1 / 9

Klaus Boehm, Wolfgang Broll, Michael Sokolewicz Presented by Ross Byers

Dynamic Gesture Recognition Using Neural Networks; A Fundament for Advanced Interaction Construction. Klaus Boehm, Wolfgang Broll, Michael Sokolewicz Presented by Ross Byers. Interaction in Virtual Spaces. Gestures can provide clear interaction between humans, despite inherent flexibility.

diallo
Download Presentation

Klaus Boehm, Wolfgang Broll, Michael Sokolewicz Presented by Ross Byers

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Dynamic Gesture Recognition Using Neural Networks; A Fundament for Advanced Interaction Construction Klaus Boehm, Wolfgang Broll, Michael Sokolewicz Presented by Ross Byers

  2. Interaction in Virtual Spaces • Gestures can provide clear interaction between humans, despite inherent flexibility. • Gestures provide opportunity for use in Virtual Spaces: Interacting with objects, as well as providing instructions • Static gestures (postures) established.

  3. Interaction in Virtual Spaces • Static Gestures are limited. • Requirement to hold pose tiring • Exact manipulation is difficult • No force feedback

  4. Dynamic Gestures • Natural Interaction is movement based, thus dynamic. • Time and movement introduce complications.

  5. Recognition of Dynamic Gestures • The authors used a Kohonen Feature Map (KFM), a type of Neural Network. • Two layers – Input and output • Unsupervised training • Output is a two-dimensional grid of neurons, where spatial proximity on the grid is correlated with similarity.

  6. Preprocessing • Because of high dimensionality (30+ in this example), the data must be preprocessed. • ‘Vertical’ preprocessing collects information for each time step • ‘Horizontal’ preprocessing filters and derives data. • Recording of training data was best assisted by a second person.

  7. First Recognition Approach • Direct Mapping • Finds match for the best gesture • Presents issues with longer and shorter gestures • Introduces lag • Requires differing buffer sizes

  8. Second Recognition Approach • Gesture Parts • Instead of Requiring ‘all at once’ recognition, recognize a library of sub-gestures. • Simplest example would use equidistant time-slices. This does not correctly model real behavior. • KFM used for part recognition only. • Second, specialized NN used for full gesture recognition.

  9. Results • Unfortunately, the authors did not share their accuracy data. • They do however, reflect on processing time. • The first approach required a reduction of input data to run in real time, and was considered ‘suitable’ for 10 gestures. • The second approach allows for more preprocessing, thus improving performance even with second neural network.

More Related