Visual Recognition of Static/Dynamic Gesture: Gesture-Driven Editing System

Human motion: modeling and recognition of actions and interactions

“…Using a kinematic hand model for hand tracking, Rehg and Kanade [49] took a least squares approach to estimate and compare stored hand models to input hand-image sequences. Min et al [33] proposed a combined approach that uses both a static representation for hand gesture and a dynamic representation for arm mo-tion for human-computer interaction.…”

“…For example, Bobick et al [52] used HMM to classify hand motions. Min et al [33] used both static representation and dynamic representation for hand gesture recognition. Yang et al [56] applied HMM to classify human action intent and to learn human skills.…”

Hand gesture modelling and recognition involving changing shapes and trajectories, using a Predictive EigenTracker

“…While some systems use multiple cameras (Hongo et al, 2000), it is more challenging to use a single uncalibrated camera. Many approaches focus primarily on motion/trajectory information (e.g., Yeasin and Chaudhuri, 2000;Min et al, 1999), or shape information (e.g., Triesch and Malsburg, 2002), while some consider both. We review their characteristics, below.…”

“…Section 2 describes this in a nutshell.) For the temporal modelling and recognition, most systems use Finite State Machines (FSMs) (e.g., Yeasin and Chaudhuri, 2000), or the more general Hidden Markov Models (HMMs) (e.g., Nam and Wohn, 1997;Kapuscinski et al, 2001;Min et al, 1999;Ng and Ranganath, 2002). HMMs have the disadvantage of a very elaborate training procedure to be effective.…”

A system for person-independent hand posture recognition against complex backgrounds

“…Second, the system must be able to deal with complex, cluttered backgrounds, making segmentation of the gesturing hand difficult. Most work in the field tries to circumvent the problem by either using colored markers, or by requiring the background to be static, or by requiring the hand to be the only skincolored object in the scene [20], [3], [7], [6], [12], [9], [14], [13]. In contrast to this, our goal was a system that works in relatively unconstrained environments, where segmentation based on primitive cues is not always possible.…”