DeepLabCut VS SuperGluePretrainedNetwork

Use DeepLabCut, I also strongly suggest that you should fund their work: https://github.com/DeepLabCut/DeepLabCut

DeepLabCut is also worth mentioning here

1 - Detecting the dog's body parts was the most difficult portion of this, and thankfully I stumbled upon DeepLabCut (https://github.com/DeepLabCut/DeepLabCut) which enables training a model to track a specific animal(s) posture. In the video, this is basically the dots that are overlayed on top of the dog, and follower her around. DeepLabCut is basically just saying that this is where it thinks it recognizes "spine" and "tail".

I used https://github.com/DeepLabCut/DeepLabCut for the core dog tracking capability, then I wrote the code that analyzes the posture (output by the model, trained via DeepLabCut) of the dog.

I built a dog poop detector for my backyard using DeepLabCut (https://github.com/DeepLabCut/DeepLabCut) and some janky poop detection heuristics I wrote that processed on the detected posture of my dog, if it's in the frame of my security camera. If it detects my dog pooping, it will record the location in a CSV and draw all the locations on an up to date image of my backyard.

Also, check out DeepLabCut. My project wouldn't have been possible without it, and it's really cool: https://github.com/DeepLabCut/DeepLabCut

Well you could try DeepLabCut - https://github.com/DeepLabCut/DeepLabCut

Code: https://github.com/magicleap/SuperGluePretrainedNetwork

This basically includes a deep learning based approach to do keypoint detection, and match them across image frames. This includes papers like SuperPoint, Superglue, and more. There is also a way to do dense matching with neural networks.

Feature matching is the way to go imo. Try out OpenCV's inbuilt feature matching methods like SIFT and FLANN. If the performance is poor, you can even try out CNN aided matching algos like SuperGlue Link (CVPR2020)

SIFT and brute force matching is your best bet in classical computer vision if you're unconcerned with runtime. There are methods from deep learning that can perform better, somewhat domain dependent. Check out superpoint and superglue from magic leap. https://github.com/magicleap/SuperGluePretrainedNetwork