Object Tracking: What is the algorithm that fits my needs

I have been searching object tracking algorithms in opencv, ranging from correlation filters , feature mapping to ones based on deep learning, but I am facing some difficulty in determining which one to fit my needs.

I have a research problem where I need a robust tracking for an object which its scale varies along time (but either in increase or in decrease “monotonic scale”). This can be modeled as a moving camera which heads (towards / away from) an object which is real-time detected. The idea is that the scale is changing in a noticeable way during the real-time.

What can I research and use in this case? Thanks in advance.

what about making an example video, and trying the benchmark program on it ?

I consider the tracking module to be abandonware.

for a visually simple task, I’d go for a correlation tracker (samples/python/mosse.py) because it can achieve subpixel localization.

for a visually challenging task, I’d go for some DNN approach. that’s where the research is. some of these networks were explicitly designed to handle scale changes.

the dnn module contains examples for SiamRPN and successor models.