I have an existing C++/opencv tracking library with neural-net based scene and object ID features. The image ID is accomplished with a back-end opencv/keras/tensorflow based image -id ajax back end..
Relevant Skills and Experience
I have built and have ready a system such as you describe, including a multi- object, multi-technique opencv c++ tracking library. I can demo this now.
Proposed Milestones
€125 EUR - initial opencv c++ prototype linux tracker and object/scene ID system
€125 EUR - additional improvements, fixes, added features
If you have sample video, I can run it through my tracker/scene/object ID sys, then send a demo back for you to evaluate. Many more features, I can show you API docs, demos, etc. All right now. I'm using various existing trained neural nets for image ID. [Now pure-C++ (*no* python) for this too.] The multiple moving-object (or person etc.) tracker has many different techniques for tracking (over 10), which can be efficiently mixed/matched per tracking task. Ask for API, demos: all ready to show you now... Visual tracking can be difficult ... what are you tracking? I have many tracking systems for different tracking situations. My tracking system also has a mode where the scene is examined, a dnn ID's the whole scene, then a specific tracking mode (track people, vehicles, or fish, etc.) is then used, depending.
[Oddly, I see others' here have bid below 250 Euro but freelancer will not let me bid any lower. I would underbid them if freelancer would let me. do so.]