http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
전진환(Jinhwan Jeon),황윤진(Yoonjin Hwang),최세범(Seibum B. Choi) 한국자동차공학회 2023 한국자동차공학회 부문종합 학술대회 Vol.2023 No.5
Recent explosive interests in autonomous driving have led to the development of various background technologies. Among them, vehicle localization and pose estimation are needed directly or can be used for advanced applications, which is why these topics have been consistently studied from the past. Although a lot of research have been done in this field, most of the studies assumed ideal driving scenarios where all sensor data are available throughout the experiment. However, this is not the case for general driving, where sensor signals can be degraded or even be lost for extreme cases. Thus, this study presents a novel sensor fusion framework to enhance vehicle motion reconstruction, even for challenging scenarios. Other than conventional INS and GNSS, learning-based lane detection system is added to the sensor configuration to enhance estimation accuracy. Robustness and performance of proposed sensor fusion framework was verified through the comparison with State-of-the-Art visual odometry algorithms, by testing on real vehicle driving dataset.
원호 스플라인 근사를 이용한 주행 기동 이벤트의 감지 및 분류
임채호(Chaeho Lim),전진환(Jinhwan Jeon),황윤진(Yoonjin Hwang),최세범(Seibum B. Choi) 한국자동차공학회 2024 한국자동차공학회 부문종합 학술대회 Vol.2024 No.6
Classification of driving maneuver events is essential for generating lane-level maps, developing advanced driver assistance systems, and analyzing driver behavior. Typically, maneuver events classification involves three key stages: segmentation of driving data, extraction of features, and classification process. However, existing studies that used thresholdbased or sliding window approaches for segmentation have not adequately considered the various durations of maneuvers and face difficulties in distinguishing consecutively occurring events. This paper presents a novel algorithm that exhibits a high classification performance through segmentation using arc spline approximation and classification employing Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), enabling the precise identification of consecutively occurred events. The effectiveness of this algorithm was validated using sensor data from real vehicle test drives in Sejong City, South Korea.