Shape recognition with point clouds in rebars

ing point clouds of reinforcing bars fr om raw point clouds The method of abstracting point clouds of reinforci ng bars Point clouds of reinforcing bars have a complicated shape. Therefore the point clouds of reinforcing bars cannot be distinguished directly from raw point clouds by using the mathematical methods such as least-square method. Hence, the authors have developed a method of abstracting point clouds of reinforcing bars from raw point clouds. The method of abstracting point clouds of reinforci ng bars in three dimensional spaces Point clouds of reinforcing bars are not a primitive shape such as plane, sphere and cylinder. Although point clouds of a primitive shape don’t include point clouds of reinforcing bars, point clouds of non-primitive shape include more much point clouds of reinforcing bars than raw point clouds data. Therefore, if point clouds of a primitive shape are removed from raw data, point clouds including reinforcing bars remain. Fig.9 Point clouds put into voxels and then replacing planes using least-square method Fig.10 Point clouds of sphere replacing a sphere model using least-square method Raw point clouds are put into voxels which represent a value on a regular grid in a three dimensional space (Figure.9). Next, the point clouds in a voxel are fitted into a sphere and plane by the least-squares method. If the sum of the squares of the errors is less than tolerance, the point clouds are fitted into a sphere (or a plane) model and then the point clouds are removed from raw data (Figure 9 and 10). Fig.11 An example of the flow diagram dealt with shape recognition in three-dimensional spaces. Remaining point clouds, which include point clouds of reinforcing bars, go on to next steps for abstracting point clouds on reinforcing bars in a cross section. Steps of abstracting point clouds on reinforcing bar in a cross section The method of abstracting point clouds on reinforcing bars in a cross section are developed by the authors consisted of the following three steps: (a) The point clouds are sliced in horizontal 2D cross sections (Figure.12) (b) In a cross section, point clouds is grouped by taking account of the continuity between pixel blocks (Figure.13 and 14) (c) The system recognizes the shape in each cluster of point clouds and then distinguishes point clouds on reinforcing bars (Figure.15) Plane