第14页 * |
| 稀疏点云道路分割与低矮路障检测的自适应方法;罗俊奇等;《遥感信息》;第108页至第115页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116524472A (en) | 2023-08-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cheng et al. | Extraction and classification of road markings using mobile laser scanning point clouds | |
| CN107045629B (en) | Multi-lane line detection method | |
| US10915779B2 (en) | Method for extracting uniform features from point cloud and system therefor | |
| Yadav et al. | Extraction of road surface from mobile LiDAR data of complex road environment | |
| Wang et al. | Real-time vehicle detection algorithm based on vision and lidar point cloud fusion | |
| US10430659B2 (en) | Method and apparatus for urban road recognition based on laser point cloud, storage medium, and device | |
| Negru et al. | Image based fog detection and visibility estimation for driving assistance systems | |
| CN110866449A (en) | Method and device for identifying target object in road | |
| CN110502983B (en) | Method and device for detecting obstacles in expressway and computer equipment | |
| Ding et al. | Fast lane detection based on bird’s eye view and improved random sample consensus algorithm | |
| CN116520353B (en) | Ground detection method, device, storage medium and equipment based on laser point cloud | |
| Zhang | Rapid inspection of pavement markings using mobile laser scanning point clouds | |
| You et al. | Lane detection algorithm for night-time digital image based on distribution feature of boundary pixels | |
| Yadav et al. | Road surface detection from mobile lidar data | |
| CN113917487A (en) | Laser radar-based method for detecting road edge and drivable area of closed road | |
| CN114882316A (en) | Target detection model training method, target detection method and device | |
| Bao et al. | Unpaved road detection based on spatial fuzzy clustering algorithm | |
| Liu et al. | Vehicle detection from aerial color imagery and airborne LiDAR data | |
| Chen | Road vehicle recognition algorithm in safety assistant driving based on artificial intelligence | |
| CN116524472B (en) | Obstacle detection method, device, storage medium and equipment | |
| CN107578037B (en) | Lane line detection method based on analog property estimation | |
| Qing et al. | Localization and tracking of same color vehicle under occlusion problem | |
| CN116681932A (en) | Object identification method and device, electronic equipment and storage medium | |
| US11555928B2 (en) | Three-dimensional object detection with ground removal intelligence | |
| CN115376106A (en) | Vehicle type identification method, device, equipment and medium based on radar map |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |