CN115359453A - Road surface obstacle detection method and system based on three-dimensional point cloud - Google Patents
Road surface obstacle detection method and system based on three-dimensional point cloud Download PDFInfo
- Publication number
- CN115359453A CN115359453A CN202210965243.3A CN202210965243A CN115359453A CN 115359453 A CN115359453 A CN 115359453A CN 202210965243 A CN202210965243 A CN 202210965243A CN 115359453 A CN115359453 A CN 115359453A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- cloud data
- cluster
- obstacle
- coordinate
- 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.)
- Pending
Links
Images
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
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06V10/30—Noise filtering
-
- 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
Abstract
The invention relates to a method and a system for detecting a road surface obstacle based on three-dimensional point cloud, which are characterized in that the method and the system carry out density-based spatial clustering on road surface three-dimensional point cloud data acquired by a four-eye camera to obtain a plurality of clusters; aiming at each cluster, calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under the three-dimensional coordinate axis to obtain the width of the barrier; calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle; and calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle. The method is directly realized through three-dimensional point cloud data generated by a four-eye camera, so that a perception module of the unmanned system obtains depth information, and the depth information can be in one-to-one correspondence with physical coordinates of a three-dimensional space. The method is applied to the perception of the road surface obstacle in the unmanned system, so that the existence of the obstacle can be detected, and the size of the obstacle and the distance between the obstacle and the unmanned vehicle can also be detected.
Description
Technical Field
The invention relates to the field of three-dimensional point cloud processing technology and computer vision, in particular to a road surface obstacle detection method and system based on three-dimensional point cloud.
Background
The improvement of the unmanned technology has remarkable research value and social value for the whole automobile industry, traffic accidents caused by drunk driving, fatigue driving and the like are avoided, so that the safety is improved, and traffic jam caused by overtaking, jamming and the like is reduced. A complete unmanned system mainly comprises three modules of perception, decision and control, wherein the perception module is the basis of the function realization of other modules, and the target detection is an important task of the perception module and is used for acquiring the position and category information (such as vehicles, pedestrians, roadblocks and the like) of an object in the space, so that the system has an important guiding function on subsequent path planning, collision avoidance and the like.
Object detection technology is one of three major tasks in the field of computer vision, which aims to obtain location and category information of an object. The detection of the road surface obstacle target is to further limit the range on the basis of the target detection, firstly setting a scene as a road on which a vehicle runs, and secondly aiming at the target such as pedestrians, vehicles, roadblocks and the like on the road surface. The detection of the road obstacle target requires finding out all obstacles in the acquired data, and positioning and classifying the obstacles.
In the conventional road surface obstacle detection method, if only two-dimensional image data is subjected to target detection, although a target in an image can be accurately identified, mapping to a three-dimensional space is lack of depth information indispensable to a perception module, namely, an unmanned vehicle perceives the existence of an obstacle, but cannot obtain the distance between the obstacle and the unmanned vehicle and the size of the obstacle, so that due help cannot be provided for later path planning and obstacle avoidance. Accordingly, there is a need for a method and system for detecting a road obstacle, so as to obtain the distance between the obstacle and itself and the size of the obstacle.
Disclosure of Invention
The invention aims to provide a road surface obstacle detection method and system based on three-dimensional point cloud, which are applied to the perception of road surface obstacles in an unmanned system, can detect the existence of the obstacles, and can detect the size of the obstacles and the distance between the obstacles and an unmanned vehicle, thereby greatly reducing the traffic accidents of the unmanned vehicle and improving the safety.
In order to achieve the purpose, the invention provides the following scheme:
a road surface obstacle detection method based on three-dimensional point cloud comprises the following steps:
carrying out density-based spatial clustering on the target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of said clusters is considered an obstacle;
aiming at each cluster, calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under a three-dimensional coordinate system to obtain the width of the obstacle; the three-dimensional coordinate system is a coordinate system established by taking a vehicle as an origin and taking the advancing direction of the vehicle as a z-axis;
calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle;
and calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the vehicle and the obstacle.
As an optional embodiment, before performing the density-based spatial clustering on the target point cloud data, the method further comprises: preprocessing three-dimensional point cloud data acquired by a four-eye camera:
extracting an interesting region of the three-dimensional point cloud data;
down-sampling the extracted point cloud data of the region of interest to obtain down-sampled point cloud data;
and carrying out outlier filtering on the down-sampling point cloud data.
As an optional implementation manner, the downsampling the point cloud data of the extracted region of interest to obtain downsampled point cloud data specifically includes:
dividing the point cloud data of the region of interest into a plurality of cubic grids;
selecting a non-empty cubic grid in the cubic grids;
calculating the centroid of all points in each non-empty cubic grid to obtain a centroid point;
and replacing all points in the non-empty cubic grid with the centroid points to obtain down-sampling points corresponding to the non-empty cubic grid.
As an optional implementation manner, the outlier filtering of the down-sampled point cloud data specifically includes:
calculating the average distance between each down-sampling point and the k nearest neighborhood points to the down-sampling point to obtain the average distance corresponding to each down-sampling point;
obtaining the mean value and the standard deviation of the average distance;
acquiring noise point data according to the average distance, the mean value, the standard deviation and a set threshold;
and filtering the noise data from the down-sampling point cloud data.
As an optional implementation manner, the acquiring noise data according to the average distance, the mean, the standard deviation, and a set threshold specifically includes:
aiming at each point p in the down-sampling point cloud data, judging whether the point p meets a formula
And if so, determining the point p as a noise point, otherwise, not determining the point p as the noise point.
As an optional implementation manner, the performing density-based spatial clustering on the road surface three-dimensional point cloud data acquired by the four-eye camera specifically includes:
acquiring a point cloud set S ground { p to be processed 1 ，p 2 ，...，p n }; the point cloud set to be processed is formed by the three-dimensional point cloud data;
initializing a core object set omega = phi, clustering number k is 0, an unvisited sample set = S, and a cluster set C = phi;
traversing the sample set S according to formula N ε (p i )＝{p i ∈S|distance(p i ,p j ) R ≦ r and Ω = { p ≦ g i ∈S||N r (p i ) And | is more than or equal to MinPts }, and obtaining a point p in the core object set i ；
Judging whether the core object set is an empty set or not to obtain a first judgment result; if the first judgment result is yes, ending; otherwise, randomly selecting a core object omega from the core object set, and initializing the core object set omega of the current cluster cur And { omega }, initializing the current cluster sample set C by making the cluster number k and k +1 k Ground { ω }, update the set of unvisited samples Φ = Φ - { ω };
judging whether the current cluster core object set is an empty set or not to obtain a second judgment result; if the second judgment result is negative, updating the core object set omega = omega-C k (ii) a Otherwise, clustering C the current k Added to cluster set C and according to Ω = Ω -C k Updating the core object set, and returning to the step of judging whether the core object set is an empty set;
from the current set of cluster core objects Ω cur Taking out a core object omega', and acquiring a neighborhood sample set N with the Euclidean distance of less than or equal to epsilon from the omega ∈ (ω'), let Δ = N r (ω') # Φ, updating the current cluster sample set C k Ground C k And U.DELTA.updating the set of unaccessed samples phi = phi-DELTA.updating omega cur ＝Ω cur And U (delta n omega) -omega' is transferred to the step of judging whether the current cluster core object set is an empty set.
The invention also provides a road surface obstacle detection system based on the three-dimensional point cloud, which comprises the following components:
the clustering model is used for carrying out density-based spatial clustering on the target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of said clusters is considered an obstacle;
the obstacle width calculation module is used for calculating the difference between the maximum value of an x coordinate and the minimum value of the x coordinate of the point cloud in each cluster under a three-dimensional coordinate system to obtain the width of an obstacle; the three-dimensional coordinate system is a coordinate system established by taking a vehicle as an origin and taking the advancing direction of the vehicle as a z-axis;
the obstacle height calculation module is used for calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of an obstacle;
and the distance calculation module is used for calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle.
Optionally, the system further includes: and the preprocessing module is used for preprocessing the three-dimensional point cloud data acquired by the four-camera before performing density-based spatial clustering on the target point cloud data.
Optionally, the preprocessing module specifically includes:
the interesting region extraction submodule is used for extracting an interesting region from the road surface three-dimensional point cloud data;
the down-sampling sub-module is used for down-sampling the extracted point cloud data of the region of interest to obtain down-sampled point cloud data;
and the outlier filtering submodule is used for filtering outliers of the down-sampling point cloud data.
Optionally, the downsampling sub-module specifically includes:
the dividing subunit is used for dividing the point cloud data of the region of interest into a plurality of cubic grids;
the non-empty cube grid selecting subunit is used for selecting a non-empty cube grid in the cube grid;
the centroid point calculation subunit is used for calculating centroids of all points in the non-empty cubic grids aiming at each non-empty cubic grid to obtain centroid points;
and the downsampling subunit is used for replacing all the points in the non-empty cubic grid with the centroid points to obtain downsampling points corresponding to the non-empty cubic grid.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a method and a system for detecting a road surface obstacle based on three-dimensional point cloud, which comprises the steps of firstly carrying out density-based spatial clustering on road surface three-dimensional point cloud data acquired by a four-camera to obtain a plurality of clusters, wherein each cluster is regarded as an obstacle; aiming at each cluster, calculating the difference between the maximum value of an x coordinate and the minimum value of the x coordinate of a point cloud in the cluster under a three-dimensional coordinate axis to obtain the width of the barrier; calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle; and calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle. The obstacle detection method is directly realized through three-dimensional point cloud data generated by the four-eye camera, so that a perception module of the unmanned system obtains depth information, and the depth information can be in one-to-one correspondence with physical coordinates of a three-dimensional space. The method is applied to the perception of the road surface barrier in the unmanned system, not only can detect the existence of the barrier, but also can detect the size of the barrier and the distance between the barrier and the unmanned vehicle, thereby greatly reducing the traffic accidents of the unmanned vehicle and improving the safety.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a flowchart of a method for detecting a road surface obstacle based on three-dimensional point cloud according to embodiment 1 of the present invention;
fig. 2 is an effect diagram corresponding to a flow of the method for detecting a road surface obstacle based on three-dimensional point cloud provided in embodiment 1 of the present invention;
fig. 3 is a diagram illustrating an effect of an original point cloud provided in embodiment 1 of the present invention;
fig. 4 is an effect diagram of region of interest selection performed on an original point cloud according to embodiment 1 of the present invention;
fig. 5 is an effect diagram of downsampling a point cloud of a region of interest according to embodiment 1 of the present invention;
fig. 6 is an effect diagram of performing outlier filtering on a down-sampled point cloud according to embodiment 1 of the present invention;
fig. 7 is a diagram illustrating an effect of clustering point clouds obtained by filtering out the point clouds according to embodiment 1 of the present invention;
fig. 8 is a diagram showing the result of detecting an obstacle according to embodiment 1 of the present invention;
fig. 9 is a diagram of an example of clustering provided in embodiment 1 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a road surface obstacle detection method and system based on three-dimensional point cloud, which are applied to the perception of road surface obstacles in an unmanned system, can detect the existence of the obstacles, and can detect the size of the obstacles and the distance between the obstacles and an unmanned vehicle, thereby greatly reducing the traffic accidents of the unmanned vehicle and improving the safety.
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof.
Example 1
The present embodiment provides a method for detecting a road obstacle based on three-dimensional point cloud, please refer to fig. 1 and 2, which includes:
s1, carrying out density-based spatial clustering on target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of the clusters is considered an obstacle.
As an optional implementation manner, before the step S1, the method further includes:
s0, preprocessing road surface three-dimensional point cloud data acquired by a four-eye camera;
wherein, the pretreatment specifically comprises:
and S01, extracting the region of interest of the road surface three-dimensional point cloud data.
As can be known from the original three-dimensional point cloud data shown in fig. 3, the point cloud data generated by the four-eye camera inevitably includes some scenes other than the road surface, and some noise points exist at positions close to the camera, which are not needed in the detection of road surface obstacles, so that not only the amount of calculation is increased, but also the detection accuracy is reduced, and therefore, an area of interest needs to be selected to perform targeted processing on the point cloud of the area of interest.
In this embodiment, a threshold segmentation method is adopted, and first, a three-dimensional coordinate system is established, so that the original road surface three-dimensional point cloud data is located under a three-dimensional coordinate axis. The three-dimensional coordinate system is a coordinate system established by taking a vehicle as an origin and taking the advancing direction of the vehicle as a z-axis. Then, a first threshold value is set along the x-axis direction, and the point cloud with the x-axis direction larger than the first threshold value is filtered, so that the effect of filtering noise points on two sides of the road surface is achieved. And then, a second threshold value is set along the z-axis direction, and noise points close to the position of the camera are filtered out, so that point cloud data only containing road surface information is obtained, as shown in fig. 4.
When the three-dimensional coordinate system is established by using the unmanned vehicle as an origin, because the camera is usually arranged on the unmanned vehicle, in order to filter noise close to the position of the camera, point clouds with a z-axis direction smaller than a second threshold value should be filtered.
And S02, performing down-sampling on the extracted point cloud data of the region of interest to obtain down-sampled point cloud data.
Specifically, in this embodiment, a voxel downsampling method may be adopted, the three-dimensional point cloud data in the region of interest is divided into a plurality of cube grids (voxels) of sxsxsxs, then a center of mass point is obtained by calculating all points in each non-empty voxel, and all points in the voxel are replaced by the center of mass point, so as to achieve a downsampling effect of the point cloud, as shown in fig. 5.
Center of mass p of voxel c (x c ，y c ，z c ) The calculation formula is as follows:
wherein n represents the number of point cloud data in a certain voxel, x i 、y i And z i And coordinates in the x-axis direction, the y-axis direction and the z-axis direction of the point cloud data in the voxel are respectively represented.
And S03, outlier filtering is carried out on the down-sampling point cloud data.
Since some outliers exist in the point cloud data after the down-sampling, the outliers also need to be filtered out.
As an alternative embodiment, the outlier filtering is a statistical-based outlier filtering method, and each point p is first calculated i (x i ，y i ，z i ) Average distance to its k neighborhood point
The calculation formula is as follows:
wherein x is j 、y j And z j Is a point p i K neighborhood of (c) j The coordinates of (a).
The average distance between all points in the three-dimensional point cloud data and the k neighborhood is subjected to statistical analysis, the obtained result is similar to a Gaussian distribution, so that the mean value mu and the standard deviation sigma of the distribution can be obtained, and the calculation formula is as follows:
wherein n represents the number of all points in the three-dimensional point cloud data.
Setting a threshold value t, judging whether the point p is retained or filtered, and calculating according to the following formula:
wherein the content of the first and second substances,
represents the average distance between point p and all points in k neighborhood, if
And if the difference value with the mean value mu is less than t times of the standard deviation sigma, keeping the point p, and if the condition is not met, taking the point p as a noise point for filtering.
Through the preprocessing process, noise data in the point cloud data acquired by the four-eye camera is filtered out, as shown in fig. 6. In order to realize the identification of the road surface obstacle, the point cloud data also needs to be clustered, that is, the point cloud data is divided into a plurality of clusters, and each cluster can be regarded as a detected obstacle. The embodiment uses a density-based spatial clustering (DBSCAN) method, which includes the following specific steps:
s11, acquiring a point cloud set S = { p) to be processed 1 ，p 2 ，...，p n H, initializing a core object set omega = phi, clustering number k to 0, an unvisited sample set = S, and a cluster set C = phi; the point cloud set to be processed is formed by the three-dimensional point cloud data of the pavement;
s12, traversing the sample set S according to a formula N ε (p i )＝{p i ∈S|distance(p i ,p j ) R ≦ r and Ω = { p ≦ g i ∈S||N r (p i ) And | is more than or equal to MinPts }, and obtaining a point p in the core object set i ；
Wherein, distance (p) i ，p j ) Representing point p i And point p j R denotes the radius length of the neighborhood, N r (p i ) Represented in the sample set S, at point p i Centered, a sample set of points within a circle of radius r, minPts represents the number of minimum points that a cluster needs to contain, |, represents the total number of collection elements | N r (p i ) | denotes N r (p i ) Number of samples in set, if set N r (p i ) If the number of p is not less than MinPts, p is added i Is denoted as a core object.
S13, judging whether the core object set is an empty set or not to obtain a first judgment result; if the first judgment result is yes, ending; otherwise, randomly selecting a core object omega from the core object set, and initializing the core object set omega cur And (C) = { omega }, the current cluster sample set C is initialized by enabling the cluster number k = k +1 k = ω, update unvisited sample set Φ = Φ - { ω };
s14, judging whether the current cluster core object set is an empty set or not to obtain a second judgment result; if the second judgment result is negative, updating the core object set omega = omega-C k (ii) a Otherwise, clustering C the current k Added to cluster set C and according to Ω = Ω -C k Updating the core object set, and returning to the step S13;
s15, from the current clusterCore object set omega cur Taking out a core object omega', and acquiring a neighborhood sample set N with the Euclidean distance of less than or equal to epsilon from the omega ∈ (ω'), let Δ = N r (ω') # Φ, updating the current cluster sample set C k ＝C k And U.DELTA.updating the set of unaccessed samples phi = phi-DELTA.updating omega cur ＝Ω cur And U (delta n omega) -omega' is transferred to the step of judging whether the current cluster core object set is an empty set. Fig. 7 and 8 show the clustered effect graph and the obstacle detection result graph, respectively.
To make the clustering process provided in this embodiment more clearly understood by those skilled in the art, an example is now provided for description, please refer to fig. 9:
executing S11:
after the previous steps, a point cloud set S = { a, b, c, d, e, f, g, ω 1, ω 2, ω 3, ω 4} is obtained. Initializing an unvisited sample set Φ = S = { a, b, c, d, e, f, g, ω 1, ω 2, ω 3, ω 4}, a core object set
Cluster number k =0, cluster set
Executing S12:
obtaining a core object set Ω = { ω 1, ω 2, ω 3, ω 4}.
Executing S13:
if the core object set omega is not an empty set, then omega 2 is randomly selected as a current processing sample from the core object set omega = { omega 1, omega 2, omega 3, omega 4}, and the current cluster core object set omega is initialized cur = { omega 2}, so that the cluster number k =1, and the current cluster sample set C is initialized 1 = ω 2, update unvisited sample set Φ = Φ - { ω 2} = { a, b, c, d, e, f, g, ω 1, ω 3, ω 4}.
Executing S14:
current cluster core object set omega cur = ω 2, not an empty set, so the core object set Ω = Ω -C is updated 1 ＝{ω1，ω3，ω4}。
Executing S15:
in the current cluster core object set omega cur Taking out a core object omega 2, and finding out a neighborhood sample set N with an Euclidean distance from omega 2 not more than r r (ω 2) = { b, ω 1, ω 3}, let Δ = N r (ω 2) # Φ = { b, ω 1, ω 3}, updating the current cluster sample set C 1 ＝C 1 U = { b, ω 1, ω 2, ω 3}, and update the set of unaccessed samples Φ = Φ - Δ = { a, c, d, e, f, g, ω 4}, update Ω cur ＝Ω cur ∪(Δ∩Ω)-ω2＝{ω1，ω3}
Executing S14:
current cluster core object set omega cur = ω 1, ω 3, not an empty set, so the core object set Ω = Ω -C is updated 1 ＝{ω4}。
Executing S15:
in the current cluster core object set omega cur A core object omega 1 is taken out, and a neighborhood sample set N with an Euclidean distance from omega 1 not more than r is found r (ω 1) = { a, b, ω 2}, let Δ = N r (ω 1) # Φ = { a }, updating the current cluster sample set C 1 ＝C 1 And U delta = { a, b, omega 1, omega 2, omega 3}, updating phi-delta = { c, d, e, f, g, omega 4} of the unvisited sample set phi and updating omega cur ＝Ω cur ∪(Δ∩Ω)-ω1＝{ω3}
Executing S14:
current cluster core object set omega cur = ω 3, not an empty set, so the core object set Ω = Ω -C is updated 1 ＝{ω4}。
Executing S15:
in the current cluster core object set omega cur Taking out a core object omega 3, and finding out a neighborhood sample set N with Euclidean distance from omega 3 not exceeding the epsilon r (ω 3) = { c, ω 2}, let Δ = N r (ω 3) # Φ = { C }, updating the current cluster sample set C 1 ＝C 1 And U is delta = { a, b, c, omega 1, omega 2, omega 3}, phi-delta = { d, e, f, g, omega 4} of the unvisited sample set phi is updated, and
executing S14:
if the current cluster core object set
Then it indicates the current cluster C 1 Has been fully generated, at which time the current cluster C needs to be clustered 1 Added to a cluster set C, i.e. C to C ^ C 1 = { { a, b, C, ω 1, ω 2, ω 3} }, re-updating the core object set Ω = Ω -C 1 = { ω 4}, and the process proceeds to step (S13).
Executing S13:
if the core object set omega is not an empty set, then randomly selecting omega 4 as the current processing sample in the core object set omega = { omega 4}, and initializing the core object set omega of the current cluster cur And (4) keeping the cluster number k =2, and initializing a current cluster sample set C 2 = ω 4, update unvisited sample set Φ = Φ - { ω 4} = { d, e, f, g }.
Executing S15:
in the current cluster core object set omega cur Taking out a core object omega 4, and finding out a neighborhood sample set N with an Euclidean distance from omega 4 not exceeding the epsilon r (ω 4) = { d, e }, let Δ = N r (ω 4) # Φ = { d, e }, updating the current cluster sample set C 2 ＝C 2 And U is equal to delta = { d, e, omega 4}, the unvisited sample set phi is updated to be equal to phi-delta = { f, g }, and the unvisited sample set phi is updated to be equal to or less than zero
Executing S14:
if the current cluster core object set
Then it indicates the current cluster C 2 Has been fully generated, at which time the current cluster C needs to be clustered 2 Added to the cluster set C, i.e. C = C ℃ 2 = { { a, b, c, ω 1, ω 2, ω 3}, { d, e, ω 4} }, and then the core object set is updated
The process proceeds to S13. Executing S13:
the core object set omega is an empty set and the algorithm ends. Get the clustering set C { { a, b, C, ω 1, ω 2, ω 3}, { d, e, ω 4} }.
Through the process, two clusters are finally obtained, wherein { a, b, c, omega 1, omega 2, omega 3} is one cluster, and { d, e, omega 4} is one cluster.
S2, aiming at each cluster, calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under the three-dimensional coordinate axis to obtain the width of the obstacle.
The calculation formula is as follows: w is a group of i ＝max(P i (x))-min(P i (x))。
And S3, calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle.
The calculation formula is as follows: h i ＝max(P i (y))-min(P i (y))。
And S4, calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle.
D i ＝mean(P i (z))
The three-dimensional coordinate system is established by taking the unmanned vehicle body as an original point, so that the average value of the cloud z coordinate values of all points in the cluster is calculated, and the distance between the body and the obstacle can be obtained.
The detection of the road surface barrier provided by the embodiment can be directly realized through three-dimensional point cloud data generated by the four-eye camera, so that the perception module of the unmanned system obtains depth information, and can be in one-to-one correspondence with physical coordinates of a three-dimensional space. The method is applied to sensing of the road surface obstacles in the unmanned system, not only can the existence of the obstacles be detected, but also the size of the obstacles and the distance between the obstacles and the unmanned vehicle can be detected, so that the traffic accidents of the unmanned vehicle are reduced to a great extent, and the safety is improved.
Example 2
The embodiment provides a road surface obstacle detecting system based on three-dimensional point cloud, includes:
the clustering model M1 is used for carrying out density-based spatial clustering on the target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of said clusters is considered an obstacle;
the obstacle width calculation module M2 is used for calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under the three-dimensional coordinate system aiming at each cluster to obtain the width of the obstacle;
the obstacle height calculating module M3 is used for calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle;
and the distance calculation module M4 is used for calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle.
Optionally, the system further includes: and the preprocessing module M0 is used for preprocessing the road surface three-dimensional point cloud data acquired by the four-eye camera before performing density-based spatial clustering on the road surface three-dimensional point cloud data acquired by the four-eye camera.
Optionally, the preprocessing module M0 specifically includes:
the interesting region extraction submodule M01 is used for extracting an interesting region of the road surface three-dimensional point cloud data;
a down-sampling sub-module M02, configured to perform down-sampling on the extracted point cloud data of the region of interest to obtain down-sampled point cloud data;
and the outlier filtering submodule M03 is used for carrying out outlier filtering on the down-sampling point cloud data.
Optionally, the downsampling sub-module M02 specifically includes:
the dividing subunit is used for dividing the point cloud data of the region of interest into a plurality of cube grids;
the non-empty cube grid selecting subunit is used for selecting a non-empty cube grid in the cube grid;
the centroid point calculation subunit is used for calculating centroids of all points in the non-empty cubic grids aiming at each non-empty cubic grid to obtain centroid points;
and the downsampling subunit is used for replacing all points in the non-empty cubic grid with the centroid point to obtain a downsampling point corresponding to the non-empty cubic grid.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the description of the method part.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (10)
1. A road surface obstacle detection method based on three-dimensional point cloud is characterized by comprising the following steps:
carrying out density-based spatial clustering on the target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of said clusters is considered an obstacle;
aiming at each cluster, calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under a three-dimensional coordinate system to obtain the width of the obstacle; the three-dimensional coordinate system is a coordinate system established by taking a vehicle as an origin and taking the advancing direction of the vehicle as a z-axis;
calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of the obstacle;
and calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the vehicle and the obstacle.
2. The method of claim 1, wherein prior to said density-based spatial clustering of target point cloud data, the method further comprises: preprocessing three-dimensional point cloud data acquired by the four-eye camera:
extracting an interesting region of the three-dimensional point cloud data;
down-sampling the extracted point cloud data of the region of interest to obtain down-sampled point cloud data;
and carrying out outlier filtering on the down-sampling point cloud data.
3. The method according to claim 2, wherein the down-sampling of the extracted point cloud data of the region of interest to obtain down-sampled point cloud data specifically comprises:
dividing the point cloud data of the region of interest into a plurality of cube grids;
selecting a non-empty cubic grid in the cubic grids;
calculating the centroid of all points in each non-empty cubic grid to obtain a centroid point;
and replacing all points in the non-empty cubic grid with the centroid points to obtain down-sampling points corresponding to the non-empty cubic grid.
4. The method of claim 2, wherein the outlier filtering of the down-sampled point cloud data comprises:
calculating the average distance between each down-sampling point and the k nearest neighborhood points to the down-sampling point to obtain the average distance corresponding to each down-sampling point;
obtaining the mean value and the standard deviation of the average distance;
acquiring noise point data according to the average distance, the mean value, the standard deviation and a set threshold;
and filtering the noise data from the down-sampling point cloud data.
5. The method according to claim 4, wherein the obtaining noise data according to the average distance, the mean, the standard deviation, and a set threshold specifically comprises:
aiming at each point p in the down-sampling point cloud data, judging whether the point p meets a formula
If yes, determining the point p as a noise point, otherwise, determining the point p not as a noise point.
6. The method of claim 1, wherein the performing density-based spatial clustering on the target point cloud data comprises:
acquiring a point cloud set S = { p) to be processed 1 ，p 2 ，...，p n }; the point cloud set to be processed is formed by the three-dimensional point cloud data;
initializing a core object set omega = phi, the number of clusters k =0, an unvisited sample set = S, and a cluster set C = phi;
traversing the sample set S according to formula N r (p i )＝{p i ∈S|distance(p i ,p j ) R ≦ r and Ω = { p ≦ g i ∈S||N r (p i ) | is more than or equal to MinPts }, and a point p in the core object set is obtained i ；distance(p i ，p j ) Representing point p i And point p j The Euclidean distance of (c); r represents the radius length of the neighborhood; n is a radical of r (p i ) Represented in the sample set S, at point p i A sample set consisting of points within a circle with a radius r as the center; minPts represents the number of minimum points contained in a cluster; | · | represents the total number of collection elements;
judging whether the core object set is an empty set or not to obtain a first judgment result; if the first judgment result is yes, ending; otherwise, randomly selecting a core object omega from the core object set, and initializing the core object set omega of the current cluster cur And (C) = { omega }, the current cluster sample set C is initialized by enabling the cluster number k = k +1 k = ω, update unvisited sample set Φ = Φ - { ω };
judging whether the current cluster core object set is an empty set or not to obtain a second judgment result; if the second judgment result is negative, updating the core object set omega = omega-C k (ii) a Otherwise, clustering C the current k Added to cluster set C and according to Ω = Ω -C k Updating the core object set, and returning to the step of judging whether the core object set is an empty set;
from the current set of cluster core objects Ω cur Taking out a core object omega', and acquiring a neighborhood sample set N with the Euclidean distance of less than or equal to epsilon from the omega ∈ (ω'), let Δ = N r (ω') # Φ, updating the current cluster sample set C k ＝C k And U.DELTA.updating the set of unaccessed samples phi = phi-DELTA.updating omega cur ＝Ω cur And U (delta ≧ n Ω) - ω', switching to the step of judging whether the current cluster core object set is an empty set.
7. A road surface obstacle detection system based on three-dimensional point cloud is characterized by comprising:
the clustering model is used for carrying out density-based spatial clustering on the target point cloud data to obtain a plurality of clusters; the target point cloud data is obtained by shooting a road scene by a four-eye camera; each of said clusters is considered an obstacle;
the obstacle width calculation module is used for calculating the difference between the maximum value of the x coordinate and the minimum value of the x coordinate of the point cloud in the cluster under a three-dimensional coordinate system aiming at each cluster to obtain the width of an obstacle; the three-dimensional coordinate system is a coordinate system established by taking a vehicle as an origin and taking the advancing direction of the vehicle as a z-axis;
the obstacle height calculation module is used for calculating the difference between the maximum value of the y coordinate and the minimum value of the y coordinate of the point cloud in the cluster to obtain the height of an obstacle;
and the distance calculation module is used for calculating the mean value of the cloud z coordinate values of all points in the cluster to obtain the distance between the body and the obstacle.
8. The system of claim 7, further comprising: and the preprocessing module is used for preprocessing the three-dimensional point cloud data acquired by the four-eye camera before performing density-based spatial clustering on the target point cloud data.
9. The system according to claim 8, wherein the preprocessing module specifically includes:
the interesting region extraction submodule is used for extracting an interesting region from the road surface three-dimensional point cloud data;
the down-sampling sub-module is used for down-sampling the extracted point cloud data of the region of interest to obtain down-sampled point cloud data;
and the outlier filtering sub-module is used for carrying out outlier filtering on the down-sampling point cloud data.
10. The system of claim 9, wherein the downsampling sub-module specifically comprises:
the dividing subunit is used for dividing the point cloud data of the region of interest into a plurality of cubic grids;
the non-empty cubic grid selecting subunit is used for selecting a non-empty cubic grid in the cubic grids;
the centroid point calculation subunit is used for calculating centroids of all points in the non-empty cubic grids aiming at each non-empty cubic grid to obtain centroid points;
and the downsampling subunit is used for replacing all the points in the non-empty cubic grid with the centroid points to obtain downsampling points corresponding to the non-empty cubic grid.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210965243.3A CN115359453A (en) | 2022-08-12 | 2022-08-12 | Road surface obstacle detection method and system based on three-dimensional point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210965243.3A CN115359453A (en) | 2022-08-12 | 2022-08-12 | Road surface obstacle detection method and system based on three-dimensional point cloud |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115359453A true CN115359453A (en) | 2022-11-18 |
Family
ID=84033187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210965243.3A Pending CN115359453A (en) | 2022-08-12 | 2022-08-12 | Road surface obstacle detection method and system based on three-dimensional point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115359453A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116184435A (en) * | 2023-02-27 | 2023-05-30 | 麦岩智能科技(北京)有限公司 | Cliff detection method, storage medium and device based on robot |
| CN117252863A (en) * | 2023-11-13 | 2023-12-19 | 山东省地质测绘院 | Quick detection and analysis method for geographic information abnormal data |
-
2022
- 2022-08-12 CN CN202210965243.3A patent/CN115359453A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116184435A (en) * | 2023-02-27 | 2023-05-30 | 麦岩智能科技(北京)有限公司 | Cliff detection method, storage medium and device based on robot |
| CN117252863A (en) * | 2023-11-13 | 2023-12-19 | 山东省地质测绘院 | Quick detection and analysis method for geographic information abnormal data |
| CN117252863B (en) * | 2023-11-13 | 2024-02-09 | 山东省地质测绘院 | Quick detection and analysis method for geographic information abnormal data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115359453A (en) | Road surface obstacle detection method and system based on three-dimensional point cloud | |
| Srinivasa | Vision-based vehicle detection and tracking method for forward collision warning in automobiles | |
| Tan et al. | Color model-based real-time learning for road following | |
| CN111260683A (en) | Target detection and tracking method and device for three-dimensional point cloud data | |
| JP7301138B2 (en) | Pothole detection system | |
| CN115049700A (en) | Target detection method and device | |
| Jia et al. | Real-time obstacle detection with motion features using monocular vision | |
| Feniche et al. | Lane detection and tracking for intelligent vehicles: A survey | |
| WO2021013227A1 (en) | Image processing method and apparatus for target detection | |
| Zhang et al. | A framework for turning behavior classification at intersections using 3D LIDAR | |
| CN113989784A (en) | Road scene type identification method and system based on vehicle-mounted laser point cloud | |
| Javadi et al. | A robust vision-based lane boundaries detection approach for intelligent vehicles | |
| John et al. | A reliable method for detecting road regions from a single image based on color distribution and vanishing point location | |
| EP3555854B1 (en) | A method of tracking objects in a scene | |
| Yiruo et al. | Complex ground plane detection based on v-disparity map in off-road environment | |
| Saleem et al. | Steering angle prediction techniques for autonomous ground vehicles: a review | |
| Ghahremannezhad et al. | A new adaptive bidirectional region-of-interest detection method for intelligent traffic video analysis | |
| CN107220632B (en) | Road surface image segmentation method based on normal characteristic | |
| El Jaafari et al. | A novel approach for on-road vehicle detection and tracking | |
| Ghahremannezhad et al. | Robust road region extraction in video under various illumination and weather conditions | |
| CN113516853B (en) | Multi-lane traffic flow detection method for complex monitoring scene | |
| Zhang et al. | Real-time obstacle detection based on stereo vision for automotive applications | |
| Dinh et al. | Image segmentation based on histogram of depth and an application in driver distraction detection | |
| Zarbakht et al. | Lane detection under adverse conditions based on dual color space | |
| Chen et al. | Vehicle tracking and distance estimation based on multiple image features |
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 |