PotreeLoaderCpp

This project provides functionality to iterate through point cloud octree data generated for the Potree renderer

Overview

"PotreeConverter generates an octree LOD structure for streaming and real-time rendering of massive point clouds."
These octree structures can be loaded and viewed using Potree which is mainly programmed using JavaScript.
This project aims to make the Potree file and data structure accessible using a simple header-only C++ library.

Features:

• Header-only C++ library (no external dependencies except stl)
• Read Potree Octree hierarchy and metadata like point attributes, offset, scale, ...
• (Conditionally) Traverse octree nodes containing points per cube, bounding boxes, spacing, ...
• Load node points from octree data on disk using the data loader

How to use

• Create an Octree object using metadata file as argument
• Initiallize data loader
• Get scale, offset and the index of the position coordinates in data
• Traverse nodes and load their respective point data
• Extract your desired point attributes

// A PointCloudItem with XYZ coordinates
struct PointCloudItem {
	double x, y, z;
};

/*----------- Directory path to PoTree converted data -------------------------*/
auto current_path = fs::current_path();
string file_searchpath = (current_path / "sample_data" / "sparse_junction").string();

/*----------- Initialize Octree class with PoTree metadata --------------------*/
auto octreeFiles = octree_files::SearchOctreeFiles(file_searchpath);
string metafile(octreeFiles["metadata"]);
Octree octree(metafile);

/*----------- Extract data with Octree Loader Class ---------------------------*/
OctreeLoader loader(&octree);
auto nodeData = loader.CreateMaxNodeData();

const int bytesPerPoint = octree.geometry.pointAttributes.bytes;
const auto scale        = octree.geometry.pointAttributes.posScale;
const auto offset       = octree.geometry.pointAttributes.posOffset;
const int posIndex     = loader.pcloud_byte_offsets.xyz;

std::vector<PointCloudItem> cloudpoints;
cloudpoints.reserve(octree.points);

/*----------- Extract data with Octree Loader Class ----------------------------*/
// Traverse Octree Nodes
octree.geometry.nodes[0]->traverse(
	[maxLevel, &loader, &nodeData, bytesPerPoint, scale, offset, posIndex, &cloudpoints](OctreeGeometryNode* node, int level) {

        // Load raw node data
        auto& data = loader.LoadNodeData(node, nodeData);
	size_t pointCountInBuffer = node->byteSize / bytesPerPoint;
	uint8_t* pBuffer = data.data();

        // Extract coordinates, apply scale plus offset and push to cloudpoints
	for (int i = 0; i < pointCountInBuffer; ++i)
	{
	     const int offsetToPointStart = i * bytesPerPoint;

             double x = std::fma(static_cast<double>(*reinterpret_cast<int32_t*>(pBuffer + offsetToPointStart + posIndex)), scale.x, offset.x);
             double y = std::fma(static_cast<double>(*reinterpret_cast<int32_t*>(pBuffer + offsetToPointStart + posIndex + sizeof(int32_t))), scale.y, offset.y);
             double z = std::fma(static_cast<double>(*reinterpret_cast<int32_t*>(pBuffer + offsetToPointStart + posIndex + 2 * sizeof(int32_t))), scale.z, offset.z);

             cloudpoints.emplace_back(x, y, z);
	}
	});

How to build sample

For an example how to additionally load point color data, build the sample source.

• Open the provided Visual Studio solution with VS2017 or newer (VS2022 is recommended)
• Select your target configuration and hit build
• Run PoTreeLoader.exe Running the program with the debugger in Debug mode is recommended. Set breakpoints to see how the code operates.

Credits

Due to the nature of working with a predefined data structure, some of the code is a direct translation of the Potree project source

• Markus Schütz and contributors of Potree and PotreeConverter