Air Writing Recognition of Geometric Shapes using CNN

Air Writing can be considered as virtually drawing or writing characters in air using hand gestures. This paper proposes a system which is a combination of Computer Vision and Convolutional Neural Network (CNN). Computer Vision is used for detecting and tracking the trajectory of the marker tip gestures performed in front of the webcam, whereas CNN is used for the recognition of geometric shapes. The proposed model recognises the geometric shapes (Square, Triangle and Circle) drawn in air using a marker of fixed color (here blue). The recognition accuracy of the model is improved further by implementing Transfer Learning in MobileNet. The final model achieved an accuracy of 98%. The system proposed in this paper does require only a laptop camera.