Hand Gesture Recognition

Abstract

This paper describes the implementation of
movement control for robotic car using hand gesture
recognition which uses deep learning algorithm.
Therefore, proposed technique is hassle free as
control is not based on joysticks or switches. There
are six conditions considered for robot car movement
control as ‘Backword’, ‘Forward’, ‘Left’, ‘No-
Motion’, ‘Right’ and ‘Stop’ using different hand
gestures. There are many researchers worked on this
area using different sensors, machine learning
algorithms and deep learning algorithms. Limitations
of the state of art techniques are studied in this paper
and designed a new modified convolutional neural
network (CNN) for gesture recognition which
controls movement of robot car. Dataset is created
which generates 1000 Gray scale images for each type
of gesture. Training modified CNN model gives
prediction accuracy of 98.4024 % while random
forest machine learning classifier gives prediction
accuracy of 69%. It is observed that proposed model
gives better accuracy compared to state of art
technique for controlling movement of robot car
using hand gesture. Obtained hand gesture class can
be send to robot using Arduino controller for
controlling movement.