Menu
Home
Log in / Register
 
Home arrow Computer Science arrow Augmented and Virtual Reality
< Prev   CONTENTS   Next >

2 Related Works

This paper relies on hand tracking, anthropomorphic robots and data transmission over the web. In this Section, state-of-the-art approaches on these topics are briefly discussed.

2.1 Hand Tracking

Object tracking techniques can be divided into two main classes: invasive approaches, based on tools physically linked to the object (e.g., sensitized gloves [18] or markers [28]), and non-invasive approaches. The former are usually fast and computationally light, but also very expensive and cumbersome. The latter require more computational resources, but are based on low-cost technologies and do not require a physical link to the object to be tracked.

Non-invasive approaches proposed in literature (as in [22]) can be classified according to the kind of information in input they need (2D or 3D) and output they provide. Obviously, as real world life is embedded in a 3D universe, best performances are obtained when 3D features characterization is performed. Moreover, relying on 3D input information makes a visual system more robust and accurate [13].

Thanks to technology evolution though, it is nowadays possible to obtain reliable features extraction from confused backgrounds by trying to isolate and segment the object to be recognized (e.g., a human hand). In fact, 3D information may be obtained by depth maps that are automatically calculated by acquisition system using cheap RGB-D cameras.

Non-invasive approaches are classified into partial tracking and full tracking: tracking is defined as partial when it requires only information on a subset of input DoF [22] (e.g., only thumb and index in case of partial hand tracking), while it is said to be full when all the DoF are taken into account for computation. Of course, full tracking approaches are the best in terms of accuracy and robustness, but they also require a lot of computational resources [10]. Full tracking solutions can be divided into model-based and appearance-based approaches.

Model-based approaches [12] are based on a 3D model representing the object to be tracked. These approaches seek for the 3D model parameters that minimize the discrepancy between the appearance and 3D structure of hypothesized instances of the model and actual observations. The problem is usually solved using probabilistic optimization approaches [25] or evolutionary algorithms [20], leading to accurate results, but requiring a lot of computational resources.

Appearance-based techniques are based on special points (features) extracted from input data. An algorithm tries to directly map the extracted features to the hand configuration (i.e., the pose). Appearance-based algorithms are often implemented using machine learning [24] or database-retrieval [4] techniques. Learning how to map features and model configurations is the most computational intensive phase. Since this task is performed off-line, appearance-based techniques easily achieve real-time performances. The accuracy of these algorithms is strongly related to the quality of the training set or database, particularly to its variety and capacity to cover the set of poses.

2.2 Anthropomorphic Haptic Interfaces

Haptic devices elicit human perception through the sense of touch. Haptics therefore extends the communication channels for human-machine interaction, in addition to the typical senses, such as vision and hearing. Haptics includes wearable devices, such as gloves, and robotic devices, such as arms and hands. With respect to robotic hands, despite the significant progress in the last decades in electronic integrated circuits and in applied computer science, current systems still lack in dexterity, robustness and efficiency, as well as in matching cost constraints [9]. Examples of dexterous robotic hand for humanoid robotics are the Awiwi Hand [14] and the Shadow hand [26].

In prosthetics, electro-actuated hands have been commercially available since the early 70s: the major manufacturer is Ottobock (Austria), followed by other few companies. Recently two commercial prosthetic hands with greater Degree of Freedom (DoF) have been introduced to the market: the i-limb (developed by Touch Bionics in 2007) and BeBionic (developed by RSL Steeper in 2010) prostheses. Both hands are capable of different grasping patterns thanks to five individually-powered digits, but their functionality is limited by the passive movement of the thumb abduction/adduction joint. Most of current commercial prosthetic hands are very simple grippers with few active DoF and poor cosmetic appearance, however major research progresses are being achieved [21].

2.3 Transmission

Remote control of robotic actuators is nowadays a well studied task, investigated above all in surgery [3]. For what concerns the transmission of human movements to the robotic hand in this project, we evaluated different scenarios trying to maintain the infrastructure as simple as possible, as this will naturally lead to a simple, fast to develop and robust pattern of communication.

 
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
 
Subjects
Accounting
Business & Finance
Communication
Computer Science
Economics
Education
Engineering
Environment
Geography
Health
History
Language & Literature
Law
Management
Marketing
Philosophy
Political science
Psychology
Religion
Sociology
Travel