Automatic hand gesture recognition using hybrid meta-heuristic-based feature selection and classification with Dynamic Time Warping

Abstract

Abstract Of late, the research world has been vigorously involved in, inventing strategy and techniques to improve the spontaneity of Human Computer Interaction (HCI). Gesture recognition is one of the most probable techniques in this area. The eventual aim here is to introduce an intelligent system for hand gesture recognition in both static and dynamic area, which is still a challenging point due to the lag of valuable beneficial methods. The main intent of this paper is to implement an efficient hand gesture recognition model considering both static and dynamic datasets for Indian Sign Languages (ISL). In static type, images are taken for processing, whereas video frames are used for processing the dynamic type. The proposed recognition model involves five main steps “(a) Image pre-processing, (b) gesture segmentation, (c) Feature extraction, (d) Optimal Feature Selection, and (e) Recognition”. In the pre-processing phase, greyscale conversion and histogram equalization are performed. The pre-processed image is subjected to the segmentation process, where the Active Contour model and Canny Edge Detection is implemented. In the feature extraction phase, both the contour image, and the edge detected image is deployed, in which Histogram of Oriented Gradients (HOG) features are extracted from the contour image, and Edge Oriented Histogram (EOH) features are extracted from edge detected images. To reduce the dimension of HOG, and EOH features, Principle Component Analysis (PCA) is applied. Further, the region props features are extracted for both contour and edge detected image. Finally, all these features are summed, and the optimal feature selection process performs here to select the unique feature giving different information with less correlation. Finally, the recognition classifier called Neural Network (NN) is adopted, where the new training algorithm is used to update network weight. Dynamic Time Warping (DTW) method helps to remove the repeated frames in the video and to reduce the time consumption of testing. In both feature selection and classification, a hybrid algorithm Deer Hunting-based Grey Wolf Optimization (DH-GWO) is used for selecting the features and weight update in NN as well. Hence, the integration of a hybrid meta-heuristic algorithm is highly efficient for recognizing the characters for images and words for videos with high recognition accuracy.