Mouse gesture recognition project using HMM.
The currently working classifier is based on Discrete HMM trained on mouse coordinates
converted to angles between models and quantized into specific ranges of angles.
The implementation exploits vectorized numpy operations to speed up training and inference time.
Future work is to consider :
1)Further exploitation of multiple training sequences based on Rabiner's approach.However the
model accuracy is > 99.5%. But this may be useful for more complex problems.
2)continuous GMM availability.
3)GUI improvements and adding features.
Note : The project is modularized (ie: quantization module or HMM module...etc can be used
seperately.
Dependencies:
1)Python 3.5
2)Numpy
3)Tkinter
Although you can collect your dataset easily through the interface and train the
model in less than half a minute, There is an already trained model for quick testing
** This trained model classifies these 7 gestures :
0 -----> Nike sign
.->--->---|
|
1 -----> | |
|
|--<---<----.
2 -----> |
|
|
3 -----> <
4 -----> >
5 ------> |--<----<--.
| |
| |
|->--->----|
6 -------> equilateral Triangle (Top then left side(up-down)
then base then right side(down-up)