« Changelog History


PCL v0.9.9 Release Notes

Release Date: 2010-02-22 // about 14 years ago

    • [[pcl]]

      • removed ConvexHull2D (API breaking change!) You need to change your code. From: pcl::ConvexHull2D<...> ...; to pcl::ConvexHull<...> ...;
      • added a new general purpose 2D/3D ConvexHull class based on QHull
      • added a new general purpose 2D/3D ConcaveHull class based on QHull
      • added helper transformPointCloud method for cloud+indices as input
      • fixed: segfaults when ICP finds no correspondences (#4618)
      • improved the PCD I/O capabilities (for binary PCD) to deal with SSE padding
      • Added possibility to create a RangeImagePlanar from a point cloud
      • Corrected is_dense to false in all range images.
      • Reimplemented big parts of the NARF keypoint extraction - should be more reliable now (and unfortunately slower) - uses polynomials to search maxima now.
      • Added helper classes for polynomial approximations to common
      • Added a new RANSAC-like algorithm: PROSAC (much faster when there is a confidence for the matches)
      • Fixed normalization factor in the VFH's scale component.
      • Made MLS more flexible: output has all fields the input has, only with XYZ smoothed, and normals are provided separately (and optionally)
      • Added multi-scale calculation to NARF keypoint to make it faster again, fixed a bug in BorderExtractor and fixed some issues in RangeImagePlanar.
      • Added functions in common to compute max distance from a point to a pointcloud. Fixed distance component of VFH, normalization is now also invariant to rotation about the roll axis.
      • Added pcl::PointSurfel to known point types.
      • eigen-decomposition for symmetric positive-semi-definite 3x3 matrices: 1) bug fix so eigenvects are orthogonal, 2) is more robust for degenerated cases

    • [[pcl_ros]]

      • MovingLeastSquares nodelet improvements
      • Changed serialization of point clouds to ship the data as-is over the wire, padding included (#4754). Implemented subscriber-side optimizations to minimize memcpys. Will do one memcpy for the whole point cloud if the data layout is an exact match.