[WIP] Add documentation for new voxel grid design

Author: tin1254

doc/tutorials/content/grid_filters.rst

@@ -0,0 +1,69 @@
+.. _grid_filters:
+
+Design of VoxelGrid code structure
+----------------------------------
+
+This document is intended to explain our new ``VoxelGrid`` design and help you implementing your own grid based filters easily by reusing our implementations.
+
+Inheritance structure of VoxelGrid
+----------------------------------
+
+.. image:: images/voxel_grid_code_structure.jpg
+    :height: 150px
+    :align: center 
+
+Structure explanation
+---------------------
+
+We will explain the pieces composing ``VoxelGrid``, then elaborate their purpose and demonstrate how they can be reused to implement your custom grid filters.
+
+In a high level view, ``VoxelGrid`` is comprised from several important classes:
+
+1. ``VoxelFilter``
+2. ``VoxelStruct``
+3. ``Voxel``
+4. ``CartesianFilter``
+5. ``TransformFilter``
+
+
+``VoxelFilter<VoxelStruct<Voxel<PointT>,PointT>>`` is aliased as ``VoxelGrid<PointT>`` in our source code, it implements the API specified for ``VoxelGrid``.
+Apart from ``PointT``, it also accepts a class ``GridStruct`` as template argument.
+
+``GridStruct`` should implement the filtering logic of a grid based filter, e.g., how to group points and add them to the same grid cell (voxel in ``VoxelGrid``), 
+which grid cells are exported and how to export them to the output point cloud. 
+They are several requirements on the member functions of ``GridStruct``, below are their requirements and their intended usage:
+
+.. note::
+
+  1. ``size_t size()``
+      - return the number of the non-empty grid cells
+  2. ``begin()`` and ``end()``
+      - used for iterating through the grid cells
+      - can return any type which can be incremented and compared, e.g., iterator, index
+  3. ``bool setUp(CartesianFilter<FilterBase, GridStruct>&)``
+      - initialize the member variables needed for filtering
+      - input point cloud and all settings of CartesianFilter can be accessed from the reference
+  4. ``void addPointToGrid(PointT&)``
+      - add a single point to a grid cell 
+  5. ``pcl::experimental::optional<PointT> filterGrid(...)``
+      - accepts output from ``begin()`` and ``end()``
+      - decide whether the iterating grid cell is eligible to be output
+      - export a point representing the iterating grid cell
+
+``VoxelStruct<VoxelT, PointT>`` is passed as ``GridStruct`` in our ``VoxelGrid``. It accepts a class ``VoxelT``, which is ``Voxel<PointT>`` in our ``VoxelGrid``. 
+``VoxelT`` is intented to store the information of a single grid cell (voxel), e.g., number of points in the cell, mean or covariance of the points.
+
+In ``VoxelFilter``, ``GridStruct`` is passed down again as template argument to ``CartesianFilter`` then to ``TransformFilter``. 
+``CartesianFilter`` simplily implements the common API for our grid based filters. 
+Note that we can pass either ``pcl::FilterIndices`` (for binary removal) or ``pcl::Filter`` (for another purposes) to ``FilterBase``.
+
+``TransformFilter`` implements the three main phases of grid based filtering:
+
+1. Initialize the filter given the input point cloud and filtering settings
+2. Iterate through all points in the input cloud and add them to grid cells one by one
+3. Iterate through all the non-empty grid cells and output a point based of the information of the grid cells
+
+It can easily be reused for other filters, as long as a filter has the similar logic of collecting points into smaller groups and transforming each group to a point.
+
+Example usage
+-------------