Skip to content

Home

Jump to bottom
Islam Mansour edited this page Jun 23, 2019 · 1 revision

Welcome to the LiDAR-PointClouds-Classification wiki!

This section aims to summarize the main LiDAR data handling techniques commonly used to calculate the high variety of LiDAR metrics.

Schematic grouping of LiDAR data handling techniques

  • Single point based (like amplitude)

  • Point neighborhood based (based on defined searching criteria)

    • Point cloud statistics

    • Normals and related properties

  • Raster based

    • Without interpolation (statistics)

      • Cell based

      • Subset related

        • Percentiles
        • Voxel

    • With interpolation (surface)

      • TIN

      • Other interpolation (IDW, kriging etc.)

      • Subset related

        • Percentiles

        • Voxel

  • Object based

Commonly used derivatives

  • Descriptive statistics: Minimum, Maximum, Mean, Median, Sum, Number of points, Percentiles, Deciles, Standard Deviation, Variance, Root mean square, Minority, Majority, Skewness, Kurtosis, Coefficient of variation

  • Filter criteria: Return number, Number of return (first, last, all echoes)

  • Feature type: geometric (using only the x,y,z information), radiometric (amplitude, echo width, backscatter cross section, backscattering coefficient, reflectance)

Point neighborhood-based

  • Point cloud statistics

One way to calculate (geometric) features are using a defined neighborhood to calculate additional attributes which represent the environment. The search criteria can be a sphere, limited cylinder or unlimited cylinder. The searching criteria are applied on every single point. Within the searching criteria, it is possible for example to derive descriptive statistics value and assigned to each point. This is a computation intensive task.

  • Calculate normal vectors and eigenvalues

Point cloud based features are often using normal vectors and normal vector related extra properties. Normal vectors can be calculated and assigned to every point using the neighborhood points within a defined radius. Furthermore, we can also compute eigenvalues and eigenvectors based on the covariance matrix of the points within the neighborhood. Based on the sorted eigenvalues further features can be derived: linearity, planarity, sphericity, anisotropy, eigen-entropy, curvature etc. Weinmann et al., 2014

Raster based

Without interpolation

In the case of cell-based features, the statistical value is derived from a grid cell. Interpolation technique is not used, the pixel value is based on the statistical value of points within cells. This is a less computational intensive solution.

  • Subset of cells

Within the defined cells the features can be calculated assigned to different height layers (vertical distribution of the point cloud) or can be calculated for voxels which determine the features within equal 3D cubes.

Interpolation based

This methodology is commonly used for deriving DTM and DSM. With the interpolation, the empty cells are interpolated based on the neighbors' value. For the interpolation, two main methodologies exist the grid-based and TIN based.The TIN based approach using every existed point to define the triangles. A grid-based approach can be defined various interpolation technique which known from geostatistics methods (nearest neighbor, IDW, kriging, etc.).

Clone this wiki locally