Skip to content

tuanhm11299/Drowsiness-Detection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Đề tài: Phát hiện hiện tượng buồn ngủ

Môn: Máy học trong thị giác máy tính - Lớp CS332.K11.KHCL

Giảng viên hướng dẫn:

Tiến sĩ Mai Tiến Dũng

Sinh viên thực hiện:

Huỳnh Minh Tuấn - 17521212

Nguyễn Thanh Tú - 17521201

Hoàng Ngọc Quân - 17520934

Link bài báo cáo của nhóm:

https://www.overleaf.com/3764583157vxbnpkssqgmk?fbclid=IwAR3F-vDtqWcmXeszBrFn9bGJGtoKg1HLtlwQKMH71z9jzqq2vm5_3VqvcAQ

Đồ án được thực hiện Ubuntu 18.04 (LTS) với tensorflow phiên bản từ 1.8 trở đi

This proposed temporal model uses blink features to detect both early and deep drowsiness with an intermediate regression step, where drowsiness is estimated with a score from 0 to 10.

Instruction to Run the Code:

THESE CODES WERE APPLIED ON THE UTA-RLDD DATASET

You can refer to the comments inside each .py file for more detailed information

0- Make sure all .py files are downloaded then install all the required packages. You can refer to the following link for a short instruction on installing some of the required packages like dlib:

https://www.pyimagesearch.com/2017/03/27/how-to-install-dlib/

Or for the conda environment you can use the following command lines:

~$ conda install -c anaconda tensorflow-gpu

~$ sudo apt-get install build-essential cmake

~$ sudo apt-get install libgtk-3-dev

~$ sudo apt-get install libboost-all-dev

~$ conda install -c anaconda scipy

~$ wget https://bootstrap.pypa.io/get-pip.py

~$ conda install -c menpo dlib

~$ conda install -c conda-forge scikit-image

~$ pip install imutils

~$ conda install scikit-learn

~$ conda install -c conda-forge opencv

1-Chạy file Blink_Video.py:

This file is fed by the input video(the directory should be given to the path variable). Then, it detects the blinks and outputs four features of all blinks in a text file.

("Trained_SVM_C=1000_gamma=0.1_for 7kNegSample.sav" is used for blink detection.)

*Use the link below to download "shape_predictor_68_face_landmarks.dat"

https://drive.google.com/open?id=1nrfc-_pdIxNn2yO1_e7CxTyJQIk3A-vX

"shape_predictor_68_face_landmarks.dat" is the pre-trained facial landmark detector inside the dlib library.

2-Chạy file Preprocessing.py

This file gets three text files (blink features in three drowsiness levels) as the main input and preprocesses them for the subsequent steps. The outputs are .npy files.

For convenience, these .npy files ({Blinks, BlinksTest, Labels, LabelsTest}_30_FoldX.npy) are provided for each X as the test fold used for five fold cross validation. For example Blinks_30_Fold4.npy is the training set consisted of all the folds except fold 4, and BlinksTest_30_Fold4.npy is the data from fold 4. If decided to apply this method to a different dataset, then the hard coded "start_indices" array in Training.py should be adjusted accordingly. More info about "start_indices is mentioned in the Training.py". Finally, to clarify, these .npy files are generated from step 1 and 2 on the UTA-RLDD dataset so one might decide to generate their own .npy files to train.

3-Chạy file Training.py:

This code is used to train based on the .npy files generated in step 2. The model details and hyperparameters are all set here. This code is also used for testing. Here, one fold from the dataset (UTA-RLDD in this case) is picked as the test fold and the other four are used for training. The output is the training and test results and accuracies based on the pre-defined metrics in the paper.

For convenience, five pre-trained models are provided, where each model used one of the folds as the test set in a five fold cross validation.

These three files are pre-trained models for the training session of fold X, where fold X had been used as the test fold:

my_modelX.data-00000-of-00001

my_modelX.index

my_modelX.meta

NOTE: References used for each code are mentioned on top of each code.

Nguồn nhóm tham khảo:

Đồ án của nhóm được dựa trên 1 bài báo của Reza Ghoddoosian về việc xây dựng mô hình phát hiện buồn ngủ sớm trên các phương tiện giao thông chỉ với camera thông thường trên điện thoại hoặc camera máy tính. Tìm hiểu thêm tại: https://arxiv.org/abs/1904.07312

Link đến UTA-RLDD dataset:

https://sites.google.com/view/utarldd/home

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages