Introduction
Driver drowsy detection dataset consists of both male and female drivers, with various facial characteristics, different ethnicities, and 5 different scenarios.
The videos are taken in real and varying illumination conditions. The scenarios contain BareFace(NoGlasses), Glasses, Sunglasses, Night-BareFace(Night-NoGlasses)
and Night-Glasses. Every video may contain 2 kinds of status: drowsy, non-drowsy. Each video is different situation with different status transition.
駕駛員疲勞檢測數據集包含了男性和女性駕駛員，涵蓋不同的面部特征、種族、以及5個不同的場景。
這些視頻在實際和不同光照的狀態下拍攝。場景中包含了BareFace(NoGlasses), Glasses, Sunglasses, Night-BareFace(Night-NoGlasses)和Night-Glasses,
每個視頻可能包含了2種狀態：drowsy, non-drowsy. 每個視頻都是在不同的狀態轉換下完成的。
--------------------------------------------------------------------------------------------------------
Camera Setting and Video Format
We used the active infrared (IR) illumination and acquire IR videos in the dataset collection.
The videos are in 640x480 pixels and 15/30 frames per second AVI format without audio.
[Notice] The videos in night_noglasses and night_glasses scenarios are 15 frame per second, and the videos in the other scenarios are 30 frame per second.
在這個數據集中我們使用主動紅外照明並獲取紅外視頻。
這個視頻集有640x480的像素，15/30fps，不帶聲音的AVI格式。
[注意] 這個視頻集在night_noglasses和night_glasses場景下是15fps，在其他場景中是30fps。
--------------------------------------------------------------------------------------------------------
Driver's Behaviors
Yawning : The driver opens his mouth wide due to tiredness.
Nodding : The driver's head falls forward when drowsy or asleep.
Looking aside : The driver turns his head left and right.
Talking & laughing : The driver is talking or laughing while driving.
Sleepy-eyes : The driver closes his eyes due to drowsiness while driving.
Drowsy : The driver looks like sleepy and lethargic.(including nodding, slowly blinking and yawning.)
Stillness : The driver drives normally.
Yawning(打哈欠): 駕駛員由於困倦張大嘴巴。
Nodding: 當疲勞和睡着時駕駛員向前低下頭。
Looking aside: 駕駛員向左或向右轉頭。
Talking & laughing: 在駕駛中，駕駛員說話或大笑。
Sleepy-eyes: 在駕駛中，駕駛員由於疲勞閉上眼睛。
Drowsy: 駕駛員看起來睡着和昏昏欲睡。（包括低頭，緩慢眨眼和打哈欠。）
Stillness: 駕駛員正常駕駛。
--------------------------------------------------------------------------------------------------------
Training Dataset
The training dataset provides 18 subject folders. Each subject will contain five different scenarios(noglasses, glasses, night_noglasses, night_glasses, sunglasses),
and each scenario will contain 4 videos with different situation and corresponding annotation files.
"yawning.avi" :The video includes yawning behaviors.
"slowBlinkWithNodding.avi" : The video includes sleepy-eyes and nodding behaviors.
"sleepyCombination.avi" : The video includes combination of drowsy behaviors, e.g. sleepy-eyes, yawning, nodding.
"nonsleepyCombination.avi" : The videos includes combination of non-drowsy behaviors, e.g. laughing, talking, looking aside.
[Notice] There is no video in Number-005 subject's night_glasses scenario.
訓練集提供了18個人的文件夾。每個人包含了5個不同場景(noglasses, glasses, night_noglasses, night_glasses, sunglasses)，
每個場景包含4個視頻，含有不同的狀況和相關的標注文件。
"yawning.avi" :包含了打哈欠的行為.
"slowBlinkWithNodding.avi" : 包含了睡着的眼睛和低頭行為。
"sleepyCombination.avi" : 包含有各種疲勞行為，如睡着的眼睛、打哈欠、低頭。
"nonsleepyCombination.avi" : 包含了不疲勞的各種行為，如大笑、說話、四處看。
[注意] 在005中沒有night_glasses場景。
--------------------------------------------------------------------------------------------------------
Evaluation Dataset
The evaluation dataset provides 4 subject folders. Each subject will contain 5 videos with different scenarios and corresponding annotation files. The videos would perform
various situations with different drowsy, non-drowsy status transition.
驗證集提供了4個人的文件夾。每個人包含5個在不同場景下的視頻和相關的標注文件。視頻展示了不同狀況，其中有疲勞、非疲勞的狀態轉換。
--------------------------------------------------------------------------------------------------------
Annotation
There are four annotations of each video and a single digit is used to indicate the status of the frame.
[video name]_drowsiness.txt : 0 means Stillness and 1 means Drowsy.
[video name]_head.txt : 0 means Stillness, 1 means Nodding and 2 means Looking aside.
[video name]_mouth.txt : 0 means Stillness and 1 means Yawning and 2 means Talking & Laughing.
[video name]_eye.txt : 0 means Stillness and 1 means Sleepy-eyes.
每個視頻中有4個標注，一個數字用來表示幀的狀態。
drowsiness: 0---Stillness, 1---Drowsy.
head: 0---Stillness, 1---Nodding, 2---Looking aside.
mouth: 0---Stillness, 1---Yawning, 2---Talking & Laughing.
eye: 0---Stillness, 1---Sleepy-eyes.
--------------------------------------------------------------------------------------------------------
Evaluation Function
The evaluation function consists of both Matlab and C++ version and is provided for user to evaluate the drowsiness performance on the evaluation dataset. It supports any environment that contains Matlab or C++ compiler.The function allows:
(1) evaluate the drowsy results(*_drowsiness.txt) for the evaluation dataset;
(2) output the accuracy score for both indivial video and overall video.
In order to use the evaluation function with your detection algorithm, you will have to make sure the format of your algorithm and compared groud-truth is indentical.
For more details, please find the instruction in the function.