• 대한임베디드공학회논문지
• /
• 제17권5호
• /
• pp.289-296
• /
• 2022

In this paper, we developed a deep learning-based recyclable object detection model. The model is developed based on YOLOv5 that is a one-stage detector. The deep learning model detects and classifies the recyclable object into 7 categories: paper, carton, can, glass, pet, plastic, and vinyl. We propose two methods for recyclable object detection models to solve problems during training. Bounding Box CutMix solved the no-objects training images problem of Mosaic, a data augmentation used in YOLOv5. Standardized Distance-based IoU replaced DIoU using a normalization factor that is not affected by the center point distance of the bounding boxes. The recyclable object detection model showed a final mAP performance of 0.91978 with Bounding Box CutMix and 0.91149 with Standardized Distance-based IoU.

• 지능정보연구
• /
• 제28권1호
• /
• pp.89-106
• /
• 2022

최근 영상 데이터의 급증으로 이를 효과적으로 처리하기 위해 객체 탐지 및 추적, 행동 인식, 표정 인식, 재식별(Re-ID)과 같은 다양한 컴퓨터비전 기술에 대한 수요도 급증했다. 그러나 객체 탐지 및 추적 기술은 객체의 영상 촬영 장소 이탈과 재등장, 오클루전(Occlusion) 등과 같이 성능을 저하시키는 많은 어려움을 안고 있다. 이에 따라 객체 탐지 및 추적 모델을 근간으로 하는 행동 및 표정 인식 모델 또한 객체별 데이터 추출에 난항을 겪는다. 또한 다양한 모델을 활용한 딥러닝 아키텍처는 병목과 최적화 부족으로 성능 저하를 겪는다. 본 연구에서는 YOLOv5기반 DeepSORT 객체추적 모델, SlowFast 기반 행동 인식 모델, Torchreid 기반 재식별 모델, 그리고 AWS Rekognition의 표정 인식 모델을 활용한 영상 분석 시스템에 단일 연결 계층적 군집화(Single-linkage Hierarchical Clustering)를 활용한 재식별(Re-ID) 기법과 GPU의 메모리 스루풋(Throughput)을 극대화하는 처리 기법을 적용한 행동 및 표정 검출용 영상 분석 시스템을 제안한다. 본 연구에서 제안한 시스템은 간단한 메트릭을 사용하는 재식별 모델의 성능보다 높은 정확도와 실시간에 가까운 처리 성능을 가지며, 객체의 영상 촬영 장소 이탈과 재등장, 오클루전 등에 의한 추적 실패를 방지하고 영상 내 객체별 행동 및 표정 인식 결과를 동일 객체에 지속적으로 연동하여 영상을 효율적으로 분석할 수 있다.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제31권11호
• /
• pp.1455-1473
• /
• 2004

본 연구는 객체 재사용과 재공학을 위해 레거시 시스템의 인터페이스 정보로부터 의미 있는 정보를 추출하고 새로운 시스템에 통합될 수 있도록 하기 위한 기존 레거시 시스템의 인터페이스에 기반 한 객체추출 기법을 제안한다. 본 논문에서 제안하는 객체추출 기법은 인터페이스 사용 사례 분석 단계, 인터페이스 객체 분할 단계, 객체구조 모델링 단계, 객체 모델 통합 단계 등 4단계로 구성되어 있다. 인터페이스 사용 사례 분석 단계는 인터페이스 구조 그리고 레거시 시스템과 사용자간의 상호작용 등의 정보를 획득하는 단계이다. 인터페이스 객체분할 단계는 인터페이스 정보를 의미 있는 필드들로 구분하는 단계이며, 객체구조 모델링 단계는 인터페이스 객체들 간의 구조적 관계와 협력 관계를 파악하여 모델링하는 단계이다. 마지막으로 객체 모델 통합 단계는 객체 단위의 단위 모델들을 통합하여 추상화된 정보를 포함한 상위 수준의 통합 모델을 유도하는 단계다. 객체추출 기법에 의해 생성된 객체 통합 모델은 역공학 기술자들의 레거시 시스템 이해와 레거시 시스템의 정보를 새로운 시스템에 적용하는데 있어 좀 더 용이한 효율성을 제공한다.

• 국제학술발표논문집
• /
• The 3th International Conference on Construction Engineering and Project Management
• /
• pp.1113-1116
• /
• 2009

The attempt to use a 3D model each field such as design, structure, construction, facilities, and estimation in the construction project has recently increased more and more while BIM (Building Information Modeling) that manages the process of generating and managing building data has risen during life cycle of a construction project. While the 2D Drawing based work of each field is achieved in the already existing construction project, the BIM based construction project aims at accomplishing 3D model based work of each field efficiently. Accordingly, the solution that fits 3D model based work of each field and supports plans in order to efficiently accomplish the relevant work is demanded. The estimation, one of the fields of the construction project, has applied BIM to calculate quantity and cost of the building materials used to construction works after taking off building quantity information from the 3D model by a item for a Quantity Take-off grouping the materials relevant to a 3D object. A 3D based estimation program has been commonly used in abroad advanced countries using BIM. The program can only calculate quantity related to one 3D object. In other words, it doesn't support the take-off process considering quantity of a contiguous object. In case of temporary materials used in the frame construction, there are instances where quantity is different by the contiguous object. For example, the formwork of the temporary materials quantity is changed by dimensions of the contiguous object because formwork of temporary materials goes through the quantity take-off process that deduces quantity of the connected object when different objects are connected. A worker can compulsorily adjust quantity so as to recognize the different object connected to the contiguous object and deduces quantity, but it mainly causes the confusion of work because it must complexly consider quantity of other materials related to the object besides. Therefore, this study is to propose the solution that automates the formwork 3D modeling to efficiently accomplish the quantity take-off of formwork by preventing the confusion of the work which is caused by the quantity deduction process between the contiguous object and the connected object.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 1998년도 추계종합학술대회 논문집
• /
• pp.927-930
• /
• 1998

This paper presents a new method for reconstructing 3 dimensional object model using a zoom camera. The proposed method uses zoom images to find the distance(D) between camera and object. Also the method uses images obtained around the object to find an $angle(\theta)$ between two connected planes of the object. With the D and $\theta,$ we can reconstruct the real sized 3-D model of object with less errors without stereo camera or rangefinder.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2003년도 ICCAS
• /
• pp.740-745
• /
• 2003

Motion tracking and object segmentation are the most fundamental and critical problems in vision tasks suck as motion analysis. An active contour model, snake, was developed as a useful segmenting and tracking tool for rigid or non-rigid objects. Snake is designed no the basis of snake energies. Segmenting and tracking can be executed successfully by energy minimization. In this research, two new paradigms for segmentation and tracking are suggested. First, because the conventional method uses only intensity information, it is difficult to separate an object from its complex background. Therefore, a new energy and design schemes should be proposed for the better segmentation of objects. Second, conventional snake can be applied in situations where the change between images is small. If a fast moving object exists in successive images, conventional snake will not operate well because the moving object may have large differences in its position or shape, between successive images. Snakes's nodes may also fall into the local minima in their motion to the new positions of the target object in the succeeding image. For robust tracking, the condensation algorithm was adopted to control the parameters of the proposed snake model called "adaptive color snake model(SCSM)". The effectiveness of the ACSM is verified by appropriate simulations and experiments.

• 한국컴퓨터산업학회논문지
• /
• 제5권9호
• /
• pp.941-946
• /
• 2004

기존의 객체지향 데이터베이스에 학습 객체에 기반을 둔 새로운 클래스를 도입한 확장된 객체 지향 데이터 베이스의 모델을 설계한다. 이를 구현하기 위해 기존의 객체 데이터 베이스에 학습객체클래스를 도입하였으며, 이 클래스들을 연산하기 위한 학습객체 연산 클래스를 설계하였다. 그리고 확장된 객체 지향의 데이터 모델상에서 데이터베이스에 저장된 학습객체의 경험적 분류 모델에 기반을 둔 검색이 가능한 질의어를 설계하였다.

• Journal of Information Processing Systems
• /
• 제14권6호
• /
• pp.1307-1317
• /
• 2018

To deal with the problems of occlusion, pose variations and illumination changes in the object tracking system, a regression model weighted multi-templates mean-shift (MS) algorithm is proposed in this paper. Target templates and occlusion templates are extracted to compose a multi-templates set. Then, the MS algorithm is applied to the multi-templates set for obtaining the candidate areas. Moreover, a regression model is trained to estimate the Bhattacharyya coefficients between the templates and candidate areas. Finally, the geometric center of the tracked areas is considered as the object's position. The proposed algorithm is evaluated on several classical videos. The experimental results show that the regression model weighted multi-templates MS algorithm can track an object accurately in terms of occlusion, illumination changes and pose variations.

• 제어로봇시스템학회논문지
• /
• 제21권11호
• /
• pp.989-995
• /
• 2015

Due to the recent increase of the importance and demand of security services, the importance of a surveillance monitor system that makes an automatic security system possible is increasing. As the market for surveillance monitor systems is growing, price competitiveness is becoming important. As a result of this trend, surveillance monitor systems based on an embedded system are widely used. In this paper, an object detection algorithm based on an embedded system for a surveillance monitor system is introduced. To apply the object detection algorithm to the embedded system, the most important issue is the efficient use of resources, such as memory and processors. Therefore, designing an appropriate algorithm considering the limit of resources is required. The proposed algorithm uses two background models; therefore, the embedded system is designed to have two independent processors. One processor checks the sub-background models for if there are any changes with high update frequency, and another processor makes the main background model, which is used for object detection. In this way, a background model will be made with images that have no objects to detect and improve the object detection performance. The object detection algorithm utilizes one-dimensional histogram distribution, which makes the detection faster. The proposed object detection algorithm works fast and accurately even in a low-priced embedded system.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2003년도 하계종합학술대회 논문집 Ⅲ
• /
• pp.1283-1286
• /
• 2003

Human can update the memory with new information not forgetting acquired information in the memory. ART(Adaptive Resonance Theory) does not need to change all information. The methodology of ART is followed. The ART updates the memory with the new information that is unknown if it is similar with the memorized information. On the other hand, if it is unknown information the ART adds it to the memory not updating the memory with the new one. This paper shows that ART is able to classify sensory information of a certain object. When ART receives new information of the object as an input, it searches for the nearest thing among the acquired information in the memory. If it is revealed that new information of the object has similarity with the acquired object, the model is updated to reflect new information to the memory. When new object does not have similarity with the acquired object, the model register the object into new memory