How to crop rs2::frame

[AlexKanakis5]

Required Info
Camera Model	D415
Firmware Version	05.13.00.50
Operating System & Version	Linux (Ubuntu 20)
Kernel Version (Linux Only)	5.15.0-53-generic
Platform	PC
SDK Version	2
Language	C++
Segment	others

Issue Description

<Describe your issue / question / feature request / etc..>

Hello,

I want to crop a frame to specific dimensions. From what I understand, I have to convert it to Mat, crop it and then convert it back to a frame. I have seen the issue #2634 on how to convert Mat to frame, and the code works when applying Gaussian blur as the example. However when I crop the Mat and then convert it, the resulting frame is not cropped. I am new and I am really struggling with this, any help would be much appreciated.
realsense_crop.txt

[MartyG-RealSense]

Hi @AlexKanakis5 There is a further discussion of this cropping method at #8512 if you have not seen it already

A couple of different approaches to cropping are at #2016 (comment) and #3894 (comment)

[AlexKanakis5]

I didn't get an answer to my original question but these suggestions really helped. What I wanted was to create a point cloud from the copped frame. What I ended up doing was to follow what the #2016 comment described with some help from the link in the #3894 comment. For anyone in a similar situation as me this is the methodology I followed:

1. Get the pixels of the box that would be used to crop the image
2. Convert them to points by using the rs2_deproject_pixel_to_point function
3. Iterate over the points and only keep the points inside the cropped part

Here is a snippet from my code that I used to convert the pixels to points. This does not work on its own but it gives an idea on how to get there.

auto profile = pipe.start(cfg);
rs2::frame depth = frames.get_depth_frame();

// get calibration data for pixel to vertix convertion 
auto stream = profile.get_stream(RS2_STREAM_DEPTH).as<rs2::video_stream_profile>();
auto intrinsics = stream.get_intrinsics(); // Calibration data
rs2_intrinsics intr = depth.get_profile().as<rs2::video_stream_profile>().get_intrinsics(); // Calibration data

// this gets pixel values of bounding box and converts them to 3D points 
float startpixel[2]; // From pixel
float startpoint[3]; // From point (in 3D)
pixel start = {171,43}; // pixel value
startpixel[0] = start.first;
startpixel[1] = start.second;
auto startdist = depth.get_distance(startpixel[0], startpixel[1]);
rs2_deproject_pixel_to_point(startpoint, &intr, startpixel, startdist);

[MartyG-RealSense]

It's great to hear that you achieved a solution, @AlexKanakis5 - thanks so much for sharing it with the RealSense community!

[MartyG-RealSense]

Case closed due to solution achieved and no further comments received.