Above image shows a plane. I would want to find the points which represent the inner spacing. Something like the fourth image at http://cgm.cs.mcgill.ca/~godfried/teaching/projects97/belair/alpha.html .
I was attempting Concavehull with an alpha of 0.1 but it does not makes sense. Attached is the hull with alpha = 0.1. On increasing alpha, the points become even sparse. What would be a good way to get inner points?

No specific reason. I tried with a bunch of values like 0.1, 0.2, 0.001, 1.0, 2.0 and so on. I found that larger and smaller values yield very strange results which do not fit my criteria.

You are using pcl::SACSegmentation before ConcaveHull and by doing so you are losing a lot of points and as result, losing lot of details. It is not wrong to do that but you'll get different result.

I use ConcaveHull with original cloud and if I want to get flat result then I compute plane and project my result points onto plane.

Both codes are ok but they give different results, depends what you want.

Indeed, that was causing a lot of points to drop. As the point cloud itself is plane so I don't need to find a plane again and project it. Can directly find hull for this.