How to get data in an array in a pattern? - c++

Question

I am trying to get the x and y values out of a PointCloud2 data. This data is stored in an array in bytes of float32's. My question is, how can I get only the x and y data?
I will add a snippet of what the PointCloud2 data looks like, but for now I'd like to just be able to get the x - data. So for every 16 values, I only want the first 4 (I want the first 4 as this is how it is ordered and because the float32's are decomposed into 4 bytes each).
My original thought process was just to use a loop, but I am worried this may be too slow. This is as my array has more than 2000 values and I am getting 15 of these arrays per second. Is there any other way of doing this?
seq: 296
stamp:
secs: 1553456947
nsecs: 421859979
frame_id: "cloud"
height: 1
width: 811
fields:
-
name: "x"
offset: 0
datatype: 7
count: 1
-
name: "y"
offset: 4
datatype: 7
count: 1
-
name: "z"
offset: 8
datatype: 7
count: 1
-
name: "intensity"
offset: 12
datatype: 7
count: 1
is_bigendian: False
point_step: 16
row_step: 12976
data: [108, 171, 8, 191, 107, 171, 8, 191, 0, 0, 0, 0, 0, 0, 113, 70, 167,
197, 8, 191, 103, 95, 10, 191, 0, 0, 0, 0, 0, 240, 115, 70, 99, 101, 9, 191,
127, 161, 12, 191, 0, 0, 0, 0, 0, 200, 106, 70, 99, 50, 5, 191, 202, 237, 9,
191, 0, 0, 0, 0, 0, 76, 82, 70, 200, 22, 235, 190, 200, 74, 246, 190, 0, 0,
0, 0, 0, 132, 52, 70, 186, 111, 255, 190, 99, 95, 7, 191, 0, 0, 0, 0, 0, 24,
60, 70, 227, 1, 8, 191, 89, 217, 17, 191, 0, 0, 0, 0, 0, 64, 93, 70, 216,
183, 8, 191, 245, 84, 20, 191, 0, 0, 0, 0, 0, 236, 112, 70, 195, 94, 8, 191,
64, 177, 21, 191, 0, 0, 0, 0 ...
I would also like to add that I am a beginner at this sort of thing. I spoke with my friend and he said using mutex threading for this, but that seems way over my head.
Thanks!

Answer 1

I suspect that you'll be surprised how fast modern hardware can loop through 30000 points.
You should start off with your loop and some basic considerations. For example, for storage of our X/Y coordinates in a vector, reserve the required space in the vector before you enter your loop to avoid extra memory allocations. Once you have your loop working, evaluate the performance and maybe post another question if you think that your code is too slow.
A helpful quote by Donald E. Knuth that may apply:
Premature optimization is the root of all evil.
Edit; to provide some context: We had an application running on a low end i5 processor collecting 76,800 3D points at 15 fps via the Kinect SDK or via OpenNI (different sensor). We'd loop through the points and apply a basic transformation before storing each point into a pcl::PointCloud. We'd then loop through the point cloud points again for some analysis. The analysis portion far outweighed the cost of any basic loop. You will probably end up worrying about optimizing whatever evaluation logic you want apply to your data rather than basic things such as looping over the points.