This document presents a system for detecting driver fatigue based on eye tracking. It begins with a literature review on existing eye tracking and fatigue detection techniques. It then describes the proposed system with a block diagram showing the workflow, which includes face and eye detection, image preprocessing, edge detection, and eye state analysis by measuring eyelid distance and template matching. Experimental results are shown for ROI extraction accuracy and system performance on detecting open and closed eyes over multiple tests. Future work ideas are provided to improve the system.
16. MEASURE THE DISTANCE OF EYELID
x = (image_width/8)
y = (image_height/3)
height = (3* image_height)/5
width = (3* image_width)/5
17. MATCH THE EYE IMAGE WITH PREDEFINED TEMPLATE
Use Euclidian Distance
If m =< threshold then we assume it as a close eye and in fatigue state otherwise discard the image
Here m = matching amount between template and image
20. LINEAR REPRESENTATION OF THE SYSTEM WITH OPEN EYE
Image pre-processing & edge detection
Eye state analysis
Eye is open and
discards the image and
continues with the next
image
21. LINEAR REPRESENTATION OF THE SYSTEM WITH CLOSE EYE
Template matching & decision taking
If
the
5consecutive
close image then
the person is in
fatigue
Match
with
template
Not an open eye
image so pass it
for
template
matching
22. ROI EXTRACTION ONLY EYE DETECTION
Case
Length
Number Number of ROI
No.
of Video of frame face image
Extraction
with Accuracy
eye(Eye Detection)
TP/(total number
TP
FN
FP
TN
face image) %
Test 1 120 sec
295
295
188
102
5
0
64%
Test 2 90 sec
220
207
140
64
3
0
67%
Test 3 60 sec
145
145
88
50
7
0
61%
Test 4 30 sec
73
63
37
26
0
0
59%
Test 5 15 sec
36
36
21
13
2
0
58%
23. ROI EXTRACTION WITH FACE BEFORE EYE DETECTION
Case No.
Length
Number
Number
of Video
of frame
of
ROI Extraction (Face & Eye Accuracy
face Detection)
TP/(total
image
number
Face
TP
FN
FP
TN
image) %
Detection
Test 1
120 sec
295
295
242
227
13
2
0
77%
Test 2
90 sec
220
207
187
153
29
5
0
74%
Test 3
60 sec
145
145
135
110
22
3
0
76%
Test 4
30 sec
73
63
50
41
9
0
0
65%
Test 5
15 sec
36
36
34
25
7
2
0
69%
face
24. SYSTEM PERFORMANCE MEASUREMENT
Case
Number
No.
of
Open
face eye
Close
Eye
eye
(TP)
image
Detected Open eye
Close eye
Fatigue Detection
(%)
Open
Close
eye
TP
FN
TP
FN
eye
Test 1
295
165
130
145
82
122
23
69
13
84%
Test 2
207
109
98
100
53
87
13
41
12
77%
Test 3
145
60
85
43
67
38
5
60
7
89%
Test 4
63
45
18
33
12
27
6
9
3
75%
Test 5
36
14
22
10
15
6
4
12
3
80%
Average 81%
25. FUTURE WORKS
Upgrade the system with eyeglass & both eyes
Upgrade the system with a self light source
Tracking eyes with the head rotation
Update template with driver’s eyes dynamically