I am designing a robot which irons and folds clothes autonomously. For this purpose, I need the robot to detect a certain number of key points in a given cloth in order to execute a certain folding or ironing algorithm. The following images best describe what I actually mean by key points
For example for a trouser, the above image shows the desired key points.
My first thoughts:
My first plans were to use neural networks detect various key points after recognising if the cloth is a t-shirt, trouser etc. But I believe, the problem with this approach is that collecting a lot of data sets to train the neural network to detect these key points is a massive task. Furthermore, I am not even sure if neural networks can produce the results I am looking for because generally, neural networks work well differentiating well defined varying classes of items like for example cats and dogs.
So my question is, is there a better way to achieve what I am trying to do? Any help is appreciated thank you.
Ideally, what I want to do is identify or recognise each of these key points precisely; meaning, for example, the system should know where the collar region for a t-shirt and track its location at all times. I need to know what each key points are (is it a zip area or shirt's shoulder or a collar region) in order to execute a certain folding algorithm. So in this case, I believe, convexity defects do not work.