Advanced driver assistant system using image processing

Name: Sorn Sooksatra

keyword: ADAS

LCSH: Automobile driving

; Drowsiness detection

LCSH: Traffic safety

; Circulation object detection

LCSH: Traffic light detection

Abstract: Advanced Driver Assistant System is the system that supports the driver while driving the car on the road to avoid car accident. By using image processing, it is used for monitoring the road environment inside and outside the car and warning the status to the driver when he/she lose attention to the road. This thesis research aim to two systems in ADAS, traffic light detection (TLD) and drowsiness detection system (DDS) used for detecting and recognizing traffic lights (TLs) and drowsiness of the driver respectively. However, we focus on the detection of red and yellow traffic lights for TLD because those signs are more crucial than green TLs for avoiding traffic accidents. Since they are installed in the car, lighting conditions need to be concerned for supporting various luminance of TLs and the driver face. For TLD, color information is used (e.g., HSV, RGB, and CIELab color model). For CIELab, their method can solve blooming effect problem, irregular intensity distribution on TLs, but has less effective to low luminance of TLs. We improve their method by excluding L component that is sensitive to luminance. The proposed method uses combination only a* and b* component for TLs extraction. The result shows that it is insensitive to various luminance of TLs and blooming effect still be solved. Then the modified fast radial symmetry transform (MFRST) is used to localize the TLs. For DDS, eye state recognition techniques mostly use pattern matching or recognition based on intensity information from the image that effects lighting conditions. We adapted eye center localization technique to be used as eye state recognition. Our method uses dark circular object detection using cross-correlation between unit distance and gradient vector. For unit gradient vector, it is known that it depends on image pattern but it is insensitive to luminance. For pre-processing step, Haar cascade method is used for detecting the face and a pair of eye or eye region. In our experiment, we use 100 street views in day and night time that has 200 TLs in total as input image for TLD. For DDS, we use 4 sample videos with 30 fps in bright, dim, and dark situation that has 600 frames for each video. IR light and IR camera are used in dark situation. The recognition performance is evaluated by F1 score. In overall case, it gives better performance compared with conventional methods in both TLD and DDS

Thammasat University. Thammasat University Library

Address: BANGKOK

Email: preserv@tu.ac.th

Name: Kondo, Toshiaki

Role: advisor

Created: 2014

Modified: 2021-12-28

Issued: 2021-12-28

วิทยานิพนธ์/Thesis

application/pdf

DOI: 10.14457/TU.the.2014.514

eng

DegreeName: Master of Engineering

Level: Master's Degree

Descipline: Information and Communication Technology for Embedded Systems

Grantor: Thammasat University

©copyrights Thammasat University

ลำดับที่.	ชื่อแฟ้มข้อมูล	ขนาดแฟ้มข้อมูล	จำนวนเข้าถึง	วัน-เวลาเข้าถึงล่าสุด
1	5620sorn.pdf	3.16 MB	1	2025-10-23 22:23:11

ใช้เวลา

0.027168 วินาที