WO2022121177A1 - 一种基于扫描线的道路点云的提取方法 - Google Patents

一种基于扫描线的道路点云的提取方法 Download PDF

Info

Publication number
WO2022121177A1
WO2022121177A1 PCT/CN2021/084078 CN2021084078W WO2022121177A1 WO 2022121177 A1 WO2022121177 A1 WO 2022121177A1 CN 2021084078 W CN2021084078 W CN 2021084078W WO 2022121177 A1 WO2022121177 A1 WO 2022121177A1
Authority
WO
WIPO (PCT)
Prior art keywords
point
scan line
ground
points
adjacent
Prior art date
Application number
PCT/CN2021/084078
Other languages
English (en)
French (fr)
Inventor
陈磊
徐忠建
朱必亮
徐云和
Original Assignee
速度时空信息科技股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 速度时空信息科技股份有限公司 filed Critical 速度时空信息科技股份有限公司
Priority to EP21901893.4A priority Critical patent/EP4120123A4/en
Publication of WO2022121177A1 publication Critical patent/WO2022121177A1/zh

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
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/176—Urban or other man-made structures
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/02—Systems using the reflection of electromagnetic waves other than radio waves
    • G01S17/06—Systems determining position data of a target
    • G01S17/42—Simultaneous measurement of distance and other co-ordinates
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/93—Lidar systems specially adapted for specific applications for anti-collision purposes
    • G01S17/931—Lidar systems specially adapted for specific applications for anti-collision purposes of land vehicles
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
    • G01S7/4808—Evaluating distance, position or velocity data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • G06V10/267—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 by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices
    • 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
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • the invention relates to the field of target recognition, in particular to a method for extracting road point clouds based on scan lines, which can be used for extracting urban roads, highways, etc. from vehicle-mounted LiDAR point clouds.
  • the algorithms are mainly divided into the following categories: (1) gradient-based filtering algorithms; (2) morphological filtering algorithms; (3) segmentation-based filtering algorithms; (4) irregular triangulation-based filtering algorithms filtering algorithm; (5) filtering algorithm for surface fitting; (6) filtering algorithm based on scan line.
  • Sun Meiling et al. proposed a one-dimensional asymptotic morphological filtering algorithm based on scan lines, which analyzes the difference between the calculated point value and its original value on the airborne LiDAR point cloud, and gradually filters out the non-ground points.
  • Wang Hui et al. proposed a mathematical morphology LiDAR point cloud filtering method for scanning lines, aiming at the shortcomings of the interpolation error of Keqi Zhang's method, and effectively dealt with the scanning polyline problem.
  • Wu Jian et al. first used the RANSAC algorithm to roughly extract the ground points, and then used the multi-feature, wide threshold, and hierarchical method to extract the road boundary feature points, and finally applied the RANSAC method to fit the road boundary curve model.
  • the algorithm of this method is complex, and there are many threshold setting items, so it is difficult to adapt to the needs of different complex scenarios.
  • Shi Wenzhong determined scan lines based on ground point information and scan line separation conditions, and then extracted road edge information from the scan lines. This method does not perform well for scenes with occluded road edges.
  • the technical problem to be solved by the present invention is to provide a method for extracting road point clouds based on scan lines, which can identify the range of urban roads and highway pavements, and at the same time can solve the discontinuity of road extraction caused by terrain fluctuations, large-scale objects occlusion, etc. and other problems to improve the efficiency and accuracy of road recognition.
  • the method for extracting road point clouds based on scan lines specifically includes the following steps:
  • S1 Select the starting point: extract the route edge point of each scan line from the vehicle point cloud as the starting point of the scan line;
  • the search is performed on the basis of the starting point of each scan line in the step S1, and according to the height difference between the starting point of the scan line and its adjacent point, the adjacent point is distinguished as ground point or feature point;
  • S3 searches for ground points along the scan line: if the adjacent point is a ground point, add the adjacent point to the set of ground points on the scan line; repeat step S2 again, and continue to judge whether the next adjacent point is a ground point or a feature point; if If the adjacent point is a ground object point, then use the existing ground point on the scan line to construct a space straight line equation, and judge that the next adjacent point of the adjacent point is a ground object point or a ground point, until it traverses to the next scan line end of the starting point;
  • S4 filter ground points cluster the ground points on each scan line, and then filter out the ground points smaller than the set length threshold, so as to obtain a coherent scan line with noise removal;
  • S5 obtains the road boundary: based on the coherent scan lines obtained in the step S4 for removing noise, select the left and right endpoints of each scan line, fit the boundaries on the left and right sides of the road, and complete the extraction of the road boundary and regularization;
  • the specific steps of determining that the adjacent point is a ground point or a feature point in the step S2 are: searching for left and right directions based on the starting point of the scanning line, and judging whether the height difference H ij between the adjacent point and the starting point is not. is less than the height difference threshold H t , if it is less than the height difference threshold H t , it is judged as a ground point, and the adjacent point is added to the ground point set of the scan line; if it is higher than the height difference threshold H t , it is judged as a ground object point,
  • the formula for calculating the height difference H ij is:
  • h j is the height of the point to be judged, that is, the adjacent point j
  • hi is the height of the current ground point i .
  • step S2 it is judged whether the height difference between the starting point of the scan line and its adjacent point is less than the set threshold, so as to judge that the adjacent point is a ground point or a feature point; Construct a set of ground points, then cluster and denoise the ground points to obtain coherent scan lines, and then obtain the boundaries on both sides of the road through fitting, complete the extraction of road boundaries, and solve the problems caused by terrain fluctuations and large-scale occlusions. Road extraction discontinuity and other problems, improve the efficiency and accuracy of road recognition.
  • , ⁇ z
  • the region growing algorithm is used to cluster the ground points on each scan line, and the specific steps of filtering out the ground points smaller than the set length threshold are as follows:
  • S41 First, randomly select a point in the ground point set of each scan line as a seed point, and judge the distance between it and surrounding adjacent points. If it is less than the set distance threshold D t , it is considered that the adjacent point belongs to a cluster class, and mark it; then use the marked adjacent point as the seed point to repeatedly search for its surrounding adjacent points, until there is no adjacent point in the surrounding distance within the range of the distance threshold D t ; if the scan line is still unmarked point, then randomly select a point as the seed point, and repeat the above operation until all points of the scan line are marked;
  • the left and right endpoints of each scan line are selected, and the boundaries on the left and right sides of the road are fitted by the least squares algorithm according to the left and right directions respectively.
  • the driving direction of the car is segmented according to the distance DS to accurately extract the boundary of the arc-shaped road.
  • the route edge point is stored in the LAS file, and its value is represented by 1, which represents the end of the previous scan line and the start of the next scan line. Therefore, all points in the LAS file with the edge point value of 1 are extracted as the starting point of each scan line.
  • the present invention has the beneficial effects as follows: the method for extracting road point clouds based on scan lines adopts one-dimensional search on the basis of scan lines, which effectively avoids the need for two-dimensional or even three-dimensional search in scenes with large data volumes. In addition, it also performs well in areas with large terrain fluctuations or slope sections, and the algorithm can also robustly extract road boundaries for areas occluded by large ground objects.
  • Fig. 1 is the flow chart of the extraction method of the road point cloud based on scan line of the present invention
  • Fig. 2 is the original data diagram of the test area of the method for extracting the road point cloud based on the scan line of the present invention
  • FIG. 3 is a result diagram of the distinction between ground points and ground objects in the test area of the method for extracting road point clouds based on scan lines of the present invention
  • Fig. 4 is the ground point map extracted by the method for extracting the road point cloud based on the scan line of the present invention
  • FIG. 5 is a road boundary line diagram extracted by the method for extracting road point clouds based on scan lines of the present invention.
  • the method for extracting road point clouds based on scan lines specifically includes the following steps:
  • Fig. 2 is the test area data of the present invention.
  • the route edge point of each scan line is extracted from the vehicle point cloud as the starting point of the scan line; the route edge point is stored in the LAS file, and its value is represented by 1, representing the end of the previous scan line and the next scan line The start of ; extract all the points with the edge point value of 1 in the LAS file as the starting point of each scan line;
  • S2 judges the ground point or the feature point: searches are performed based on the starting point of each scan line in the step S1, and judge whether the height difference between the starting point of the scan line and its adjacent points is less than the set threshold, Thereby distinguishing that the adjacent point is a ground point or a feature point;
  • the specific steps of determining that the adjacent point is a ground point or a feature point in the step S2 are: searching for left and right directions based on the starting point of the scanning line, and judging whether the height difference H ij between the adjacent point and the starting point is not. is less than the height difference threshold H t , if it is less than the height difference threshold H t , it is judged as a ground point, and the adjacent point is added to the ground point set of the scan line; if it is higher than the height difference threshold H t , it is judged as a ground object point,
  • the formula for calculating the height difference H ij is:
  • h j is the height of the point to be judged, that is, the adjacent point j, and hi is the height of the current ground point i ;
  • the height difference threshold H t is determined by the smoothness of the road surface; through experiments, it is found that if the height difference threshold value H t is set to L, the ground point is easily mistakenly judged as a feature point, which increases the complexity of the algorithm; If the threshold H t is set to 3L, the feature points are easily misjudged as ground points; if the height difference threshold H t is set to 2L, the extraction of ground points is more in line with the actual situation; therefore, the setting of the height difference threshold is directly related to the ground.
  • the specific threshold value is mainly related to the distance L between adjacent points, and is determined by the smoothness of the road surface.
  • S3 searches for ground points along the scan line: if the adjacent point is a ground point, add the adjacent point to the set of ground points on the scan line; repeat step S2 again, and continue to judge whether the next adjacent point is a ground point or a feature point; if If the adjacent point is a ground object point, then use the existing ground point on the scan line to construct a space straight line equation, and judge that the next adjacent point of the adjacent point is a ground object point or a ground point, until it traverses to the next scan line end of the starting point;
  • , ⁇ z
  • S4 filter ground points cluster the ground points on each scan line, and then filter out the ground points smaller than the set length threshold, so as to obtain a coherent scan line with noise removal;
  • step S4 the area growth algorithm is used to cluster the ground points on each scan line, and the specific steps of filtering out the ground points smaller than the set length threshold are as follows:
  • S41 First, randomly select a point in the ground point set of each scan line as a seed point, and judge the distance between it and surrounding adjacent points. If it is less than the set distance threshold D t , it is considered that the adjacent point belongs to a cluster class, and mark it; then use the marked adjacent points as seed points to repeatedly search for its surrounding adjacent points, until no adjacent points are within the range of the distance threshold D t ; if the scan line is still unmarked point, then randomly select a point as the seed point, and repeat the above operation until all points of the scan line are marked;
  • S5 obtains the road boundary: based on the coherent scan lines obtained in the step S4 for removing noise, select the left and right endpoints of each scan line, fit the boundaries on the left and right sides of the road, and complete the extraction of the road boundary and regularization;
  • step S5 the left and right endpoints of each scan line are selected, and the boundaries on the left and right sides of the road are fitted by the least squares algorithm according to the left and right directions respectively.
  • segmented fitting is performed according to the distance DS according to the driving direction of the car, which can effectively solve the inaccurate extraction of the curved road boundary; the distance DS can be selected according to the actual situation, and it is recommended to select 1m as the threshold of the distance DS.
  • Figure 5 5.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Artificial Intelligence (AREA)
  • Electromagnetism (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Processing Or Creating Images (AREA)

Abstract

本发明公开了一种基于扫描线的道路点云的提取方法，步骤包括：S1从车载点云中提取每条扫描线的航线边缘点作为扫描线的起始点；S2以起始点进行搜索，根据起始点与其相邻点之间的高差，判断地面点或地物点；S3若相邻点是地面点，则将相邻点加入该扫描线地面点集合；再重复步骤S2，判断下一个相邻点；若相邻点是地物点，则以该扫描线上已有的地面点构建空间直线方程，并判断相邻点的下一个相邻点，直至遍历至下一条扫描线的起始点结束；S4将每条扫描线上的地面点进行聚类，并过滤掉噪声，得到连贯的扫描线；S5以连贯的扫描线为基础，选取每条扫描线的左端点和右端点，拟合出道路左右两侧的边界，完成道路边界的提取和规则化。

Description

一种基于扫描线的道路点云的提取方法 技术领域
本发明涉及目标识别领域，具体涉及一种基于扫描线的道路点云的提取方法，可用于车载LiDAR点云的城市道路、公路等的提取。
背景技术
道路作为城市或城郊地面的重要特征，在城市规划、灾害预报、资源勘探、地理信息系统库更新、数字化城市以及军事侦察等领域都有着重要的作用。近年来，随着LiDAR技术的快速发展，基于车载和机载LiDAR技术探测道路点云的研究越来越多。根据滤波原理的不同，算法主要分为以下几类：(1)基于坡度的滤波算法；(2)基于形态学滤波算法；(3)基于分割的滤波算法；(4)基于不规则三角网的滤波算法；(5)曲面拟合的滤波算法；(6)基于扫描线的滤波算法。
孙美玲等提出一种基于扫描线的一维渐近式形态学滤波算法，在机载LiDAR点云上分析运算后的点值与其原始值的差异，逐步滤除非地面点。王慧等针对Keqi Zhang方法在内插误差的不足，提出一种扫描线的数学形态学LiDAR点云滤波方法，并对扫描折线问题进行有效处理。
吴坚等先采用RANSAC算法粗提取地面点，再采用多特征、宽阈值、分层次的方法提取道路边界特征点，最后应用RANSAC方法拟合道路边界曲线模型。该方法算法复杂，阈值设置项众多，难以适应不同复杂场景的需求。史文中等基于地面点信息和扫描线分离条件确定扫描线，再从扫描线中提取道路边缘信息。该方法对于道路边缘被遮挡的场景表现不佳。
现有研究大多考虑到了点云间的局部特征，通过迭代窗口尺寸和定义高差阈值滤除地面点，但对于大型地物或地物遮挡此类方法表现不佳。同时，由于需要搜索邻近点导致计算量很大，效率不高。
因此，有必要开发一种基于扫描线的道路点云的提取方法，能够对城市道路、公路路面范围进行识别，同时可以解决地形起伏、大型地物遮挡等造成的道路提取不连续等问题，提高道路识别效率和正确率。
发明内容
本发明要解决的技术问题是提供一种基于扫描线的道路点云的提取方法，能够对城市道路、公路路面范围进行识别，同时可以解决地形起伏、大型地物遮挡等造成的道路提取不连续等问题，提高道路识别效率和正确率。
为了解决上述技术问题，本发明采用的技术方案是：该基于扫描线的道路点云的提取方法，具体包括以下步骤：
S1选取起始点：从车载点云中提取每条扫描线的航线边缘点作为扫描线的起始点；
S2判断地面点或地物点：以所述步骤S1中每条扫描线的起始点为基础进行搜索，根据该扫描线的起始点与其相邻点之间的高差，区分该相邻点是地面点或地物点；
S3沿扫描线搜索地面点：若相邻点是地面点，则将相邻点加入该扫描线地面点集合；再重复步骤S2，继续判断下一个相邻点是地面点或地物点；若相邻点是地物点，则以该扫描线上已有的地面点构建空间直线方程，并判断相邻点的下一个相邻点是地物点或地面点，直至遍历至下一条扫描线的起始点结束；
S4过滤地面点：将每条扫描线上的地面点进行聚类，再过滤掉小于设定长度阈值的地面点，从而得到去除噪声的连贯的扫描线；
S5获得道路边界：以所述步骤S4获得的去除噪声的连贯的扫描线为基础，选取每条扫描线的左端点和右端点，拟合出道路左右两侧的边界，完成道路边界的提取和规则化；
所述步骤S2判断该相邻点是地面点或地物点的具体步骤为：以扫描线的起始点为基础搜索左右两个方向，判断相邻点与起始点之间的高差H ij是否小于高差阈值H t，若小于高差阈值H t则判为地面点，并将该相邻点加入该扫描线的地面点集合；若高于高差阈值H t则判为地物点，其中高差H ij的计算公式为：
H ij＝|h j–h i|；
其中h j为待判断点即相邻点j的高，h i为当前地面点i的高。
采用上述技术方案，步骤S2中根据判断该扫描线的起始点与其相邻点之间的高差是否小于设定的阈值，从而判断该相邻点是地面点或地物点；从而通过地面点构建地面点集合，再对地面点进行聚类和去噪，获得连贯的扫描线，再经过拟合得到道路两侧的边界，完成道路边界的提取，解决地形起伏、大型地物遮挡等造成的道路提取不连续等问题，提高道路识别效率和正确率。
作为本发明的优选技术方案，所述步骤S3中若相邻点为地物点时，则以该扫描线已检测到的地面点集合构建在OXY空间和OYZ空间的直线方程y＝k*x+b和z＝m*y+n，并求解参数k、b、m和n，将下一个相邻点即待判断点p的坐标(x p，y q)代入方程求解直线上的点q的坐标(y q，z q)；计算待判断点p和直线上点q在y方向和z方向的差值Δy＝|y q-y p|、Δz＝|z q-z p|，若Δy小于阈值Y t且Δz小于高差阈值H t，则该待判断点p判为地面点，否则判为地物点。
作为本发明的优选技术方案，所述步骤S4中采用区域生长算法对每条扫描线上的地面点进行聚类，过滤掉小于设定长度阈值的地面点的具体步骤为：
S41：首先在每条扫描线的地面点集合中随机选择一个点作为种子点，判断其与周边相邻点的距离，若小于设定的距离阈值D t，则认为该相邻点属于一个聚类，并对其进行标记；再以已标记的相邻点为种子点重复搜索其周边相邻点，直至周边没有相邻点在距离阈值D t范围内；若该扫描线还有未被标记的点，则再随机选择一个点作为种子点，重复上述操作直至该扫描线所有点都被标记；
S42：然后设定长度阈值D s，按标记的点计算地面点的聚类长度，若小于设定长度阈值D s，则将小于长度阈值D s的点从地面点集合中剔除。
作为本发明的优选技术方案，所述步骤S5中选取每条扫描线的左端点和右端点，按照左右两个方向分别采用最小二乘算法拟合出道路左右两侧的边界，拟合时按照汽车行驶方向按间距DS进行分段拟合，以准确提取弧形道路的边界。
作为本发明的优选技术方案，所述航线边缘点存储于LAS文件中，其值用1表示，代表上一次扫描线的结束和下一次扫描线的开始。因此，提取LAS文件中所有航线边缘点值为1的点作为每条扫描线的起始点。
与现有技术相比，本发明具有的有益效果为：该基于扫描线的道路点云的提取方法在扫描线的基础上采用一维搜索，有效避免了二维乃至三维搜索在大数据量场景下的低效率问题，而且在地形起伏较大的地区或坡路路段也表现优异，对于有大型地物遮挡区域算法也能鲁棒的提取道路边界。
附图说明
图1为本发明的基于扫描线的道路点云的提取方法的流程图；
图2为本发明的基于扫描线的道路点云的提取方法的试验区原始数据图；
图3为本发明的基于扫描线的道路点云的提取方法的试验区地面点地物点区分结果图；
图4为本发明的基于扫描线的道路点云的提取方法的提取的地面点图；
图5为本发明的基于扫描线的道路点云的提取方法的提取的道路边界线图。
具体实施方式
下面将结合本发明的实施例图中的附图，对本发明实施例中的技术方案进行清楚、完整的描述。
实施例：如图1所示，该基于扫描线的道路点云的提取方法，具体包括以下步骤：
S1选取起始点：图2为本发明的试验区数据。从车载点云中提取每条扫描线的航线边缘点作为扫描线的起始点；所述航线边缘点存储于LAS文件中，其值用1表示，代表上一次扫描线的结束和下一次扫描线的开始；提取LAS文件中所有航线边缘点值为1的点作为每条扫描线的起始点；
S2判断地面点或地物点：以所述步骤S1中每条扫描线的起始点为基础进行搜索，判断该扫描线的起始点与其相邻点之间的高差是否小于设定的阈值，从而区分该相邻点是地面点或地物点；
所述步骤S2判断该相邻点是地面点或地物点的具体步骤为：以扫描线的起始点为基础搜索左右两个方向，判断相邻点与起始点之间的高差H ij是否小于高差阈值H t，若小于高差阈值H t则判为地面点，并将该相邻点加入该扫描线的地面点集合；若高于高差阈值H t则判为地物点，其中高差H ij的计算公式为：
H ij＝|h j–h i|；
其中h j为待判断点即相邻点j的高，h i为当前地面点i的高；
所述高差阈值H t由路面平整度决定；通过实验发现，若高差阈值H t设定为L，地面点易被错判为地物点，加大了算法的复杂度；若高差阈值H t设定为3L，地物点易被错判为地面点；若高差阈值H t设定为2L，地面点的提取比较符合实际情况；因此，高差阈值的设定直接关系地面点的初步提取效果，具体阈值数值主要和相邻点间距L有关，由路面平整度决定。
S3沿扫描线搜索地面点：若相邻点是地面点，则将相邻点加入该扫描线地面点集合；再重复步骤S2，继续判断下一个相邻点是地面点或地物点；若相邻 点是地物点，则以该扫描线上已有的地面点构建空间直线方程，并判断相邻点的下一个相邻点是地物点或地面点，直至遍历至下一条扫描线的起始点结束；
所述步骤S3中若相邻点为地物点时，则以该扫描线已检测到的地面点集合构建在OXY空间和OYZ空间的直线方程y＝k*x+b和z＝m*y+n，并求解参数k、b、m和n，将下一个相邻点即待判断点p的的坐标(x p，y q)代入方程求解直线上的点q的坐标(y q，z q)；计算待判断点p和直线上点q在y方向和z方向的差值Δy＝|y q-y p|、Δz＝|z q-z p|，若Δy小于阈值Y t且Δz小于高差阈值H t，则该待判断点p判为地面点，否则判为地物点；通过实验发现，若Y方向阈值Y t设定为L，有较多的地面点被错判为地物点；若Y方向阈值Y t设定为2L，仍有部分地面点被错判为地物点；若Y方向阈值Y t设定为3L，地面点的提取比较符合实际情况(见图3、图4)；若Y方向阈值Y t设定为4L，地物点易被判为地面点；高差阈值H t与步骤S2中一致；
S4过滤地面点：将每条扫描线上的地面点进行聚类，再过滤掉小于设定长度阈值的地面点，从而得到去除噪声的连贯的扫描线；
所述步骤S4中采用区域生长算法对每条扫描线上的地面点进行聚类，过滤掉小于设定长度阈值的地面点的具体步骤为：
S41：首先在每条扫描线的地面点集合中随机选择一个点作为种子点，判断其与周边相邻点的距离，若小于设定的距离阈值D t，则认为该相邻点属于一个聚类，并对其进行标记；再以已标记的相邻点为种子点重复搜索其周边相邻点，直至周边没有相邻点在距离阈值D t范围内；若该扫描线还有未被标记的点，则再随机选择一个点作为种子点，重复上述操作直至该扫描线所有点都被标记；
S42：然后设定长度阈值D s，按标记的点计算地面点的聚类长度，若小于设定长度阈值D s，则将小于长度阈值D s的点从地面点集合中剔除；若距离阈值D t设定为L，会有椒盐现象，导致聚类过于分散；若距离阈值D t设定为2L，聚类效果较好，满足后续处理要求；
S5获得道路边界：以所述步骤S4获得的去除噪声的连贯的扫描线为基础，选取每条扫描线的左端点和右端点，拟合出道路左右两侧的边界，完成道路边界的提取和规则化；
所述步骤S5中选取每条扫描线的左端点和右端点，按照左右两个方向分别采用最小二乘算法拟合出道路左右两侧的边界。拟合时按照汽车行驶方向按间 距DS进行分段拟合，可有效解决弧形道路边界提取不准确的情况；间距DS可根据实际情况选取，建议选取1m作为间距DS的阈值，提取的道路边界如图5。
以上所述仅为本发明的较佳实施例而已，并不用于限制本发明，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (5)

  1. 一种基于扫描线的道路点云的提取方法，其特征在于，具体包括以下步骤：
    S1选取起始点：从车载点云中提取每条扫描线的航线边缘点作为扫描线的起始点；
    S2判断地面点或地物点：以所述步骤S1中每条扫描线的起始点为基础进行搜索，根据该扫描线的起始点与其相邻点之间的高差，区分该相邻点是地面点或地物点；
    S3沿扫描线搜索地面点：若相邻点是地面点，则将相邻点加入该扫描线地面点集合；再重复步骤S2，继续判断下一个相邻点是地面点或地物点；若相邻点是地物点，则以该扫描线上已有的地面点构建空间直线方程，并判断相邻点的下一个相邻点是地物点或地面点，直至遍历至下一条扫描线的起始点结束；
    S4过滤地面点：将每条扫描线上的地面点进行聚类，再过滤掉小于设定长度阈值的地面点，从而得到去除噪声的连贯的扫描线；
    S5获得道路边界：以所述步骤S4获得的去除噪声的连贯的扫描线为基础，选取每条扫描线的左端点和右端点，拟合出道路左右两侧的边界，完成道路边界的提取和规则化；所述步骤S2判断该相邻点是地面点或地物点的具体步骤为：以扫描线的起始点为基础搜索左右两个方向，判断相邻点与起始点之间的高差H ij是否小于高差阈值H t，若小于高差阈值H t则判为地面点，并将该相邻点加入该扫描线的地面点集合；若高于高差阈值H t则判为地物点，其中高差H ij的计算公式为：
    H ij＝|h j–h i|；
    其中h j为待判断点即相邻点j的高，h i为当前地面点i的高。
  2. 根据权利要求1所述的基于扫描线的道路点云的提取方法，其特征在于，所述步骤S3中若相邻点为地物点时，则以该扫描线已检测到的地面点集合构建在OXY空间和OYZ空间的直线方程：
    y＝k*x+b；
    z＝m*y+n；
    k、m为直线的斜率，b、n为直线的截距；
    并求解参数k、b、m和n，将下一个相邻点即待判断点p的坐标(x p，y q)代入方程求解直线上的点q的坐标(y q，z q)；计算待判断点p和直线上点q在y方 向和z方向的差值Δy＝|y q-y p|、Δz＝|z q-z p|，若Δy小于阈值Y t且Δz小于高差阈值H t，则该待判断点p判为地面点，否则判为地物点。
  3. 根据权利要求2所述的基于扫描线的道路点云的提取方法，其特征在于，所述步骤S4中采用区域生长算法对每条扫描线上的地面点进行聚类，过滤掉小于设定长度阈值的地面点的具体步骤为：
    S41：首先在每条扫描线的地面点集合中随机选择一个点作为种子点，判断其与周边相邻点的距离，若小于设定的距离阈值D t，则认为该相邻点属于一个聚类，并对其进行标记；再以已标记的相邻点为种子点重复搜索其周边相邻点，直至周边没有相邻点在距离阈值D t范围内；若该扫描线还有未被标记的点，则再随机选择一个点作为种子点，重复判断其与周边相邻点的距离是否在阈值范围内，直至该扫描线所有点都被标记；
    S42：然后设定长度阈值D s，按标记的点计算地面点的聚类长度，若小于设定长度阈值D s，则将小于长度阈值D s的点从地面点集合中剔除。
  4. 根据权利要求2所述的基于扫描线的道路点云的提取方法，其特征在于，所述步骤S5中选取每条扫描线的左端点和右端点，按左右两侧分别顺次连接道路端点拟合道路左右两侧的边界，拟合时按照汽车行驶方向按间距DS进行分段拟合，以准确提取弧形道路的边界。
  5. 根据权利要求2所述的基于扫描线的道路点云的提取方法，其特征在于，所述航线边缘点存储于LAS文件中，其值用1表示，代表上一次扫描线的结束和下一次扫描线的开始。
PCT/CN2021/084078 2020-12-07 2021-03-30 一种基于扫描线的道路点云的提取方法 WO2022121177A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP21901893.4A EP4120123A4 (en) 2020-12-07 2021-03-30 STREET POINT CLOUD EXTRACTION METHOD BASED ON SCANLINES

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011415577.0 2020-12-07
CN202011415577.0A CN112200171B (zh) 2020-12-07 2020-12-07 一种基于扫描线的道路点云的提取方法

Publications (1)

Publication Number Publication Date
WO2022121177A1 true WO2022121177A1 (zh) 2022-06-16

Family

ID=74033845

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/084078 WO2022121177A1 (zh) 2020-12-07 2021-03-30 一种基于扫描线的道路点云的提取方法

Country Status (3)

Country Link
EP (1) EP4120123A4 (zh)
CN (1) CN112200171B (zh)
WO (1) WO2022121177A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117173424A (zh) * 2023-11-01 2023-12-05 武汉追月信息技术有限公司 一种点云坡面边缘线识别方法、系统及可读存储介质
CN117475002A (zh) * 2023-12-27 2024-01-30 青岛亿联建设集团股份有限公司 基于激光扫描技术的建筑倾斜度测量方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112200171B (zh) * 2020-12-07 2021-03-23 速度时空信息科技股份有限公司 一种基于扫描线的道路点云的提取方法
CN112862844B (zh) * 2021-02-20 2024-01-05 园测信息科技股份有限公司 基于车载点云数据的道路边界交互式提取方法
CN113762067B (zh) * 2021-07-21 2024-03-26 上海圭目机器人有限公司 一种机场板块的识别方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012056255A1 (en) * 2010-10-25 2012-05-03 Seraphim Amvrazis Method of mapping and control of surfaces of tunnels during the construction project
CN103778429A (zh) * 2014-01-24 2014-05-07 青岛秀山移动测量有限公司 一种车载激光扫描点云中道路信息自动提取方法
CN104197897A (zh) * 2014-04-25 2014-12-10 厦门大学 一种基于车载激光扫描点云的城区道路标线自动分类方法
CN108062517A (zh) * 2017-12-04 2018-05-22 武汉大学 基于车载激光点云的非结构化道路边界线自动提取方法
CN109684921A (zh) * 2018-11-20 2019-04-26 吉林大学 一种基于三维激光雷达的道路边界检测与跟踪方法
CN112200171A (zh) * 2020-12-07 2021-01-08 速度时空信息科技股份有限公司 一种基于扫描线的道路点云的提取方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8825260B1 (en) * 2013-07-23 2014-09-02 Google Inc. Object and ground segmentation from a sparse one-dimensional range data
US10031231B2 (en) * 2016-09-12 2018-07-24 Delphi Technologies, Inc. Lidar object detection system for automated vehicles
TWI652449B (zh) * 2017-12-11 2019-03-01 財團法人車輛研究測試中心 三維感測器之動態地面偵測方法
CN110378173A (zh) * 2018-07-10 2019-10-25 北京京东尚科信息技术有限公司 一种确定道路边界线的方法和装置
CN109741450B (zh) * 2018-12-29 2023-09-19 征图三维（北京）激光技术有限公司 一种基于扫描线的路面点云自动提取方法及装置
CN110598541B (zh) * 2019-08-05 2021-07-23 香港理工大学深圳研究院 一种提取道路边缘信息的方法及设备
CN111208495A (zh) * 2020-02-28 2020-05-29 燕山大学 一种基于激光雷达点云特征线与平面校准的地面提取方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012056255A1 (en) * 2010-10-25 2012-05-03 Seraphim Amvrazis Method of mapping and control of surfaces of tunnels during the construction project
CN103778429A (zh) * 2014-01-24 2014-05-07 青岛秀山移动测量有限公司 一种车载激光扫描点云中道路信息自动提取方法
CN104197897A (zh) * 2014-04-25 2014-12-10 厦门大学 一种基于车载激光扫描点云的城区道路标线自动分类方法
CN108062517A (zh) * 2017-12-04 2018-05-22 武汉大学 基于车载激光点云的非结构化道路边界线自动提取方法
CN109684921A (zh) * 2018-11-20 2019-04-26 吉林大学 一种基于三维激光雷达的道路边界检测与跟踪方法
CN112200171A (zh) * 2020-12-07 2021-01-08 速度时空信息科技股份有限公司 一种基于扫描线的道路点云的提取方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117173424A (zh) * 2023-11-01 2023-12-05 武汉追月信息技术有限公司 一种点云坡面边缘线识别方法、系统及可读存储介质
CN117173424B (zh) * 2023-11-01 2024-01-26 武汉追月信息技术有限公司 一种点云坡面边缘线识别方法、系统及可读存储介质
CN117475002A (zh) * 2023-12-27 2024-01-30 青岛亿联建设集团股份有限公司 基于激光扫描技术的建筑倾斜度测量方法

Also Published As

Publication number Publication date
EP4120123A1 (en) 2023-01-18
CN112200171A (zh) 2021-01-08
EP4120123A4 (en) 2023-08-30
CN112200171B (zh) 2021-03-23

Similar Documents

Publication Publication Date Title
WO2022121177A1 (zh) 一种基于扫描线的道路点云的提取方法
CN105184852B (zh) 一种基于激光点云的城市道路识别方法及装置
EP3171292B1 (en) Driving lane data processing method, device, storage medium and apparatus
WO2018068653A1 (zh) 点云数据处理方法、装置及存储介质
WO2018133851A1 (zh) 一种点云数据处理方法、装置及计算机存储介质
CN112801022B (zh) 一种无人驾驶矿卡作业区道路边界快速检测更新的方法
CN111487641B (zh) 一种利用激光雷达检测物体的方法、装置、电子设备及存储介质
CN111079611B (zh) 一种道路面及其标志线的自动提取方法
CN109584294B (zh) 一种基于激光点云的路面点云提取方法和装置
Huang et al. On-board vision system for lane recognition and front-vehicle detection to enhance driver's awareness
CN108898672A (zh) 一种制作三维高清道路图车道线的半自动点云方法
CN101916373B (zh) 基于小波检测和脊线跟踪的道路半自动提取方法
CN114812581A (zh) 一种基于多传感器融合的越野环境导航方法
CN111881790A (zh) 一种高精度地图制作中道路人行横道自动化提取方法和装置
CN110363771B (zh) 一种基于三维点云数据的隔离护栏形点提取方法及装置
JP4940177B2 (ja) 交通流計測装置
WO2023216470A1 (zh) 一种可行驶区域检测方法、装置及设备
CN102982304A (zh) 利用偏光图像检测车辆位置的方法和系统
JP2012255703A (ja) 道路勾配推定装置
Jiang et al. Effective and robust corrugated beam guardrail detection based on mobile laser scanning data
KR101910256B1 (ko) 카메라 기반 도로 곡률 추정을 위한 차선 검출 방법 및 시스템
CN112435336B (zh) 一种弯道类型识别方法、装置、电子设备及存储介质
Hu et al. LiDAR-based road extraction for UGV in high definition map
CN114170579A (zh) 一种路沿检测方法、装置及汽车
CN109583417B (zh) 基于fcw系统的前方车辆候选框提取方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21901893

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021901893

Country of ref document: EP

Effective date: 20221003

NENP Non-entry into the national phase

Ref country code: DE