Using mini minimum jerk model for human activity classification in home-based monitoring

This paper proposes a method for human activity classification in home based monitoring. The proposed approach is based on minimum jerk (MinJerk), a primary model for smooth path planning employed by human motor control in upper-extremity motion. Based on new evidences that show common control strategies in lower and upper extremity, MinJerk is adapted in our study to estimate the foot motion with fifth order polynomial functions. Experimental data are recorded during walking and going up and down the stairs using a single inertial measurement unit. Features of interest in this study are the optimized curve fitting coefficients. Using a structured support vector machine with radial basis function, an overall accuracy of 98.6% is achieved for activity classification. The proposed method is also capable of detecting the transitions between the movements with accuracy of 99.96%.