Driver_Drowsiness_Detection

Marmara University - Computer Engineering Graduation Project

Developed by

• Ayşenur YILMAZ @Aysenuryilmazz
• Mahmut AKTAŞ @mahmutaktas
• Mustafa Abdullah HAKKOZ @mustafahakkoz

Driver_Drowsiness_Detection

Driver drowsiness is one of the causes of traffic accidents. According to the statistics; highway road crashes hold 11.09% of the total number of accidents. There are several reasons of drowsy driving such as: a lack of quality of sleep, may be overnight driving or having sleep disorders e.g. sleep apnea. However; all people should know that: People can not fight against to sleep.

Using Image Processing and both classical and brand-new Machine Learning techniques such as SVM, k-NN, XGBoost, and also LSTM; we are trying to predict beforehand the driver’s drowsiness and warn him/her by an alert before any crash happened.

Aims of the Project:

Real-time Application:

Pipeline:

Hand-made Features:

Frame Based Models Classification Results:

Sequential Models Regression and Classification Results:

Technologies that We Used:

Online links of notebooks and input/output files:

1.construct-df

2.feature importances

• feature-importances-wrf.ipynb

3.normalization

• final_step2_scaled.pkl

4.ML classification

5.Process RLDD

• merged_normalized_scaled.pkl
• rldd_normalized_scaled.pkl

6.experiments

frame-based models:

• adaboost_RLDD.ipynb
• adaboost_MERGED.ipynb
• bagging_MERGED.ipynb
• bagging_RLDD.ipynb
• cat_MERGED.ipynb
• cat_NTHU.ipynb
• cat_RLDD.ipynb
• cat_MERGED.ipynb
• dt_MERGED.ipynb
• dt_RLDD.ipynb
• extratrees_MERGED.ipynb
• extratrees_NTHU.ipynb
• extratrees_RLDD.ipynb
• gradientboost_RLDD.ipynb
• gradientboost_MERGED.ipynb
• knn_MERGED.ipynb
• knn_RLDD.ipynb
• lgbm_MERGED.ipynb
• lgbm_NTHU.ipynb
• lgbm_RLDD.ipynb
• logisticregression_NTHU.ipynb
• logisticregression_MERGED_RLLD.ipynb
• nb_MERGED.ipynb
• nb_RLDD.ipynb
• rf_MERGED.ipynb
• rf_RLDD.ipynb
• simple-thresholding_NTHU_RLDD_MERGED.ipynb
• svm_MERGED.ipynb
• svm_NTHU.ipynb
• svm_RLDD.ipynb
• xgboost_RLDD.ipynb
• xgboost_MERGED.ipynb

sequential models:

• data-preparation_NTHU.ipynb
• data-preparation_RLDD.ipynb
• lstm-vanilla_NTHU_classification.ipynb
• lstm-vanilla_NTHU_regression.ipynb
• lstm-stacked_NTHU_classification.ipynb
• lstm-stacked_NTHU_regression.ipynb
• bi-lstm_NTHU_classification.ipynb
• bi-lstm_NTHU_regression.ipynb
• cnn-lstm_NTHU_regression.ipynb
• cnn-lstm_NTHU_classification.ipynb
• conv-lstm_NTHU_regression.ipynb
• conv-lstm_NTHU_classification.ipynb

7.demo app

backend

input

• extra_best_estimator_compressed.pkl
• lstm_vanilla.h5
• lstm_vanilla_v2.h5
• scaler.sav
• shape_predictor_68_face_landmarks.dat

8.alternatives

I3D

• i3D.ipynb

blink detection

• eye-blink-detection-1-simple-model.ipynb
• eye-blink-detection-2-adaptive-model.ipynb
• eye-blink-detection-3-ml-model-part1.ipynb
• eye-blink-detection-3-ml-model-part2.ipynb
• eye-blink-detection-4-comparison.ipynb