Investigate potential models to use with deep-sort

[cczarnuch]

Create a list of potential models that we can use to detect items from this list.

• Various sized trucks
• Cars
• Busses
• Pedestrians
• Cyclists
• People with mobility devices

We will probably need to take and existing model and train it further for accuracy in different conditions. (nighttime, rain, snow, etc.) If there is not a sufficient model for this purpose, we will have to train our own.

[cczarnuch]

Take a look at this repo: https://github.com/pythonlessons/TensorFlow-2.x-YOLOv3

It may provide some guidance as to what kinds of models we can use. We don't have to find a model that is specifically cars, trucks, pedestrians, etc. We can just ignore the objects that we don't need.

[icecap360]

@cczarnuch I want to make sure that my understanding of the issue is correct: Deepsort an algorithm that given boundary boxes classifies objects, using its own association metrics. The subject of this issue is to find DL architectures that find the boundary boxes in the first place.

[cczarnuch]

@cczarnuch I want to make sure that my understanding of the issue is correct: Deepsort an algorithm that given boundary boxes classifies objects, using its own association metrics. The subject of this issue is to find DL architectures that find the boundary boxes in the first place.

Yeah sorry, I could have worded it better. But yeah that's pretty much what we need. I linked the other repo because they have an example where they were able to use the tensor flow implementation of YOLOv3, ...v4, etc. That would be able to generate the bounding boxes so we can run the tracker on them.