Emblematic Gestures Recognition

Abstract

This paper presents a framework allowing emblematic gestures detection, segmentation and their recognition for human-robots interaction purposes. This framework is based on a new coding of arms’ kinematics reflecting both the muscular activity of the performer and the appearance of arm seen by the recipient when a gesture is performed. Following that, gestures can be seen as sequences of torques activations leading arm’s parts to express a comprehensive meaning. In addition, these sequences have very stable topologies and shapes regardless to performers. This facilitates the generalization of the recognition process with a minimalistic learning effort for online usages. Promising results were obtained for a set of 5 classes of gestures performed by 19 different persons.

Gestures vs. Gesticulations: Change Point Models Based Segmentation for Natural Interactions

“…HMM use a time based representation that also works as implicit segmentation. Support Vector Machine [1], neural networks [15] and dynamic time warping [16] also achieve various level of success on segmented data. However, most of these techniques are interchangeable, benchmarks [17] tends to find only small difference in performance between the different techniques.…”

“…Most of these approach are based on a fixed length analysis window. In [1], they used a representation that transform a pose sequence of arbitrary length into a fixed length set of scalar representing the arm position and the presence of cycles. Such an approach is extremely sensible to the gesture segmentation.…”

“…In many works , those two process are treated simultaneously The segmentation of gestures has not been studied as extensively as its recognition counter part. As shown in [3], segmentation methods falls in one of the following categories : extrema detection [1], [4], where physical characteristics like speed or position are used as triggers. Hidden Markov Model [5]- [7] can also be trained specifically for detecting transition between strokes, but the lack of a baseline gesture grammar prevents such approach from perfect generalisation.…”

“…In the following, we state the nature of arms movements and detail accordingly, the stochastic technique allowing to segment time series describing arms movements into consistent segments. In the second part, we show how we reuse a previous work in the recognition phase [1] and the assessment of the segmentation/recognition phases on public datasets. In the last part, we describe the scenario we built to allow novice users to built their own interface with a minimum of efforts.…”

Characterisation of gestural units in light of human-avatar interaction

“…Indeed, the preparation of motion consists of a ballistic motion that brings the arm(s) to the core of the motion [8]. This ballistic motion involves acceleration followed by deceleration as the final position is approached, then symmetrical acceleration and deceleration to the first set, and a return to the resting position.…”