Static Hand Gesture Recognition Using Multi-Layer Neural Network Classifier on Hybrid of Features
Kolawole G. Akintola, Jadesola Adejoke Emmanuel · American Journal of Intelligent Systems · 2020
Hand gesture recognition has gotten so many areas of application such as in human-computer interaction, hearing impaired communication and systems control. Recognizing gestures in videos however is a challenging task. Many techniques and features have been adopted in the literature but some of the methods still need to be improved upon. There are basically two types of gestures: the static and the dynamic gestures. In this work, a computer vision-based system for recognition of static hand gesture is proposed using a fusion of the histogram of oriented gradient and the Hu invariant moments as features. The proposed system consists of three phases: preprocessing, feature extraction and classification. The images are first pre-processed using segmentation and morphological operations. The histogram of oriented gradient and the Hu invariant moments are then extracted as features. The extracted features are concatenated and passed as input into a multilayer neural network model to recognize the static hand gesture. The proposed system is implemented and tested on the hand gesture database collected online. The model was trained using 500 features which consist of 20 gestures each from 25 gesture types. The model is then tested with another 500 gestures which also consist of 20 gestures each from the 25 gesture types. The experimental results show that the proposed system is able to recognize the static gestures with accuracy of 96.4%.