-
Notifications
You must be signed in to change notification settings - Fork 4
/
README_HMM
30 lines (21 loc) · 1.29 KB
/
README_HMM
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
### Readme for the HMM (Hidden Markov Model) classifier in the gesture recognition application
The HMM is a stochastic Model for sequential Data, which uses markov model to model hidden states.
Each HMM represents a gesture in the application, every state (frame of a gesture) is represented
by an GMM (Gaussian Mixture Model)
The HMM-Module provides training and classifying given record data. It also can be used to plot the raw,
preprocessed and sample data from trained GMM.
### Usage:
Config parameters are explained in config/default.cfg. On first start a config/personal.cfg file is created. Changes can be made here.
Two already trained networks are provided and specified in "classifier.hmm.config.py"
You can choose:
0 all gestures provided by the application, trained with all data. gesture 0, 1, 2, 4, 6, 7
1 3 gestures trained with well formed data - classifying work kinda good.
gesture 0, gesture 1
gesture 5 changed to "Bottom-To-Top-One-Hand"
some commands for console:
u hmm - loads HMM module, with configured gesture set
c - start classification
l ds gesture_number - (Overloaded) starts plot class and show raw training data
right click - change mode: raw, preprocessed or gmm samples
mousewheel - change data index
0-7 - change gesture