Among practically all nations, the occurrence of falls in the elderly is currently a big concern. Severe injuries occur as a result of the fall and sometimes even to mortality. Physiotherapy is an analysis of human joints and bodies, which gives remedy for any pains or injuries. Kinect Xbox One is a low - cost 3D camera which can be used for human motion tracking. The human movements are tracked in real time by using Kinect Xbox One and use the human exercises to calculate certain values in parameters. This paper contains, a Multi - Path convolutional Neural Network (MP - CNN) is recommended for doing rehabilitation exercises recognition using Kinect Xbox One sensor data. It contains two important components. They are a Dynamic - convolutional Neural Network (D - CNN) and a State transition probability convolutional Neural Network (S - CNN). Gaussian Mixture Regression (GMR) is used in D - CNN to record the sensor data's regarding the way the body moves at the time exercising for rehabilitation. The input signal and the GMR are appeared in various segments. In S - CNN we use Lossless Information Compression Encoder Algorithm (LICE) to take use of the hidden states of distinct motions' the transition probabilities. The combination of D - CNN and S - CNN creates the MP - CNN. We use Augmented Reality (AR) technique in this method. When comparing the results obtained from MP - CNN, KNN and Logistic Regression it is found that MP - CNN is better than Logistic Regression and KNN.