I was wondering whether it was possible to recognise if an object is in a desired pose with some visual recognition tools (in particular, I was hoping to use IBM Watson services). For instance, I need to check if a cylinder is in the desired position relatively to other objects near it. Hence, given enough features in the environment around the cylinder, would this be possible?
Sorry, but I am really new to visual recognition field and I still need to learn lots of things. Also, I haven't been able to find anything online regarding this issue. Thank you!
whether it was possible to recognise if an object is in a desired pose with some visual recognition tools
Yes, you can train a custom classifier in Watson Visual Recognition, just by giving it example images.
It cannot solve every problem in this space of pose classification, but depending on what type of object and what pose, and the surrounding environment, it could solve a lot of them. To say, for instance, if we are looking at the frontal or rear view of a car is pretty easy if the car is the main thing in the scene. You would just need to give it a few dozen examples.
This is how to train a custom classifier: https://console.bluemix.net/docs/services/visual-recognition/tutorial-custom-classifier.html#creating-a-custom-classifier
Keep in mind, it can only be trained to recognize distinct classes you have given examples of, such as "upright cylinder" vs. "cylinder laying on its side". It would not report a continuously varying output like "37 degree rotated" cylinder.
The best approach is try it out experimentally. There is also a demo GUI you can train https://visual-recognition-demo.ng.bluemix.net/train without an API key (though your classifier would not persist) and a GUI training tool (requires API key, so your classifier would persist): https://watson-visual-recognition.ng.bluemix.net/