Dynamic Hand Gesture Recognition with Self-co-articulation removal in Indoor and Outdoor Environment

In a continuous video sequence, recognition of dynamic hand gestures is a challenging task and more difficulties arise when the recording environment is also dynamic in nature. This paper proposes a system which solves the problem of the complex environment. It also tries to solve the problem of presence of unwanted strokes during the formation of the gesticulation pattern. Gestures used in this paper are recorded in both indoor and outdoor environment. Detection is done by a two-level algorithm using skin color and motion information. Hand tracking is done using modified tracker which overcomes the difficulties arising in existing KLT due to movement of hand. Incremental feature selection is utilized to use only those features for recognition which provides best results in terms of accuracy. SVM is utilized for recognition of the gestures and the recognition accuracy has been compared with deep learning algorithm also. It has been observed that 96.67% recognition accuracy is being achieved using deep learning which is better than existing machine learning algorithm.