지능형 자동차 비전모듈의 성능 검증을 위한 가상 차선비디오 생성 방법

Abstract

최근 다양한 카메라 기술들이 개발되고 자동차에 대한 소비자의 인식 및 요구가 달라짐에 따라 이에 대응하기 위한 미래형 지능형 자동차 기술 개발이 활발히 이루어지고 있다. 미래형 지능형 자동차는 차량에 장착된 다양한 센서를 이용하여 물체를 검출 및 인식하고 이를 기반으로 운전자에게 필요한 정보를 제공한다. 그 중에서 카메라는 최근 가장 많이 사용되고 있는 센서 중에 하나이다. 최근 차량 전방에 설치된 카메라를 이용한 다양한 형태의 비전기반 지능형 자동차 안전기술들이 개발되고 있다. 비전기반 지능형 자동차 안전기술에는 차선을 인식하여 비정상적 차선이탈을 경고해주는 차선이탈경고시스템(Lane Departure Warning System: LDWS)이나 전방 차량을 인식하여 주행 중 추돌을 방지해주는 전방 충돌 경고 시스템(Forward Collision Warning System: FCWS) 등이 있다. 이러한 비전기반 지능형 자동차 안전기술들이 실제 차량에 창작되어 활용되기 위해서는 개발 단계에서 다양한 실험 및 검증을 통해 높은 수준의 신뢰성을 확보해야 한다. 그러나 이런 실험 및 검증들은 시간적으로나 공간상의 제약으로 인해 많은 어려움이 따른다. 따라서 이러한 실험 및 검증을 보다 효율적으로 수행하기 위한 노력들이 진행되고 있으며, 그 중에 대표적인 것이 시뮬레이터(Simulator) 개발이다. 본 논문에서는 비전기반 지능형 자동차 시스템의 성능검증을 위한 성능검증용 가상의 비디오 생성방법을 제안하였다. 그 중에서도 특히 차선이탈경고시스템에 초점을 맞추어 다양한 이미지합성기법들을 활용하여 실제 도로상의 차선에 대해 오염된 차선 환경을 가상적으로 구현하였다. 다양한 크기와 모양의 오염 이미지를 생성하기 위해 가우시안 오염 패턴(Gaussian Noise Pattern)을 제안하였다. 2차원 이미지의 합성 기법을 비디오로 확장함으로써 실제 도로 상에서 주행하는 것과 같은 환경을 구축하였다. 본 논문은 지능형 자동차 비전 시스템의 성능검증을 위한 가상의 오염 차선 비디오 생성방법을 제안했다는데 학문적 의의가 있다. 지능형 자동차 비전 시스템의 신뢰성 확보를 위해 본 논문에서 제안하는 방법을 통해 생성한 비디오를 기반으로 다양한 실험을 함으로써 보다 효과적인 성능검증이 가능하다. 또한 본 논문은 지능형 자동차 비전 시스템의 성능검증을 위한 시뮬레이터 개발의 기반을 제공했다는 점에서도 학문적 중요성을 가진다 하겠다. 다양한 분야의 성능검증용 시뮬레이터 개발에 본 논문을 활용함으로써 시뮬레이터 개발에 들어가는 시간과 노력들을 조금이나마 줄일 수 있을 것이라 기대된다. 향후 본 논문을 기반으로 다양한 조명 및 기상조건을 추가로 생성하는 연구를 진행한다면 보다 우수한 검증 시스템 개발이 가능할 것으로 사료된다.|Recently, there is an active development in the field of future intelligent vehicle because of the advancement in the technology of camera and the high public awareness for vehicle safety. The intelligent vehicle detects and recognizes objects using various sensors mounted on the vehicle and provides the necessary information to driver based on that detection and recognition. In particular, the camera is one of the most widely used sensors. Lately, various safety technologies of vision-based intelligent vehicle have been developed using a camera installed in front of the vehicle. Some representative examples of vision-based intelligent safety technology are LDWS (Lane Departure Warning System), FCWS (Forward Collision Warning System) and so on. These safety technologies should be obtained high level of reliability through various experiments and validations in the development stage in order to be used commercially. However, these experiments and validations are not easy to be implemented due to constraints such as time and space. Therefore, many efforts are being made to run these experiments and validations efficiently. One of the most representative methods is by developing a simulator to evaluate the performance of the system. In this dissertation, an augmented video generation method for performance evaluation of intelligent vehicle vision systems was proposed. In particular, we focused on the performance verification of LDWS. To evaluate the performance of LDWS, a virtual lane video which included contamination lane was generated by utilizing a variety of image synthesis techniques based on the real road scene. In order to produce contamination in a variety of shape and sizes, a contamination generation method based on the Gaussian function was also proposed. A simulation situation of driving on the road is generated by applying image synthesis techniques to the video. This dissertation has the following academic significances. First, this dissertation proposed an augmented video generation method for the performance evaluation of intelligent vehicle vision systems. Through various experiments based on videos generated from the proposed system more effective performance verification can be. This dissertation also has a significant role in developing simulator for the performance verification of the intelligent vehicle vision systems. By taking advantage of this dissertation in the development of various simulators for performance verification, time and effort to develop the simulator can be reduced. In the future, by taking into consideration about the variety of lighting and weather conditions on the road based on this dissertation, it is promising to develop a more reliable and effective performance verification system.; Recently, there is an active development in the field of future intelligent vehicle because of the advancement in the technology of camera and the high public awareness for vehicle safety. The intelligent vehicle detects and recognizes objects using various sensors mounted on the vehicle and provides the necessary information to driver based on that detection and recognition. In particular, the camera is one of the most widely used sensors. Lately, various safety technologies of vision-based intelligent vehicle have been developed using a camera installed in front of the vehicle. Some representative examples of vision-based intelligent safety technology are LDWS (Lane Departure Warning System), FCWS (Forward Collision Warning System) and so on. These safety technologies should be obtained high level of reliability through various experiments and validations in the development stage in order to be used commercially. However, these experiments and validations are not easy to be implemented due to constraints such as time and space. Therefore, many efforts are being made to run these experiments and validations efficiently. One of the most representative methods is by developing a simulator to evaluate the performance of the system. In this dissertation, an augmented video generation method for performance evaluation of intelligent vehicle vision systems was proposed. In particular, we focused on the performance verification of LDWS. To evaluate the performance of LDWS, a virtual lane video which included contamination lane was generated by utilizing a variety of image synthesis techniques based on the real road scene. In order to produce contamination in a variety of shape and sizes, a contamination generation method based on the Gaussian function was also proposed. A simulation situation of driving on the road is generated by applying image synthesis techniques to the video. This dissertation has the following academic significances. First, this dissertation proposed an augmented video generation method for the performance evaluation of intelligent vehicle vision systems. Through various experiments based on videos generated from the proposed system more effective performance verification can be. This dissertation also has a significant role in developing simulator for the performance verification of the intelligent vehicle vision systems. By taking advantage of this dissertation in the development of various simulators for performance verification, time and effort to develop the simulator can be reduced. In the future, by taking into consideration about the variety of lighting and weather conditions on the road based on this dissertation, it is promising to develop a more reliable and effective performance verification system.