Controlling the motor direction by tilting of head and using KNN classifier to identify the pattern
G. Akhil, I. Satyakumar, K. Sarath, M. Rahul, Prasanth M. Warrier · 2017
This paper presents a methodology for controlling the direction of motor taking a video sample from a camera as input. To control the direction the subject has to move his head in a direction which he would want the motor to rotate. The main challenge would be classifying the test sequence which has the data of the activity performed by the subject The actions are recognized in the frontal view by tracking the centroid of the head over consecutive frames in a video. Input sequence is matched against the sample points and the action is labeled by using the neighborhood. Once the class is labeled control signals can be sent to the motor to rotate it in the desired direction. In the present implementation we tried to classify the bending right and left directions since the motor has only two degrees of freedom. The paper uses a basic K-Nearest Neighbor(KNN) classifier to classify the actions. This methodology can find an application where an individual without arms can control the direction of a wheel chair using a camera interface. It discusses the reason for preferring the KNN classifier which has more error probability than Bayesian frame work for the classification of activity of a person in a given video sample.