winter's PCL-Trimble Code Sprint， PCL Code Sprint 的主要方向

1. Automated decimation: automatically reduce the number of points in the point cloud based upon the definition of a feature (i.e., not a generic spatial sampling). For example reduce the number of points used to define simple surfaces, such as a wall, cylinder, etc. while retaining sufficient points to define all surfaces and edges. The purpose of this project is to make the data size smaller to allow for improved performance with large data sets / streaming over the Internet.
2. Automated noise filtering: clean parasitic points, such as measurement noise, interruptions (objects moving across scanner), vegetation (primarily grass patches to identify land surfaces). PCL already contains tools for removing registration occlusions and outliers, but we are removing edges as well. The purpose of this project is to deal with larger scenes, perform filtering, but retain edges.
3. Automated feature extraction based on type of object (i.e. pipes, steel structures, walls, valves, etc). For example identify all objects that are pipes and remove everything else so that the pipes can be examined and modelled in more detail using tools specific to the type of object.
4. Out-of-core point cloud database: develop an out-of-core database system for storing and interactively accessing massive point clouds (e.g. 100 TB to 1 PB) using techniques like hierarchical level-of-detail, out-of-core simplification, parallel database access, proxy scenegraphs, etc. Applications accessing the database would run on laptops and mobile devices with typically 1 GB of graphics memory.
5. Best-fit feature estimation: based on sensor characteristics (noise, biases, range-dependent errors) and data-collection geometry (sensor location and orientation relative to the objects in the point cloud), estimate best-fit for common geometry types (e.g. surfaces, spheres, cylinders, etc.). Identify all points likely to be part of the object, including potentially identifying outliers that are likely part of the object, but not used in the best-fit solution.
6. Change detection: automatically determine differences between objects scanned over time. For example scan a car or vessel on two occasions and automatically compare and highlight / report differences. This has to improve upon our current existing work on change detection and efficiently deal with large datasets.
7. Automated point cloud registration: improve the current set of registration/scan matching tools in PCL. The purpose of this project is to augment the new registration API (based on Correspondences) in PCL with new methods, and evaluate when to apply the right rejector/estimation method based on the input data. This has to improve upon our current existing work and deal efficiently with very large datasets.

   1。自动抽取：根据特征的定义（不是通用的空间采样）自动减少点云中的点的数量。例如定义简单的表面来减少点的数目，如墙壁，缸等，同时保留足够的点来定义所有的表面和边缘。这个项目的目的是用于互联网上大型数据集/流媒体时能缩小数据量，以便提高性能。
   2。自动噪声过滤：寄生干净点，如测量噪声，干扰（移动整个扫描仪中的对象），植被（主要是基层的修补程序，以确定土地表面）。 PCL已经包含删除登记闭塞和异常的工具，但将移除边缘。这个项目的目的是处理较大的场景，执行过滤，但保留边缘。
   3。基于自动特征提取对象的类型（如管道，钢结构，墙面，阀门等）。例如，找出所有的对象都管，并消除一切使管道可以使用特定的对象类型的工具，在更详细检查和建模。
   4。核心点云数据库：发展为核心的数据库系统存储和交互地访问大量的点云（例如100 TB为1 PB）使用像分层技术水平的细节，简化核心代理scenegraphs等应用程序访问数据库，并行数据库访问，通常为1 GB的绘图记忆体的笔记本电脑和移动设备上运行。
   5。最佳拟合功能估计：基于传感器的特性（噪音，偏见，范围依赖错误）和数据收集几何（点云对象的传感器的位置和方向），估计最适合常见的几何类型（如表面，球体，圆柱体等）。确定所有可能的对象的一部分，包括潜在确定离群值可能是对象的一部分，但不能用在最合适的解决方案。
   6。变化检测：自动判断随着时间的推移而扫描的对象之间的差异。例如，扫描两次的汽车或船只，自动比较突出/报告的差异。这已经改善我们现有的工作电流变化检测和有效地处理大型数据集。
   7。自动点云注册：提高登记/扫描工具，在PCL相匹配的电流。这个项目的目的是与新方法，以增加新的注册API（书信）在PCL和评估时，申请权抑制器/基于输入数据的估算方法。这已经改善我们目前现有的工作，并有效地处理具有非常大的数据集。