Enhanced hand-gesture recognition by improved beetle swarm optimized probabilistic neural network for human–computer interaction

The dynamic hand gesture recognition is a new research area due to the emerging nature of pervasive somatosensory devices, which has been gained more attention and been broadly utilized in human–computer interaction (HCI). This paper tactics to develop an intelligent hand gesture recognition model through considering the use of hand gestures for HCI. The user-friendly, less intrusive, intuitive, and more natural HCI for controlling an application by executing hand gestures are presented. The overall models cover diverse steps like data collection, pre-processing, segmentation, feature extraction, and recognition. The hand gestures videos from different benchmark sources are collected, and further, the pre-processing is performed by median filtering. Further, the goal of the segmentation is to evaluate the temporal hand trajectories from the detected hand poses, which is carried out through the adaptive region-based active contour (ARAC) method based on the meta-heuristic basis by the opposition strategic velocity updated beetle swarm optimization (OSV-BSO). Feature extraction is the next step, which is done by combining a set of features like oriented FAST and rotated BRIEF (ORB), histogram of oriented gradients (HOG), and regionprops, which are further given to the principal component analysis (PCA) for dimensionality reduction. With these relevant features, the recognition phase is accomplished by the optimized probabilistic neural network (PNN). To improvise the existing performance of PNN as a deep learning model, the weights in training are tuned based on the meta-heuristic basis by the OSV-BSO. The objective model of the optimized PNN is to minimize the error between the desired and actual outcomes. Finally, the designed “gesture recognition approach” makes the HCI process more user-specific and intuitive, which is proved by comparing it over the existing approaches.