Driver drowsy detection dataset

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.

驾驶员疲劳检测数据集包含了男性和女性驾驶员，涵盖不同的面部特征、种族、以及5个不同的场景。

这些视频在实际和不同光照的状态下拍摄。场景中包含了BareFace(NoGlasses), Glasses, Sunglasses, Night-BareFace(Night-NoGlasses)和Night-Glasses,

每个视频可能包含了2种状态：drowsy, non-drowsy. 每个视频都是在不同的状态转换下完成的。

--------------------------------------------------------------------------------------------------------

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.

在这个数据集中我们使用主动红外照明并获取红外视频。

这个视频集有640x480的像素，15/30fps，不带声音的AVI格式。

[注意] 这个视频集在night_noglasses和night_glasses场景下是15fps，在其他场景中是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): 驾驶员由于困倦张大嘴巴。

Nodding: 当疲劳和睡着时驾驶员向前低下头。

Looking aside: 驾驶员向左或向右转头。

Talking & laughing: 在驾驶中，驾驶员说话或大笑。

Sleepy-eyes: 在驾驶中，驾驶员由于疲劳闭上眼睛。

Drowsy: 驾驶员看起来睡着和昏昏欲睡。（包括低头，缓慢眨眼和打哈欠。）

Stillness: 驾驶员正常驾驶。

--------------------------------------------------------------------------------------------------------

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.

训练集提供了18个人的文件夹。每个人包含了5个不同场景(noglasses, glasses, night_noglasses, night_glasses, sunglasses)，

每个场景包含4个视频，含有不同的状况和相关的标注文件。

"yawning.avi" :包含了打哈欠的行为.

"slowBlinkWithNodding.avi" : 包含了睡着的眼睛和低头行为。

"sleepyCombination.avi" : 包含有各种疲劳行为，如睡着的眼睛、打哈欠、低头。

"nonsleepyCombination.avi" : 包含了不疲劳的各种行为，如大笑、说话、四处看。

[注意] 在005中没有night_glasses场景。

--------------------------------------------------------------------------------------------------------

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.

验证集提供了4个人的文件夹。每个人包含5个在不同场景下的视频和相关的标注文件。视频展示了不同状况，其中有疲劳、非疲劳的状态转换。

--------------------------------------------------------------------------------------------------------

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.

每个视频中有4个标注，一个数字用来表示帧的状态。

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.