Hello,
I’ve just recently started reading about DeepStream and RTSP, so I’m not completely sure how to work my way out of the following issue: I’m projecting a solution where we’d like to apply an object detection neural network to a live feed IP camera and present the bounding boxes of the detected objects to users across the internet. Multiple users are expected to be watching this stream simultaneously, each, perharps, wanting to see a different set of bounding boxes.
I know this can be achieved with nvdsosd, but so far my understanding is that this will “hardcode” all the rectangles on the image frame. What I would like to do is to let the user choose how and when to draw a bounding box, so I’m leaning towards sending the inference results (bbs, confidence, class labels) and more metadata (strings, floats, int and the like) along with the image frame, which would need to get decoded on the client-side and finally be displayed using javascript.
I’m trying to work my way through deepstream_test_3 and deepstream_test1_rtsp_out samples to achieve this, but haven’t been able to decode the metadata from the stream in order to draw them solely on the client-side. Is this possible? Is there a better way to achieve this? Should I be focusing more on syncronizing the stream and metadata with RTSP than trying to pack both with DeepStream?
Let me know if I can give you any further insights into this.
Thank you for your time.