One of the most promising topics of research in the field of artificial intelligence is the application of data captured from human motion using sensors processed with various algorithms to achieve successful data analysis. This project aims to design and develop a method to judge human motion and allows the users to see the score while they are performing motion. The Neural Network is trained to follow human expert scores on all players' motion profiles as compared with master player. This Neural Network is trained for its best performance by changing the number of iterations and the dataset passes through the network. Its sets the time period to train the Neural Network and sets the gradient so that error generated is a minimal or changing mean square error. The user data derived from a Yoga performance or any other motion was collected from Kinect sensors' data reports. The data collected contained X, Y and Z rotational and positional points. Kinect sensors captures 20 joints. The master data and user data were used as input to the Dynamic Time Warping algorithm. It compensated for the speed and time between the master and user data and gave a value that suggests the relative similarity of both motion profiles. The output of this algorithm was fed to Neural Network that was trained with a human judgment expert's data on motion profiles. This project is an attempt to train a Neural Network that will eventually judge like an expert and determine the success level that a user's motion profile exhibits.
Shah, Priyank, "Motion Learning Using The Neural Network" (2012). Master's Projects. 263.