Gesture recognition by learning local motion signatures using smartphones

In recent years, gesture or activity recognition is an important area of research for the modern health care system. An activity is recognized by learning from human body postures and signatures. Presently all smartphones are equipped with accelerometer and gyroscopes sensors, and the reading of these sensors can be utilized as an input to a classifier to predict the human activity. Although the human activity recognition gained a notable scientific interest in recent years, still accuracy, scalability and robustness need significant improvement to cater as a solution of most of the real world problems. This paper aims to fill the identified research gap and proposes Grid Search based Logistic Regression and Gradient Boosting Decision Tree multistage prediction model. UCI - BAR dataset has been used to perform Gesture recognition by learning local motion signatures. The proposed approach exhibits improved accuracy over preexisting techniques concerning to human activity recognition.