阅读说明：本技术 一种基于轮廓分割的车库点云地图构建方法 (Garage point cloud map construction method based on contour segmentation ) 是由 张朝昆 郭钰 周勃麟 单涛涛 于 2021-06-22 设计创作，主要内容包括：本发明公开了一种基于轮廓分割的车库点云地图构建系统,该系统包括依序连接的竖直平面提取模块、边缘墙面提取模块和墙面点云配准模块,以激光雷达扫描到的连续点云帧为输入,所述竖直平面分割模块将点云帧中的竖直平面分割出来,所述边缘墙面提取模块将边缘墙面提取出来,所述墙面点云配准模块将两帧点云配准到一起从而处理所有点云帧、最终得到完整点云地图。与现有技术相比,本发明1)改善了车库中点云地图构建的精度以及相邻两帧的配准精度；2)在不依赖其他高精度定位解决方案以及高精度传感器设备的情况下提高大范围点云地图构建的精度；3)在重复结构多的车库场景下,实现点云地图的构建。(The invention discloses a garage point cloud map construction system based on contour segmentation, which comprises a vertical plane extraction module, an edge wall surface extraction module and a wall surface point cloud registration module which are sequentially connected, wherein continuous point cloud frames scanned by a laser radar are used as input, the vertical plane segmentation module segments a vertical plane in the point cloud frames, the edge wall surface extraction module extracts an edge wall surface, and the wall surface point cloud registration module registers two frames of point clouds together so as to process all the point cloud frames and finally obtain a complete point cloud map. Compared with the prior art, the method has the advantages that 1) the construction precision of the point cloud map in the garage and the registration precision of two adjacent frames are improved; 2) the method has the advantages that the construction precision of the large-range point cloud map is improved under the condition of not depending on other high-precision positioning solutions and high-precision sensor equipment; 3) under the garage scene with a plurality of repeated structures, the point cloud map is constructed.)

1. The utility model provides a garage point cloud map construction system based on outline is cut apart which characterized in that, this system includes vertical plane extraction module, edge wall extraction module and the registration module of wall point cloud that connects gradually, wherein:

the method comprises the steps that continuous point cloud frames scanned by a laser radar are used as input, a vertical plane segmentation module segments a vertical plane in the point cloud frames, an edge wall surface extraction module extracts an edge wall surface, and a wall surface point cloud registration module registers two frames of point clouds together so as to process all the point cloud frames and finally obtain a complete point cloud map.

2. A garage point cloud map construction method based on contour segmentation is characterized by comprising the following steps:

step 1: assuming that the current time is t, extracting specific components from data with a large amount of noise for the point cloud conforming to the point cloud model to be used as pre-registered two-frame point cloud P_t-1And P_tThe point cloud obtained after filtering the noise isAndpoint cloud hereinThe model is a plane point cloud of an xOy coordinate plane which is vertical to a coordinate system of the current extracted point cloud;

two frames of point clouds after filteringAndpoint-to-point cloud by using RanSaC algorithmAndextracting the vertical plane to obtain two vertical plane point cloud setsAnd

step 1.3: constructing a vertical planar point cloud setAndindex I of_t-1、I_t；

Step 2: extracting edge wall surfaces, namely utilizing an alpha-shape algorithm to obtain a vertical plane point cloud set from the step 1Andextracting the edge contour of the midpoint set, wherein after the extraction is finished, the edge contour of the midpoint set is extractedAndedge wall set inAnd

and step 3: carrying out wall point cloud registration, namely firstly carrying out the edge wall set extracted in the step 2Andare combined into two wall point clouds W_t-1And W_t；

Step 3.2: using ICP algorithm to W_t-1And W_tPoint cloud registration is carried out, the ICP registration step is shown in figure 3, and the final registration result is P_t-1And P_tThe result of the registration of (1).

Technical Field

The invention relates to the technical field of high-precision map construction, in particular to a garage point cloud map construction method.

Background

The three-dimensional point cloud map has a very wide application prospect, and the application fields include but are not limited to intelligent driving, intelligent home, three-dimensional reconstruction, digital earth, urban planning, disaster prevention and reduction, ocean mapping and the like. The current optional point cloud map construction scheme comprises a laser SLAM mapping technology, a point cloud splicing mapping technology and the like.

The traditional garage interior point cloud map building technology adopts a laser SLAM map building technology. Because the algorithm uses a uniform motion model as motion prediction, the mapping accuracy can only be ensured within a short distance, and the accumulated drift error is larger along with the longer running time of the whole system. Under the condition that the garage scene range is large, the map constructed by the laser SLAM is not accurate enough and is not suitable for being used as a map construction basis.

The point cloud splicing and composition technology mainly depends on a point cloud registration algorithm to splice continuous point cloud frames, and compared with a laser SLAM, the point cloud splicing and composition technology has the advantage that the map is stable in the construction process and has higher precision. When the point cloud splicing composition technology is used, the point cloud registration algorithm is mainly relied on for registering two adjacent frames of point clouds, the repeatability of the environment in a garage scene is high, the repeatability of the scanned point cloud structure is also high, and the mismatching is easily generated only by relying on the registration algorithm.

The GPS is not available in a garage scene, so that the problem of mismatching caused by repeated structures in the garage scene cannot be solved by combining a GPS-RTK map construction technology in the garage scene.

Disclosure of Invention

In view of the above, the invention provides a garage point cloud map construction system and method based on contour segmentation, which realize garage point cloud map construction by using continuous point cloud frames scanned by a vehicle-mounted laser radar sensor and utilizing the extraction, registration and other processing of the collected garage point cloud frames.

The invention is realized by the following technical scheme:

the utility model provides a garage point cloud map construction system based on contour segmentation, this system includes vertical plane extraction module, edge wall extraction module and the registration module of wall point cloud that connects gradually, wherein:

the method comprises the steps that continuous point cloud frames scanned by a laser radar are used as input, a vertical plane segmentation module segments a vertical plane in the point cloud frames, an edge wall surface extraction module extracts an edge wall surface, and a wall surface point cloud registration module registers two frames of point clouds together so as to process all the point cloud frames and finally obtain a complete point cloud map.

A garage point cloud map construction method based on contour segmentation comprises the following steps:

step 1: assuming that the current time is t, extracting specific components from data with a large amount of noise for the point cloud conforming to the point cloud model to be used as pre-registered two-frame point cloud P_t-1And P_tThe point cloud obtained after filtering the noise isAndthe point cloud model is a plane point cloud of an xOy coordinate plane which is vertical to a coordinate system of the current extracted point cloud;

two frames of point clouds after filteringAndpoint-to-point cloud by using RanSaC algorithmAndextracting the vertical plane to obtain two vertical plane point cloud setsAnd

step 1.3: constructing a vertical planar point cloud setAndindex I of_t-1、I_t；

Step 2: extracting edge wall surfaces, namely utilizing an alpha-shape algorithm to obtain a vertical plane point cloud set from the step 1Andextracting the edge contour of the midpoint set, wherein after the extraction is finished, the edge contour of the midpoint set is extractedAndedge wall set inAnd

and step 3: carrying out wall point cloud registration, namely firstly carrying out the edge wall set extracted in the step 2Andare combined into two wall point clouds W_t-1And W_t；

Step 3.2: using ICP algorithm to W_t-1And W_tPoint cloud registration is carried out, the ICP registration step is shown in figure 3, and the final registration result is P_t-1And P_tThe result of the registration of (1).

Compared with the prior art, the invention can achieve the following beneficial effects:

1) the construction precision of the point cloud map in the garage and the registration precision of two adjacent frames are improved;

2) the method has the advantages that the construction precision of the large-range point cloud map is improved under the condition of not depending on other high-precision positioning solutions and high-precision sensor equipment;

3) under the garage scene with a plurality of repeated structures, the point cloud map is constructed.

Drawings

Fig. 1 is a schematic structural diagram of a garage point cloud map construction system based on contour segmentation.

Fig. 2 is a schematic flow chart of a garage point cloud map construction method based on contour segmentation.

Fig. 3 is a flowchart of the ICP registration algorithm.

Detailed Description

The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

Fig. 1 is a schematic structural diagram of a garage point cloud map construction system based on contour segmentation according to the present invention. The system comprises a vertical plane extraction module 10, an edge wall extraction module 20 and a wall point cloud registration module 30. The method comprises the steps of taking continuous point cloud frames scanned by a laser radar as input, then segmenting a vertical plane in the point cloud frames through a vertical plane segmentation module, then extracting an edge wall surface through an edge wall surface extraction module, registering two frames of point clouds together through a wall surface point cloud registration module, and processing all the point cloud frames to finally obtain a complete point cloud map. Wherein:

the vertical plane extraction module extracts a vertical plane set in the point cloud by using a RanSaC algorithm, and the wall surface point cloud set is a subset of the set.

The edge wall surface extraction module integrates the previously extracted vertical planes together according to frames, calculates the mass center of each vertical plane, concentrates the mass center in a coordinate system and projects the mass center to the xoy plane, and extracts the edge wall surface through an alpha-shape algorithm.

The wall surface point cloud registration module is used for preprocessing the extracted wall surface, fusing point clouds of the same wall surface, registering the preprocessed wall surface point cloud result, and taking the registered result as the final registered result between two frames of point clouds.

As shown in fig. 2, the implementation process of the garage point cloud map construction method based on contour segmentation specifically includes the following steps:

step 1: assuming that the current time is t, using RanSaC (random Sample consensus) algorithm to pre-register two adjacent frames P according to a preset parameter model_t-1And P_tExtracting specific components, namely vertical plane point cloud sets from data with a large amount of noise by point clouds conforming to point cloud models in point cloudsAndstep 1.1: and removing the vehicle point cloud and other noise points in the point cloud frame. When the point cloud is collected, due to the collecting equipment, completely invalid points of the collecting equipment can be generated and eliminated through the condition filtering of fixed parameters. Taking the currently used data as an example: data are collected by a Velodyne-32 laser radar carried by a three-compartment type Beijing sedan, and the range of conditional filtering parameters in the data is-2<x<1.5、 -0.7<y<0.7、-1.5<z<0. In addition, the invalid point in the point Cloud frame needs to be eliminated, and a removeNaNFromPoint Cloud party in PCL (Point Cloud library) is usedThe invalid point can be removed. The point cloud obtained after the filtering in the step isAnd P^f；

Step 1.2: two frames of point clouds after filteringAndperforming planar extraction, comprising:

the parameters in the RanSaC algorithm are set as follows:

(1) determining a minimum number of samples M;

(2) determining a reference vector

(3) The extraction model is set to a constant SACMODEL _ PARALLEL _ PLANE which represents the extraction mode as PARALLEL to the reference vector;

utilizing RanSaC algorithm to point cloud according to the set parametersAndextracting the vertical plane to obtain two vertical plane point cloud setsAndextracting a vertical plane which is vertical to an xOy coordinate plane in a laser radar coordinate system by using a RanSaC algorithm;

step 1.3: constructing a vertical planar point cloud setAndindex I of_t-1、I_tTo do so byFor the purpose of example only,the elements in (1) begin with t, and the extraction time sequence is used as the differentiation to construct an index I_tE.g. ofThe n-th extracted vertical plane is at index I_tThe content of the indication in (1) is P_t,nThe subsequent treatment and use are convenient;

step 2: extracting an edge wall surface, and extracting an edge wall surface set from the vertical plane point cloud set obtained in the step 1 by utilizing an alpha-shape algorithm so as toFor example, the edge wall surface is extracted from the wall surfaceIs provided withThe number of the medium element is N_tUsing S_i(i＝1,2,3,······,N_t) To represent a vertical planar point cloudA vertical plane of; the alpha-shape algorithm is a method for abstracting an intuitive shape of a discrete space point set, and specifically is a method for finding out a stack of unordered discrete point outlines through the algorithm;

step 2.1: representing a vertical plane by a mass center, and calculating a point cloud set of the vertical planeEach of which is a vertical plane S_iCenter of mass m_iThe formula is as follows:

wherein p is_jIs a vertical plane S_iA point of (1);

storing mass center m by taking plane point cloud serial number as index_i(x, y, z) which is then mapped to the xOy coordinate plane to reduce its coordinates to m'_i(x,y)；

The centroid point set calculation method is the same as above.

Step 2.2: performing edge wall surface extraction, namely, using an alpha-shape algorithm to perform the vertical plane point cloud set obtained in the step 1Andextracting the edge contour of the centroid point set of the middle vertical plane, and after extraction is finished, extractingAndedge wall set inAnd

and step 3: carrying out wall point cloud registration, namely firstly carrying out the edge wall set extracted in the step 2Andintegrating into two Point cloud sets, then registering by using Iterative Closest Point (ICP) algorithm, calculating a feature descriptor for each Point in the two input Point clouds to be registered, searching corresponding points in the two Point clouds according to the feature descriptor, then removing the corresponding points which do not meet requirements, and then calculating pose transformation between the two Point clouds by using the remaining points;

step 3.1: integrating wall point clouds, namely merging the edge wall surfaces extracted in the step 2 by using an addition overloading method in PCL (polycaprolactone), and merging a plurality of edge wall surfaces extracted from the same vertical plane point cloud set into two wall point clouds W_t-1And W_tMerging, the formula is as follows:

wherein M is_t-1And M_tAre respectively asAndthe number of elements in the middle edge wall set;

step 3.2: using ICP algorithm to W_t-1And W_tPoint cloud registration is carried out, and the final registration result is P_t-1And P_tThe result of the registration of (1).

Fig. 3 shows a flowchart of the ICP registration algorithm. The registration algorithm comprises the following specific steps:

input devicePoint cloud W_t-1、W_t(ii) a Calculating a feature descriptor of each point; finding out corresponding points in the two point clouds; and removing corresponding points by using the existing method, realizing pose change estimation and calculating a transformation matrix T.

The invention has the advantages that: (1) only a laser point cloud frame acquired by a laser radar sensor is needed, and other sensors such as an IMU (inertial measurement Unit) and the like are not needed to maintain the drawing construction precision; (2) the contour extraction is used for ensuring the matching of two frames of point clouds in the registration process, and the mismatching caused by a repetitive structure when the point cloud splicing composition technology is directly used is avoided; (3) as the contour extraction uses the traditional point cloud processing algorithm, the calculation power can be saved more.