| 78 | |
| 79 | |
| 80 | void ApproximateCoplanarFilter::filter(PointView& view) |
| 81 | { |
| 82 | const KD3Index& kdi = view.build3dIndex(); |
| 83 | |
| 84 | for (PointRef p : view) |
| 85 | { |
| 86 | // find the k-nearest neighbors |
| 87 | PointIdList ids = kdi.neighbors(p, m_knn); |
| 88 | |
| 89 | // compute covariance of the neighborhood |
| 90 | Matrix3d B = math::computeCovariance(view, ids); |
| 91 | |
| 92 | // Check if the covariance matrix is all zeros |
| 93 | if (B.isZero()) |
| 94 | { |
| 95 | log()->get(LogLevel::Info) |
| 96 | << "Skipping point " << p.pointId() |
| 97 | << ". Covariance matrix is all zeros. This suggests a large " |
| 98 | "number of redundant points. Consider using filters.sample " |
| 99 | "with a small radius to remove redundant points.\n"; |
| 100 | continue; |
| 101 | } |
| 102 | |
| 103 | // perform the eigen decomposition |
| 104 | Eigen::SelfAdjointEigenSolver<Matrix3d> solver(B); |
| 105 | if (solver.info() != Eigen::Success) |
| 106 | throwError("Cannot perform eigen decomposition."); |
| 107 | Vector3d ev = solver.eigenvalues(); |
| 108 | |
| 109 | // test eigenvalues to label points that are approximately coplanar |
| 110 | if ((ev[1] > m_thresh1 * ev[0]) && (m_thresh2 * ev[1] > ev[2])) |
| 111 | p.setField(Id::Coplanar, 1u); |
| 112 | else |
| 113 | p.setField(Id::Coplanar, 0u); |
| 114 | } |
| 115 | } |
| 116 | |
| 117 | } // namespace pdal |
nothing calls this directly
no test coverage detected