The code of this article has been uploaded to GitHub
. In the previous article, due to factors such as the conversion on the left, the accuracy was lost, and the original point cloud image was not restored. After a day of thinking, I found that the previous method is not particularly desirable. This article will use a new method to restore the original image to the greatest extent. Before that, make a summary of the two methods:
Method 1: Since the point cloud xyz format corresponds to coordinates and is a floating point number, then convert *10 to an int type number. Then use the maximum number of x-axis and y-axis as the image size.
Method 2: In fact, the point cloud image is consistent with the two-dimensional image, and each point is ordered. We only need to count how many numbers appear on the x-axis and y-axis to determine the size of this graph. For example, a point cloud map.
| x | y | z |
|---|---|---|
| 1.1 | 1.1 | 1.2 |
| 1.1 | 2.0 | 1.3 |
| 2.0 | 2.0 | 1.3 |
Can you see it, in fact, the conversion of this point cloud image into a grayscale image is 2*2 pixels. Because the x-axis only has (1.1, 2.0) y (the same goes for the y-axis). So for example, the first coordinate (1.1, 1.1, 1.2) corresponds to (0, 0, 1.2) on the grayscale image. 0 comes from 1.1 is the 0th element of the x-axis.
Just to mention here, the x-axis and y-axis of the previous article are reversed. How many x-axes indicate how many columns there are, and how many y-axes indicate how many rows there are
Note that the way of indexing yx and map is about twice as fast as np.argwhere.
