Indian Sign Language Recognition Using Weighted Motion History RGB Images Through Modified LeNet-5

Gesture plays a dynamic role in the area of computer vision for effective communication. In recent years, Human-Computer interface has become research of interest in gesture segmentation, identification and recognition. Sign language as an application of gesture plays a major role among the deaf-mute public for communication. Sign Language being the native language of deaf-mute community, there is a need for interpreter for normal people to communicate with them. This paper focuses on Indian Sign Language recognition which overcomes the barrier between deaf-mute people and normal people for easy communication using vision-based technique. Vision based systems dependent on traditional cameras are preferred rather than the multi-modal data such as skeletal and depth information. Existing sign language recognition systems employ complex algorithms for extracting spatiotemporal information along with deep learning architectures. A novel temporal condensation method Weighted Motion History based RGB (WMH-RGB) is presented. The proposed technique is used to convert input video into a single RGB image at the pre-processing stage and a modified LeNet-5 is used for classification. The proposed method reduces the problem of video classification to RGB image classification with WMH-RGB method, which reduces the system complexity. Henceforth the proposed method is implemented using Indian Sign Language INCLUDE dataset for ISL words achieved an average accuracy of 91%.