http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
의약 용기용 다중 카메라 인라인 검사 시스템에서 정확도 향상을 위한 딥러닝 네트워크 및 레이어에 대한 연구
이태윤(Tae-Yoon Lee),라승탁(Seung-Tak Ra),오승진(Seung-Jin Oh),오준혁(Jun-Hyeok Oh),신인영(In-Young Shin),이승호(Seung-Ho Lee) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
In this paper, we propose a study on the effect of deep learning networks and layers to improve the accuracy of inline inspection system for medicinal appliances. The base network was tested while crossing CNN, Resnet50 and Vision Transformer. As a result of the experiment, the cost and accuracy of the learning result of Resnet50 was lower than the learning result of CNN and Vision Transformer. Therefore, it seems appropriate to use Resnet50 to improve the accuracy of multi-camera inline inspection.
12M의 고해상도 360° 카메라를 사용한 주차장의 14면 주차 상태 판단 프로그램
이영지(Young-Ji Lee),이희열(Hee-Yeol Lee),고태영(Tae-Young Ko),곽동훈(Dong-Hoon Kwak),김재형(Jae-Hyung Kim),김주호(Joo-Ho Kim),오승진,이태윤(Tae-Yoon Lee),박상민(Sang-Min Park),이승호(Seung-Ho Lee) 대한전자공학회 2019 대한전자공학회 학술대회 Vol.2019 No.6
In this paper, we propose a program to identify 14 parking status in a parking lot using a high resolution 360° camera of 12M. The proposed program consists of three steps: Match with plane image, parking area detection algorithm, and discrimination of parking using learning method. Tests on a model car to evaluate the program to identify 14 parking status in a parking lot using a high-resolution 360° camera of 12M showed 100% accuracy for both parking and double parking. Therefore, the effectiveness of a program to identify 14 parking status in a parking lot using a high-resolution 360° camera of 12M proposed in this paper has been proved.
Depth 추정 기법을 이용한 쓰러짐 및 낙상 감지 방법
오승진(Seung-Jin Oh),라승탁(Seung-Tak Ra),이태윤(Tae-Yoon Lee),오준혁(Jun-Hyeok Oh),신인영(In-Young Shin),이승호(Seung-Ho Lee) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
In this paper, we propose a falling down detection method using Depth estimation techniques[1]. The proposed method consists of four processes: Depth estimation using Depthformer[2], three points on the floor plane based on the generated Depth Map, equation derivation of the plane, three-dimensional relative distance coordinate calculation, and falling down detection using calculated relative distance coordinates. As a result of the experiment, 9 out of 10 falling down distribution videos distributed by the Korea Internet & Security Agency (KISA) were detected, proving effectiveness.
라승탁(Seung-Tak Ra),오승진(Seung-Jin Oh),이태윤(Tae-Yoon Lee),오준혁(Jun-Hyeok Oh),신인영(In-Young Shin),이승호(Seung-Ho Lee) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
In this paper, loitering and intrusion algorithms for intelligent CCTV were developed. First, object detection was performed based on Yolo_X, a deep learning model capable of real-time object detection. Second, the false detection removal algorithm was applied to remove falsely detected objects. Finally, the event situation is determined by applying loitering and intrusion algorithms to each detected object. As a result of the experiment, loitering and intrusion events were detected at the correct time in 59 out of 60 images.
NeRF 기법에서 사용되는 UV Position Map 생성을 위한 Auto Encoder와 Variational Auto Encoder 비교에 관한 연구
김홍직(Hong-Jik Kim),이희열(Hee-Yeol Lee),라승탁(Seung-Tak Ra),김정윤(Jeong-Yoon Kim),오승진(Seung-Jin Oh),김기범(Gi-Beom Kim),유하영(Ha-Young Yoo),이태윤(Tae-Yoon Lee),오준혁(Jun-Hyeok Oh),이승호(Seung-Ho Lee) 대한전자공학회 2022 대한전자공학회 학술대회 Vol.2022 No.11
In this paper, Auto Encoder and Variational Auto Encoder were compared in generating UV Position Map, which is one of the important factors for 3D face reconstruction. Both models were trained from the same MNIST data, and as a result of training, the performance of Variational Auto Encoder was better. This seems to be the effect of the reparameterization trick that Auto Encoder does not have. Since the encoder extracts the mean and variance of the input data and uses them, the decoder knows the distribution information of the input data, so more sophisticated images can be created. Through this, by using the flow field of the continuous UV position map generated by VAE, it can be added as a new input to NeRF, and a novel view with more natural and various angles can be created than that of the existing NeRF.