DATASET AND CODE FOR FITTING CYLINDER

1. Data set

This data set uses to evaluate the fitting cylinder from point cloud data of cylindrical objects. The format of all data sets is '.pcd' file (file format of point cloud data (pcd)). You can download the data set in this link 'Generated cylinder data set'. This data set consists of six different subsets, denoted from dC1 to dC6. For each subset dCi, inlier ratio is increased by a step of 5% from 10 to 80 % as Table 1. Therefore, there are fifteen point clouds. They are denoted dS1 to dS15. A point cloud dSi consists of three thousand points. Each point of a dSi has the 3-D coordinate (x, y, z) which generated from MATALB based on the cylinder equation. An inlier data point (xi,yi, zi) of dSi is lying on the cylinder's curved-surface which is generated as follows: xi = cos(θi); zi or yi= sin(θi) respectively; yi or zi is randomly selected in [0, 1] respectively; θi is randomly selected from [0, 2π]. That means the cylinder's axis of generating cylinder following the direction of y-axis or z-axis of the original coordiante system. The radius of the generated cylinder is 2, the coordiantes of a point on the cylinder's axis are (0,0,0) and the height of cylinder is 1. Therefore, all of inlier points are generated from a cylinder, that has the equation (x2 + y2 + z2 = 22). Outliers are generated randomly as a normal distribution from [-3, 3] or [-4, 4]. The detail of each subset is presented in Table 1.  Some illustrating images of dC1 to dC6 at the inlier ratio 50% (dS5) are presented as Figure 1.

Dataset Characteristics of the generalized data
Height/
Radius		 Direction
of main
axis
 Spatial
distribution
of inliers		 Spread of
outliers
dC1,
dC4		 1 /
2		 parallel
withthe z-axis		 Around of
a cylinder		 [-3, 3],
[-4, 4]
dC2,
dC5		 1 /
2		 parallel with
the y-axis		 Around of
a cylinder		 [-3, 3],
[-4, 4]
dC3,
dC6		 1 /
2		 parallel with
the y-axis		 one half of
a cylinder		 [-3, 3],
[-4, 4]
Table 1: The characteristics of the first dataset (synthesized data)
Figure 1: Point clouds of dC1, dC2, dC3,dC4, dC5, dC6 of the first dataset (the synthesized data) in case of 50% (dS5) inlier ratio.
The red points are inliers, whereas blue points are outliers.

2. GCSAC Code
This code is improved of MLESAC algorithm. You can download the code in this link GCSAC code. The proposed frame-work is deployed for fitting a cylinder object from a point cloud data. The corresponding normal vectors are already prepared in advance. The most critical step is the first one - sampling and/or searching good samples procedure, whereas other steps are deployed on the MLESAC algorithm. In this code we have attached a tutorial file for using the software with the name 'readme.txt'.

3. Illustrating Results
Some result images of fitting cylinder on the generated data set are presented in Figure 3 and Figure 4. They also have in the folder 'Image' of  'GCSAC' code.

Figure 3: Illustrating the fitting cylinder on the generated data has 10% inlier. Red points are defined inlier points; blue points are defined outlier points; green cylinder is the estimated cylinder from data points. Left is the result of estimating a cylinder based on the MLESAC original algorithm, right is the result of estimating a cylinder based on the our algorithm.

Figure 4: Illustrating the fitting cylinder on the generated data has 20% inlier. Red points are defined inlier points; blue points are defined outlier points; green cylinder is the estimated cylinder from data points. (a), (b), (c) are the result of estimating a cylinder based on the MLESAC original algorithm, the FILSAM algorithm, the our algorithm respectively.