Human-Computer Interaction is a flourishing area in terms of research and has many real-world applications. Keeping this in mind, we came up with an idea to develop Human- Computer interaction for proper conduct of physical exercises at home, using information sensed by an RGB-D camera, namely the Microsoft Kinect. Along with Kinect, we also make use of a Machine Learning technique to perform operations on captured data to predict the accuracy of a performed physical exercise. Our approach is based on the study of the movement of various joints in the human body, which we examine with the use of the Kinect. We take into account an algorithm for our implementation - Hidden Markov Model (HMM). We combine these and detect the posture of a user while he performs a particular exercise, before comparing it with our ideal database of postures. Based on this comparison, we predict the accuracy of the exercise and aim to improve and correct the form of the user in terms of performance of the exercise.