송봉섭 김형태 2015-02 19558 https://aurora.ajou.ac.kr/handle/2018.oak/18625 학위논문(석사)--아주대학교 일반대학원 :기계공학과,2015. 2 1. 서 론 1.1 연구 배경 1.2 연구 내용 2. 시스템 구성 2.1 센서 구성(sensor configuration) 2.2 소프트웨어 아키텍쳐 3. 전방 차량 인식 기술 3.1 가드레일 형태 속성 판단을 통한 차량 인지 기술 3.2 차선 신뢰도 판단을 통한 고장 시 안전동작 4. 후방 차량 인식 기술 5. 측방 차량 예측 기술 6. 실험적 검증 7. 결론 8. 참고 문헌 Abstract kor The Graduate School, Ajou University 아주대학교 논문은 저작권에 의해 보호받습니다. 운전지원시스템을 위한 레이더와 비전센서 융합기반 전 방위 차량 인식 기술 개발 All-around Vehicle Recognition based on Radar and Vision Sensor Fusion for Driver Assistance System Thesis 아주대학교 일반대학원 일반대학원 기계공학과 2015. 2 Master http://dcoll.ajou.ac.kr:9080/dcollection/jsp/common/DcLoOrgPer.jsp?sItemId=000000019558 운전지원시스템 비전센서 This paper proposes all-around multiple vehicle recognition algorithm based on radar and vision sensor fusion for driver assistance system. In order to realize driver assistance system which considers collision safety toward surrounding vehicles, it is necessary to recognize the vehicles in accordance with longitudinal and lateral position of the vehicles. For this purpose, there are three challenging problems with sensor configuration this thesis presented. First of all, front radar detects objects about guardrail, and it could be recognized left and/of right vehicle. In order to solve this problem, guardrail recognition algorithm which is based on shape and motion attribute is presented. Secondly, due to lack of vision sensor to detect backward, it is impossible to measure radius of curvature in backward. It causes wrong vehicle recognition in curved road especially. To solve the problem, it is suggested that radius of curvature toward rear direction is estimated based on the radius measured by front vision and time delay. The last problem is misdetection in side of ego vehicle, and it is caused by limit of field of view of environmental sensors. In order to improve, tracking and estimation algorithm is proposed. It is realized by discrete Kalman filter based on Constant Velocity (CV) model and Constant Turn Rate and Velocity (CTRV) model. Finally, the suggested algorithm for all-around multiple vehicle recognition is validated via field test data on real road.