Tracking camera for gardening

I use a Jetson Nano B01
Deeepstream version is 5.0
Jetpack is 4.4
raspberry pi cam V2
2 motors connected to Jetson GPIO pins
motor drives to control said motor using GPIO pins

Project explantion
My needs are quite basic but the python deepstream python apps ae quite daunting.
I just need to move a camera on a rail (vertically) the camera will stop and look at the tomatoes ( as i want to pick them ) and see if its red. Using the pretrained yolov3-spp model it will decide to pick it. Where a scissor or blade controlled by another motor on a rail (horizontal) will cut it .

*** Technical Problem explanation***
I just need to get the rect params ,confidence value and ***obj meta ***name. how should I go about this should I extract each frame and from that frame cast the meta data.

Update: I use open-cv and numpy .
So the I have re-used some code from post processing image app in python samples. There is a lot of stuff that I don’t think is necessary for my problem. So I wanted to trim it down, but it is like looking at a mine-field. Its difficult to navigate and I am having trouble to know what to keep and not to.

Your question is too abstract to be answered.

Can you study DeepStream document first? https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Overview.html#

Sure