Gesture Recognition using Cognitive AI and Deep Learning

Real-time hand tracking differs greatly from RGB video tracking of frequently monitored objects. Because it is a biological object, the hand experiences both physical and behavioral changes as it moves. If a basic RGB camera is used to record the video, the hand also occupies a very small portion of the image frame. It has an unpredictable pattern of movement, which greatly degrades the image quality. We provide a hybrid architecture in this chapter to monitor hand motion in RGB sequences of video. The framework incorporates the distinctive characteristics of the Scale-Invariant Features Transform (SIFT) method and Residual Network-based Rapid Region-based cognitive network (Faster R-CNN). Deep neural networks' discriminative learning capacity and SIFT's hand-crafted features' quick detection speed enhance this combo. As a result, our online technique adjusts to variations in hand movement that occur in real time and has a high level of cognitive hand trajectory detection efficiency. The empirical findings presented in this research show that the method is capable of handling both the inherent and extrinsic difficulties involved in the visual observation of hand motions in RGB films.