Skip to content

Latest commit

 

History

History

svm

README.md

Eye Blink using SVM

The SVM uses eye aspect ratio (ear) in consecutive video frames to predict eye blink.

Data

Eyeblink8 is used to train.

Preprocess

Get training data ready

Train

Be sure training data is prepared in eyeblink8 and run:

python3 preprocess.py

Then train svm:

from utils import *
from utils_frame_based import *
from train import *

# uses 7 frames to predict
X, y = build_svm_data('./train/training_set.pkl', 7)

norm_X = transform_svm_data(X)

# adjust on your need
params = {
    'kernel': ['rbf'],
    'C': [10],
    'gamma': ['scale'],
    'max_iter': [5000],
    'class_weight': [None],
}
svm = train(norm_X, y, params)

save_model(svm, 'svm_7.pkl')

Reference

Largely modified from Mustafa A. Hakkoz's kaggle: