Vision-based Two-Stage Approach for Lane Departure Warning on Embedded Systems
- 발행기관 한동대학교 일반대학원
- 지도교수 이강
- 발행년도 2019
- 학위수여년월 2019. 2
- 학위명 석사
- 학과 및 전공 일반대학원 정보통신공학과
- 원문페이지 40
- 실제URI http://www.dcollection.net/handler/handong/200000211288
- UCI I804:47030-200000211288
- 본문언어 영어
- 저작권 한동대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
Lane detection is basis of ADAS (Advanced driver assistance systems) or autonomous vehicles because the result of lane detection is used for functions such as blind spot monitoring, road marker detection, or lane departure warning. This paper proposes two-step approach to detect lane markers. The first step is initial lane marker detection. It provides the initial lane location information for the tracking process as the starting point of the algorithm. To extract accurate lane marker, this step includes binarization, reliability and consistency estimation. This process achieves high detection accuracy while consuming high computational recourse in the embedded environment. The second step is lane marker detection by tracking. This step is based on the result of first step. Limiting the lane marker candidates search area around the initial lane marker reduces the amount of calculation. Tracking information decides the final lane marker with accordance of equations of Kalman Filter. Tracking-based lane detection maintains high accuracy with low computational complexity and adapts detection to changes in the environment. The lane marker detected by the proposed method is used for lane departure warning system. Lane departure warning prevents dangerous situations for drivers or self-driving cars. Typically, successive detection of lane marker makes the detection rate of the lane departure warning also increase. By comparing slopes and angles of current lane marker with reference lane marker, the departure rate is calculated. The high degree of the departure generates alarm signals. The experimental results show that the accuracy is normally about 97.82% on average and the processing time is 20 fps on Raspberry Pi 3. The precision of the LDW is 90.5% and the recall is 92.3% while the processing time is about 10ms. According to experimental results, the proposed algorithm has availability to be operated in the embedded system.
more목차
I. Introduction
1. Research background 9
2. Related works 9
3. Problem definition 11
4. Paper organization 12
II. Lane Detection Algorithm
1. Overview of the proposed algorithm 13
2. Initial lane marker detection 14
2.1 Preprocessing 15
Figure 3. Preprocessing images 16
2.2 Binarization 16
2.3 Analysis of reliability 20
2.4 Selection to lane marker 22
3. Lane marker detection by tracking 22
3.1 Kalman Filter 23
3.2 Preprocessing 24
3.3 Analysis of similarity 25
3.4 Selection to lane marker 26
III. Lane Departure Warning Algorithm
1. Lane departure criteria 27
2. Lane departure detection 28
IV. Experimental results
1. Accuracy 31
1.1 Lane Detection Accuracy 31
1.2 LDW accuracy 32
2. Processing time 32
V. Conclusion
Reference