| To the right are the specific components of the dynamical gesture recognition and control system from an architectural and implementational viewpoint. In module G, a human moves a flashlight against a black background to create a gesture. Sensor module S uses a vision system to detect the gesture and outputs position and velocity information to module P. The predictor module P consists of seeded parameter predictor bins tuned to specific gestures. A final gesture classification is sent to module T, which transformed the gesture to a reference trajectory for an attached actuated mechanism, R, to track. |
|
|
To the left is an illustration of me generating a large slow circular gesture (part of module G). A total of 24 oscillating circle and line motion gestures are identifiable by this system with a 90% overall recognition rate. The system was also expanded to identify non-linear "come here" motions. |
|
Predictor module P has a bank of predictor bins, as shown to the right. Each bin contains a dynamical system model with parameters preset to a specific gesture, with the data decoupled in x and y. Each bin's model, which has parameters that tune it to a specific gesture, is used to predict the future position and velocity of the motion. The bin which most accurately predicts the gesture's next motion is the best match. |
|