Finding Special Feature in 3-D points

I have 3-D points data as following. Could you suggest a way to locate the two features circled? I can use regression and dy/dx to achieve this task if it's 2-D.

[Walter Roberson]

To my eye, there is nothing special about those points, as compared to a little distance away. You will need to be more specific about what you are looking for.

[Hg]

I am looking for a region on the surface where there's a sudden/big change in normal vector.

[Walter Roberson]

Normal vector has no meaning for a point cloud. Do you have a mesh rather than a point cloud? Even if you do, I do not see any sudden change in normal at the locations you have indicated.

[Hg]

I don't have a mesh. I'm thinking of doing it in another way. Get y-z plane, fit some lines, and find the intersections of the lines. Is it possible to get these lines?

[Walter Roberson]

That might be possible. A difficulty that is commonly encountered with something like that is in deciding where to place the segment, taking into account that the points must be taken to be noisy coordinates. Still, given a specific number of segments, it is linear programming to decide where to place the breaks to minimize the fitting error -- but it is still tricky to figure out which points should be included in the fitting.

[Hg]

Is there any function I can use to achieve that if I get the edge image?

[Walter Roberson]

One would need to be written but it would be easier than the non-edge case

[Hg]

I see. So there's no existing function to do this. I guess I have to figure it out by myself, Thanks for the advice