A Comparative Analysis of Cloud Based Watson System and CNN for Gesture Recognition Systems

According to census 2011, the number of disabled people in India is 2.68 crores. Out of those, about 19 percent have a problem in hearing. With the advent of Convolutional Neural Networks (CNN) deployed on the host system and multi cloud platforms like IBM Watson, an important challenge faced by the developers is selection of suitable architecture for deployment. The ability of CNN to model non-linear relationships enables it to be used widely in biomedical domain and thus for the problem of disabled people. Recognition models deployed on Cloud offer on-demand secure storage, analysis and rapid scalability of services. This paper aims at providing a comparative study between the two architectures. For the first type of architecture the gesture of a mute person is recognized using image processing and CNN. Whereas second architecture uses cloud based visual recognizer to recognize the gestures. The prominent parameters such as recognition accuracy, angled detection and response time that play an important role when deploying the two architectures are measured and provide a perspective over the selection of architecture. The accuracy obtained for the CNN model is 98% and 97% for the cloud-based Watson model for the trained tested classes.