Real-time Kinect-based air-writing system with a novel analytical classifier

Air-writing is an attractive method of interaction between human and machine due to lack of any interface device on the user side. After removing existing limitations and solving the current challenges, it can be used in many applications in the future. In this paper, using the Kinect depth and color images, an air-writing system is proposed to identify single characters such as digits or letters and connected characters such as numbers or words. In this system, automatic clustering, slope variations detection, and a novel analytical classification are proposed as new approaches to eliminate noise in the trajectory from the depth image and hand segmentation, to extract the feature vector, and to identify the character from the feature vector, respectively. Experimental results show that the proposed system can successfully identify single characters and connected characters with the average recognition rate of 97%. It provides a better result than other similar approaches proposed in the literature. In the proposed system, the character recognition time is quite low, about 3 ms, because of using a novel analytical classifier. Evaluation of 4 classifiers shows that the proposed classifier has a higher speed and precision than the SVM, HMM, and K-nearest neighbors classifiers.