http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
샴 네트워크를 사용하여 추적 레이블을 사용하지 않는 다중 객체 검출 및 추적기 학습에 관한 연구
강정규,송유승,민경욱,최정단,Kang, Jungyu,Song, Yoo-Seung,Min, Kyoung-Wook,Choi, Jeong Dan 한국ITS학회 2022 한국ITS학회논문지 Vol.21 No.5
Multi-object tracking has been studied for a long time under computer vision and plays a critical role in applications such as autonomous driving and driving assistance. Multi-object tracking techniques generally consist of a detector that detects objects and a tracker that tracks the detected objects. Various publicly available datasets allow us to train a detector model without much effort. However, there are relatively few publicly available datasets for training a tracker model, and configuring own tracker datasets takes a long time compared to configuring detector datasets. Hence, the detector is often developed separately with a tracker module. However, the separated tracker should be adjusted whenever the former detector model is changed. This study proposes a system that can train a model that performs detection and tracking simultaneously using only the detector training datasets. In particular, a Siam network with augmentation is used to compose the detector and tracker. Experiments are conducted on public datasets to verify that the proposed algorithm can formulate a real-time multi-object tracker comparable to the state-of-the-art tracker models.
검출 데이터셋으로부터 자동 형성된 Instance Segmentation Label을 사용한 추적기 구성 방법
강정규(Jungyu Kang),이동진(Dongjin Lee),민경욱(Kyoung-Wook Min) 한국자동차공학회 2023 한국자동차공학회 학술대회 및 전시회 Vol.2023 No.11
Object tracking in computer vision is essential for various applications but faces challenges in distinguishing similar objects and generating re-identification (Re-ID) based trackers due to limited datasets. We propose a solution using the Segment Anything Model (SAM) to automatically create instance segmentation labels from detection data. We employ Test-time Augmentation for label accuracy and utilize YOLOX for detection, instance segmentation, and Re-ID feature generation. Our model achieves a 56.50 mean average precision (mAP) on the ETRI State Recognition Dataset and runs in real-time on an NVIDIA TITAN RTX GPU. Qualitative assessments demonstrate the models effectiveness in urban environments, showcasing its potential for practical object tracking.
도시내 다양한 야간 조명 환경에 강인한 자율주행 알고리즘 개발을 위한 데이터 증강 기법
강정규(Jungyu Kang),안택현(Taeghyun Ahn),민경욱(Kyoung-Wook Min) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
Recently, datasets for training a CNN model have become very important in the field of self-driving vehicle research. However, most of the currently released deep learning datasets are focused on daytime environments. In this paper, we present a study on how to enhance the robustness of the CNN model against night environment by augmenting the autonomous driving dataset using CycleGAN. The experimental results showed that the proposed framework shows a significant improvement in robustness without adding any additional annotated datasets.
김휘(Whui Kim),윤창락(Changrak Yoon),강정규(Jungyu Kang),김경호(Kyong-Ho Kim) 대한전자공학회 2015 대한전자공학회 학술대회 Vol.2015 No.11
We propose pedestrian detection rate improvement method on Tegra K1 Processor considering vehicle application. We improved the OpenCV HoG-SVM detection algorithm to reduce computation time. Using regions of interest defined by the calibration information and multi-scale classifiers that share HoG Features, computation time of HoG-SVM detector was reduced by 34% on Geforce GTX 980 processor and by 44% on Tegra K1 processor.