Gesture Tracking using Hidden Markov Models

Rad se bavi detekcijom gesti pomoću matematičkog modela naziva skriveni Markovljev model. Pomoću kamera robota i računalnog vida na robotu se dobivaju podaci koji se naknadno kvantiziraju i obrađuju. Nakon toga pomoću velikog broja podataka koji označavaju gestu trenira model geste te stvara granica za prepoznavanje gesti. Pomoću navedene granice i modela omogućeno je detektiranje gesti u realnom vremenu. Testirana je detekcija gesti nad trima gestama definiranima u protokolu o dijagnostici autizma.The main subject of this bachelor thesis is gesture detection by the mathematical implementation of hidden markov models.Through robot’s camera and computer vision we get data that’s additionally processed and quantized.Thanks to a large set of data that corresponds to the defined gesture through learning we define a model for the gesture and the gesture threshold.By using the defined model we enable realtime gesture detection. The gesture detection has been tested on three gestures defined through te autism diagnostic protocol