I am working on a project where I need to track 20 moving objects in real time (30 FPS) in a HD video stream (1920 x 1080). The objects are about 45 x 45 pixels in size and are a black character on a white background (For example the letter 'A' printed on a white label). These objects are robotic cars moving around in an area about 4' x 8' in size (1.2 x 2.4 meters). Each car is about 4 x 4 inches (100 x 100 mm). The cars will be moving at a speed of approximately 450 Inches / minute (11 meters / minute.
I have done a little bit of object detection using a Raspberry Pi 4 and OPEN CV but from that experience, I do not believe those are up to the task.
I have been looking into the Jetson Orin Nano, it may be up to the task, but have not seen any example videos showing the JON doing that or similar.
For this application, I would be able to use a desktop PC or the Jetson Orin Nano, the hardware selection isn't critical.
The next area where I can make a decision is what software tool to use. There are a lot of terms and method names that seems to be expanding faster than I can comprehend (I am in the 60+ crowd unfortunately). Tensor Flow, YOLO, Inference, Open CV, etc. etc. etc.
I sure could use a little direction on this so that I don't keep getting overwhelmed everytime I start researching. Anything at all the would narrow down the possible choices would be of great value to me.
Chris