#Smartphone Gesture Recognize app
This repo is programmed for a UAV competition in SJTU 2015.05. The whole app can recognize a gesture, and then send instructions to the uav. And this repository extract the gesture recognition part. With a simple theory of pattern recognition. This APP can recognize up to 10 already-recorded gestures with 3-axis accelerators and cosine similarity algorithm. Wish these codes would help others.
Firstly, You should record your gestures with this app's instructions. Including input your gesture's id , perform the same gestures for times so that APP can remember the gesture's features.(You should keep tapping the screen while you are performing gestures, and release as soon as the gesture ends).
Secondly, tap the recognize button to enter recognize part. Tap the screen when performing gestures as before. Then this app will give the recognize result in both UI and logcat.
The main idea of this app is to compare the similarities between gestures with cos similarity algorithm , and use a 3-dimension vector to represent a gesture.
1.Represent a gesture: Take advantage of the 3-axis accelerator in Android devices, keep sampling the accelerate while screen is being tapped every 50ms, and map them into a vector).
2.Calculate the similarities value between the recorded gestures and the to-be-recognized gesture vector. The gesture will be considered the class of gesture that has the minimum similarity value. Normalization: To make vectors having different lengths , I normalize all vectors in to a 60 sample accelerates with linear interpolation algorithm.