Democritus University of Thrace (DUTH)
Ανίχνευση κινήσεων και εφαρμογές τους στην επικοινωνία ανθρώπου-υπολογιστή
Abstract
dc:descriptionNowadays, intelligence computing plays an important role in almost every significant aspect of everyday life. Education, entertainment, communication, business, health and art are some of the fields of human activity that are modified and enriched by the use of a variety of applications. Nonetheless, the interaction with computers is still based on traditional means, i.e. keyboard, mouse and touch screen, that fail to be intuitive. Therefore, researchers have become more interested in taking advantage of the inherent human means of communication, such as facial expressions, eye blinking, gazing, body posture, body movement and gestures. In this context, the thesis presents new and innovative image and video processing techniques that detect, analyze and recognize gestures. Then the techniques are applied in the development of three different human - computer interaction systems.Specifically, the thesis proposes techniques concerning the: real time hand detection in a complex background, hand posture analysis via a self-organized neural network and hand posture recognition.The hand detection in a complex background technique consists of four main stages: motion detection, skin color detection, morphology description and combination of the extracted information. Motion detection is based on image differencing, which defines the region of motion, and on background subtraction, that extracts the moving objects. Skin color detection is accomplished through a color classification technique that utilizes online color training, so as to adapt better to the variety of the lighting conditions and the user’s skin color. Morphology description of the detected hand in the previous frame is used to specify the possibility of a pixel in the current frame to belong to the hand. Finally, the derived information is combined. The main innovation of this technique is that the combination is accomplished in a region based manner, in order to take into consideration the information that is provided not just from a single pixel or a window neighborhood, but from an arbitrarily shaped area of pixels with similar color. This is achieved through a color reduction algorithm that segments the input frame in regions of similar color. To each one of the resulting areas, a rating system is applied. This system uses motion, color and morphology data in order to define the possibility of a region to be part of the hand. The output is a grayscale image that through binarization results to the detected hand. The hand posture analysis is based on the approximation of the topology of the hand by applying the Self Growing and Organized Neural Gas (SGONG) on the segmented (binary) image. The SGONG is an unsupervised neural classifier that consists of two layers, i.e. the input and the output layer. The dimensions of the input space and the output lattice of neurons are always identical. Thus, the structure of neurons in the output layer approaches the structure of the input data. In the proposed technique, the input data of the SGONG are the coordinates of random samples of the hand. As a result, the output neurons and their connections approximate its morphology. Based on the SGONG output lattice, the palm of the hand, the palm center, the hand slope and the number of raised fingers are determined. Finally, robust features that describe successfully the properties of the fingers are extracted. The recognition process consists of three stages: the off–line calculation of the features’ distributions, the likelihood based classification, the final classification. It is based on the choice of the most probable finger combination of a set of feasible gestures. This is accomplished by classifying the raised fingers into five classes (thumb, index, middle, ring, little) according to their features. The classification depends on the probabilities of a finger to belong to the above classes. The probabilities derive from the features’ distributions.The primary outcome of this thesis is the combination of the above proposed techniques in the development of a vision-based hand posture recognition system in a complex background. As an extension, a dynamic gestures recognition system and a recognition system based on depth sensing technology are created. All of the three systems are innovative and meet the expectations of a natural and simple communication, the goal of the latest research efforts in the scientific field of human-computer interaction systems.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stergiopoulou, Aikaterini
- Στεργιοπούλου, Αικατερίνη
Subjects
dc:subject × 10- Αναγνώριση χειρονομιών
- Ψηφιακή επεξεργασία βίντεο
- Επικοινωνία ανθρώπου-υπολογιστή
- Gesture recognition
- Digital video processing
- Human-computer interaction
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/39874