A selective region-based detection and tracking approach towards the recognition of dynamic bare hand gesture using deep neural network

GCR‐Net: A deep learning‐based bare hand detection and gesticulated character recognition system for human‐computer interaction

“…Most researchers 2,3,5‐11 experimented and proved their claim in a controlled environment for bare‐hand detection for detecting the bare hand. Initially, researchers 6 used skin‐color information and Viola Jones's algorithm to locate the skin color pixels for bare‐hand detection.…”

“…However, the work is experimented with in the simple‐fixed environment (excluding illumination, pattern, and occlusion kinds of variations) and reported an accuracy of 96.11%. To investigate and examine the performance of DCNN models, researchers 2,3 designed a character recognition model to recognize 58/60 gestures using AlexNet as the backend network. They reported that most of the characters were misclassified due to a similar pattern in gesticulation style.…”

“…Character recognition has gained much attention in recent years due to its potential applications 1‐10 like document analysis, language translation, number plate recognition, text‐to‐speech converter, air‐writing, and virtual text entry systems. Particularly, gesticulated character recognition is still needed to be explored further.…”

“…Particularly, gesticulated character recognition is still needed to be explored further. Few researchers 2‐5 worked to design gesticulated characters recognition system due to its complexity in bare‐hand detection and gesticulated trajectory recognition. The initial stage of these systems is bare‐hand detection, which should be implemented precisely.…”

“…The ImageNet large‐scale visual recognition challenge (ILSVRC) 12 has provided an explorer to detect and classify the object using deep learning. It encouraged researchers 13‐18 to design and develop numerous bare‐hand detection models (e.g., deep convolutional neural network (DCNN), region‐based convolutional neural networks (RCNN), and single‐shot detector (SSD)) in various fields such as air‐writing, 11 laparoscopic surgery, 13 automatic driver system, 14 gesture recognition 2 and so forth. The deep learning models have the ability to extract more robust features directly from the image and reduce the manual workload of feature extraction.…”