I am working on an emotion recognition using OpenCV and machine learning and in the process tried this out today. This is a landmark detector and shape predictor. It makes the use of haarcascades and 68 defined face points to detect the exact coordinates of the shapes. This facilitates us to learn and observe the changes in facial characters by training a model of different emotions and then comparing the change observed in the video feed to the trained model. I will be using Support Vector Machines (SVM) for the first part of it and will later include more complex training models like Deep-Belief Networks to do the same. Small things like these give you a boost to study and work in the area of machine learning.