Gestural Interfaces

2021-04-11
1 min read

To design a gestural interfaces

Creating a gesture set and defining a gesture-command mapping

Laban feature

Laban Movement Analysis (LMA)

The LMA system provides models for the interpretation of movement, its function and its expression through 4 components: − Body (what) − Effort (how) − Space (where) − Shape (relation with the environment) Effort has 4 Factors thought as a continuum with 2 opposite ends: − Weight : strong, light − Time : quick, substained − Space : flexible, direct − Flow : bound , free

Choosing the device

Structured light

Depth from focus: the further the blurrer Depth from stereo: view different from different angle

Time of Flight (TOF) imaging:

Use phase difference between the radiated and reflected IR wavesis to compute the distance of objects.

Building a recognizer

Rubine algorithm

There is a set of C gesture classes (from 0 to C-1). Each class is specified by example gestures. Given an input gesture g, determine the class to which g belongs.
Statistical gesture recognition is done in 2 steps: first a vector of features f is extracted from the input gesture. Then, f is classified as one of the C possible gestures.
Recognition of 2D single-stroke gestures with a clear start and a clear end

$ Family (Template-based)

$1 one stroke recoginzer
对输入的手势轨迹进行重新采样,将采样轨迹中心点与x轴的连线旋转到0度,将旋转后的手势缩放到标准的正方形大小并进行手势平移。然后,对处理过的手势和搜索匹配模板使用“黄金选择搜索”(GSS),得到每个模板的最佳匹配得分。得分最高的手势是最终识别的手势

  1. Resampling the point path (gesture trajectory)
  2. Rotate once based on the « indicative angle » (parallel to x axis)
  3. Scale and traslate
  4. Recognition ( Golden Section Search (GSS))
    Limitations:
  5. No way to distinguish a rectangle by a square
  6. No way to distinguish an ellipse by a circle
  7. No way to distinguish the orientation of an arrow

DTW (Template-based)

Machine Learning approaches: kNN, SVM…

Providing a teaching method

When

Feedback vs feedforward information: − Feedback: informs about the effects of the actions already performed
− Feedforward: provides information prior to any action, i.e. showing or anticipating the possible future actions.

Evaluating your gestural interface