GitLab

[FIX #373] Due to the update change of #373 there was a minor update bug: When new data was connected, no update took place.

 * This approach describes the surface detection of Zahra Lari and Ayman Habib (2014).
 * Paper: "An adaptive approach for the segmentation and extraction of planar
 * andlinear/cylindrical features from laser scanning data", ISPRS Journal of
 * Photogrammetry and Remote Sensing, 2014
 *
 * Eventually the process covers also the linear and cylindrical feature segmentation
 * but in our requirements only the planar segmentation is required. Steps:
 *   - Original laser point cloud as input data
 *   - PCA-based classification of planar features
 *   - Selection of appropriate representation models for the detected planar features
 *   - Local point density estimation along the detected planar features
 *   - Precise estimation of the planar features segmentation attributes
 *   - Parameter-domain segmentation and extraction of planar features
 *   - Boundary tracking for resolving the parameter-domain segmentation ambiguities of
 *     planar features
 *   - Extracted planar features in spatial domain
 *
 * Estimations (Based on the paper because I haven't completely impelemented yet):
 *   - Fair segmentation results
 *   - Relatively slow
 *
 * Implementation progress:
 *   - Done to the step "Precise estimation of the planar features segmentation
 *     attributes".
 *
 *
 * Other updates:
 *   - Renamed module "Pooints - Crop" to "Points - Transform". It optained the feature
 *     to translate and to rotate the point set.
 *   - Wew code refactoring.

 * Currently the module Surface detection by PCL only gives the ability to detect
 * surfaces using point clouds (WDataSetPoints) what is a method of a fair quality.
 *
 * Some few changes across other modules:
 *   - Improvement on the octree neighbor search algorithm. Now you can look for
 *     neighbors using the neighborship of 6, 18 or 26
 *   - Very first steps on the surface detection algorithm "An adaptive approach for the
 *     segmentation and extraction of planar and linear/cylindrical features from laser
 *     scanning data" of Lari/Habib (2014).
 *   - Added an unidimensional kd tree structures with some processing methods such as
 *     looking for nearest nghbors.

 * I changed in the WPrincipalComponentAnalysis analyzeData(const vector<WPosition>) to
 * analyzeData(const vector<WPosition>*) in order to avoid unnecessary copies. Therefore
 * some classes had to be changed to remain compilable.

 * Added a new outline method to the wall detection. Voxels or properties of its
 * input point data can be displayed using rhombs. They display the wall node group,
 * Eigen Vectors, Eigen Values and the mean coordinate.
 *
 * Applied fixes:
 * The wall detection failed by bleeding through several surfaces. The problem was
 * that no angle condition was applied during merging wall groups.
 * Also small cleanups were applied on the code.

 * The algorithm is intended to grouup voxels that have surfaces with the same
 * normal vector. This feature helps to remove small groups. Obviously trees
 * by nature have points that lead the Principal Component Analysis to
 * surfaces that have very different Eigen Vectors of the smallest Eigen Value
 * that is the normal vector in the case of surfaces.
 *
 * Other feature: Removing voxels by a point count per voxel.

 * The new module firstly puts point data into a voxel grid. For each of these areas
 * the Principal Component Analysis is run. For each voxel the Eigen Vector of the
 * smallest Eigen Value is taken. In the case of surface kins data it's the normal
 * vector. In the last steps voxels with the similar vector are grouped.
 * Currently the region growing often bleeds over through many surfaces.

 * This algorithm puts input point coulds to a voxel grid and applies the Principal
 * Component Analysis to each of these areas. It's calculated how isotropic voxels are.
 * Either planar voxels have the blue color and isotropic ones are red. There's a
 * feature to cut avay voxels from the output triangle mesh for the view. You can either
 * set the minimal point count per voxel or the quotient of the smallest Eigen value over
 * the biggest one as the maximal isotropic level criteria.

[CHANGE] Many improvements over time. Outlining, Stratified and regular samplers, shader improvements and shader debugging code,...

[ADD] Global Boy coloring, as well as custom colors in StartEnd Coloring. This enables illumination for singe color as well.

Following bugfixes new features:
  - Points group selector plugin: Can output selected buildings as
    WDataSetPointsGrouped in order to be able to view data in other plug-ins
  - Elevation image export plugin: Features to switch elevation display
    in triangle mesh height and color.
  - Fixes in documentation and few code cleanup.

The buildings detection plugin generally was a mix of many things. These things of
it were split up into these plutins:
  - Elevation image export
  - Points group selector which transforms building groups into a voxel structure
    and exports it to a triangle mesh.

Added features:
  - Elevation image can be output in a triangle mesh
  - Elevation image (file and triangle mesh output) and building outline got colored
    building outline

Code style:
  - Fixed prooblems with documentation (make doc)
  - Purged last code style (make stylecheck)

[FIX] RGBA used an invalid alpha for lineAO, so as we do not need it here, we just use RGB for coloring the start-end.

[ADD] New start-end color coding for fibers. Currently only red-white is possible and hard coded as I am not aware of any colorscales as Properties. It would be possible to define one, but this is deferred to the time when really needed.

Some simplistic algorithms are done in order to group buildings. Evaluation:
  + Important data structures as Octree and Quadtree are implemented with important functions
  + Grouping neighbor voxels runs even if still not 26-neighborship (few steps required)
     + Algorithm can easily be transferred to the quadtree set.
  + First new plugin for cutting of outlier points using the algorithm that calculates
    figuring out cube connectivity. Other points are cut off that don't belong to the
    largest voxel group
  - Possibly vaporized effort
      - The algorithm that takes relative elevation minimums has problems detecting low buildings.
          - Probably the algorithm won't be taken
          + The great deal of the code amount used for it can be very useful for the final code.
      + It's easy to differ trees from buildings
          - Not thought to the end to differ trees that are very close to buildings

 * This is a first important intermediate step for making sure having a
 * consistent data structure that has an elevation image which has an
 * important role in building detection.

Removing a file that I've created by mistake

Done few very first steps for building detection:
  - Areas that contain data set points, are outlined by octree nodes
  - Ability to draw that outline as boxes in order to be able to draw very
    huge las files at all.

Just started very first building detection implementation steps:
  - Transforming point data to octree nodes
  - Ability to display LAS data as boxes, but without color intensity display.
    Now you don't need vast amount of RAM (32GB for a 500MB las file) and GPU
    memory to be able to display larger las files at all.