Extracting bounding boxes in deepstream

How can I extract bounding boxes from detection.
I am testing deepstream_test3_app with multiple sources of urls.

I like to have bounding boxes from each frame in object detection so that I can do mapping the pixel locations in images to outside location using homographic transformation.

How can I have bounding boxes from images in detection with multiple sources?

nvdsosd has NvDsBatchMeta input. NvDsBatchMeta has all meta data. How can I have boxes extracted from NvDsBatchMeta? For that, do I need to make another plugin?
I found this info as I need to make a plugin.
For batches of multiple sources, how can I know boxes are from which frame of which source?

Now I know I can follow the sample in osd_sink_pad_buffer_probe in test1.
source id can be taken from framemeta and bboxes are from objectmeta.

Yes, so have you solved your problem?

Yes, thanks