Hand gesture detection using connected Component and manhattan distance Transform on svm and cnn classifiers

Abstract

The automotive sectors and many consumer electronics divisions use the gesture-based machine operating system without any human interaction. Hence, the detection and recognition of hand gestures are very important in automotive industry. The present hand gesture recognition system is affected by its surrounding environments. This led to misclassification of hand gestures for automotive functioning system. The hand gesture can be detected and recognition under any background environment. Hence, there is a requirement for developing a hand gesture recognition system under different environment. newlineIn this thesis, automatic detection and classifications of hand gesture recognition system is proposed. It consists of pre-processing, transformation, feature extraction and classifications. The RGB image is converted into grey scale image as a pre-processing method and this image is converted into multi resolution image using Gabor transform. The features are extracted from Gabor transformed image and these features are classified using Adaptive Neuro Fuzzy Inference System (ANFIS) classifier. newlineGesture can be used as a tool of communication between computer and human. Hand gesture recognition has great value in sign language recognition. The Adaptive Histogram Equalization (AHE) method is used as enhancement method for improving the contrast of each pixel in an image. Connected Component (CC) analysis algorithm is used in order to segment the finger tips from hand image. The finger tip from each hand image is detected and noted as finger peak. In this thesis, the human hand gestures are detected and recognized using Convolutional Neural Networks (CNN) classification approach. This process flow consists of hand Region of Interest (ROI) segmentation using mask image, fingers segmentation, normalization of segmented finger image and finger recognition using CNN classifier. The proposed CNN architecture consists of 5 Convolutional layers and 1 fully connected layer with 1024 units newline