UCF CAP 5937 - Dynamic Gesture Recognition Using Neural Networks

Unformatted text preview:

1Dynamic Gesture Dynamic Gesture Recognition Using Neural Recognition Using Neural Networks; A Fundament for Networks; A Fundament for Advanced Interaction Advanced Interaction ConstructionConstructionKlaus Boehm, Wolfgang Klaus Boehm, Wolfgang BrollBroll, Michael , Michael SokolewiczSokolewiczPresented by Ross ByersPresented by Ross ByersInteraction in Virtual SpacesInteraction in Virtual Spaces►►Gestures can provide clear interaction Gestures can provide clear interaction between humans, despite inherent between humans, despite inherent flexibility.flexibility.►►Gestures provide opportunity for use in Gestures provide opportunity for use in Virtual Spaces: Interacting with objects, as Virtual Spaces: Interacting with objects, as well as providing instructionswell as providing instructions►►Static gestures (postures) established.Static gestures (postures) established.2Interaction in Virtual SpacesInteraction in Virtual Spaces►►Static Gestures are limited.Static Gestures are limited.►►Requirement to hold pose tiringRequirement to hold pose tiring►►Exact manipulation is difficultExact manipulation is difficult►►No force feedbackNo force feedbackDynamic GesturesDynamic Gestures►►Natural Interaction is movement based, Natural Interaction is movement based, thus dynamic.thus dynamic.►►Time and movement introduce Time and movement introduce complications.complications.3Recognition of Dynamic GesturesRecognition of Dynamic Gestures►►The authors used a The authors used a KohonenKohonenFeature Map Feature Map (KFM), a type of Neural Network.(KFM), a type of Neural Network.►►Two layers Two layers ––Input and outputInput and output►►Unsupervised trainingUnsupervised training►►Output is a twoOutput is a two--dimensional grid of dimensional grid of neurons, where spatial proximity on the grid neurons, where spatial proximity on the grid is correlated with similarity.is correlated with similarity.PreprocessingPreprocessing►►Because of high dimensionality (30+ in this Because of high dimensionality (30+ in this example), the data must be preprocessed.example), the data must be preprocessed.►►‘‘VerticalVertical’’preprocessing collects information preprocessing collects information for each time stepfor each time step►►‘‘HorizontalHorizontal’’preprocessing filters and derives preprocessing filters and derives data.data.►►Recording of training data was best assisted Recording of training data was best assisted by a second person.by a second person.4First Recognition ApproachFirst Recognition Approach►►Direct MappingDirect Mapping►►Finds match for the best gestureFinds match for the best gesture►►Presents issues with longer and shorter Presents issues with longer and shorter gesturesgestures►►Introduces lagIntroduces lag►►Requires differing buffer sizesRequires differing buffer sizesSecond Recognition ApproachSecond Recognition Approach►►Gesture PartsGesture Parts►►Instead of Requiring Instead of Requiring ‘‘all at onceall at once’’recognition, recognition, recognize a library of subrecognize a library of sub--gestures.gestures.►►Simplest example would use equidistant timeSimplest example would use equidistant time--slices. This does not correctly model real behavior.slices. This does not correctly model real behavior.►►KFM used for part recognition only.KFM used for part recognition only.►►Second, specialized NN used for full gesture Second, specialized NN used for full gesture recognition.recognition.5ResultsResults►►Unfortunately, the authors did not share their Unfortunately, the authors did not share their accuracy data.accuracy data.►►They do however, reflect on processing time.They do however, reflect on processing time.►►The first approach required a reduction of input The first approach required a reduction of input data to run in real time, and was considered data to run in real time, and was considered ‘‘suitablesuitable’’for 10 gestures.for 10 gestures.►►The second approach allows for more The second approach allows for more preprocessing, thus improving performance even preprocessing, thus improving performance even with second neural with second neural


View Full Document

UCF CAP 5937 - Dynamic Gesture Recognition Using Neural Networks

Documents in this Course
Load more
Download Dynamic Gesture Recognition Using Neural Networks
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Dynamic Gesture Recognition Using Neural Networks and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Dynamic Gesture Recognition Using Neural Networks 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?