The invention provides a three-dimensional dynamic finger text input
system and method based on a depth image. The method comprises the following steps: firstly, obtaining 3D coordinate information ofhand contour information and
hand joint points; matching with a specific static gesture, extracting a descending fingertip amplitude characteristic, and judging a knocking action; obtaining all possible words by knocking finger sequences; and knocking a 3D coordinate of a
tail end of a finger, knocking finger information and a
language model based on the word frequency, obtaining candidate wordsby using a Bayesian model, calculating the confidence of a word
phrase with the greatest probability relative to the word, obtaining a maximum probability
phrase of the word, and finally, and finally,outputting the candidate words and possible phrases. Specific gestures are recognized, and the output candidate words and the phrases are selected and operated, or the selected words are deleted. Thethree-dimensional dynamic finger text
input method based on the depth image provided by the invention improves the speed and accuracy of air text input.