Earticle

최근 범죄예방과 안전문제 등으로 CCTV와 같은 영상장비가 다양하게 활용되고 있다. 영상기기들은 대부분 24시 간 작동되기 때문에 경비 인력을 절감할 수 있지만, 녹화된 영상에서 특정 인물과 같은 객체를 검색하는 업무는 여전히 수동으로 이루어지고 있어, 실시간 검색이 요구되는 상황에서는 정확하고 빠른 대처가 미흡하다. 본 논문에서는 최신 딥러닝 기술과 OpenCV 라이브러리를 이용하여 사용자의 의해 입력된 의상정보를 바탕으로 특정인물을 영상에서 빠르 게 검색하고, 그 결과를 실시간으로 전송하는 기술을 제안한다. 개발된 시스템은 YOLO 라이브러리를 이용하여 실시간 으로 인물객체를 탐지한 후, 딥러닝 기술을 이용하여 인간의 의상을 상/하의로 구분하고 OpenCV 라이브러리를 통해 색을 검출하여 특정 인물객체를 자동으로 인식하도록 구현하였다. 본 논문에서 개발한 시스템은 특정 의상을 갖춘 인물 객체를 정확하고 빠르게 인식할 뿐만 아니라 기타 객체 인식에도 활용할 수 있는 확장성을 갖추고 있어 다양한 용도의 영상감시시스템에 활용될 수 있을 것으로 기대된다.

Recently, video-taping equipment such as CCTV have been seeing more use for crime prevention and general safety concerns. Since these video-taping equipment operates all throughout the day, the need for security personnel is lessened, and naturally costs incurred from managing such manpower should also decrease. However, technology currently used predominantly lacks self-sufficiency when given the task of searching for a specific object in the recorded video such as a person, and has to be done manually; current security-based video equipment is insufficient in an environment where real-time information retrieval is required. In this paper, we propose a technology that uses the latest deep-learning technology and OpenCV library to quickly search for a specific person in a video; the search is based on the clothing information that is inputted by the user and transmits the result in real time. We implemented our system to automatically recognize specific human objects in real time by using the YOLO library, whilst deep learning technology is used to classify human clothes into top/bottom clothes. Colors are also detected through the OpenCV library which are then all combined to identify the requested object. The system presented in this paper not only accurately and quickly recognizes a person object with a specific clothing, but also has a potential extensibility that can be used for other types of object recognition in a video surveillance system for various purposes.

Human-object interaction (HOI) detection is a popular computer vision task that detects interactions between humans and objects. This task can be useful in many applications that require a deeper understanding of semantic scenes. Current HOI detection networks typically consist of a feature extractor followed by detection layers comprising small filters (eg, 1 × 1 or 3 × 3). Although small filters can capture local spatial features with a few parameters, they fail to capture larger context information relevant for recognizing interactions between humans and distant objects owing to their small receptive regions. Hence, we herein propose a three-stream HOI detection network that employs a context convolution module (CCM) in each stream branch. The CCM can capture larger contexts from input feature maps by adopting combinations of large separable convolution layers and residual-based convolution layers without increasing the number of parameters by using fewer large separable filters. We evaluate our HOI detection method using two benchmark datasets, V-COCO and HICO-DET, and demonstrate its state-of-the-art performance.

인간-객체 상호작용 탐지는 객체 탐지와 상호작용 인식을 함께 풀어야하는 분야로 탐지 모델의 학습을 위해서 많은 데이터를 필요로 한다. 현재 공개된 데이터셋은 규모가 부족하여 데이터 증강 기법에 대한 요구가 커지고 있으나, 대부분의 연구에서 기존의 객체 탐지, 이미지 분할분야에서 활용하는 증강 기법을 활용하고 있는 실정이다. 이에 본 연구에서는 인간-객체 상호작용 탐지 분야에서 활용하는 데이터셋의 특성을 파악하고, 이를 통해 인간-객체 상호작용 탐지 모델 성능 향상에 효과적인 데이터 증강 기법을 제안한다. 본 연구에서 제안한 증강 기법에 대한 검증을 위하여 실험 환경을 구축하고, 기존의 학습 모델에 적용하여 증강 기법을 적용할 경우에 탐지 모델의 성능 향상이 가능함을 확인하였다.

Human-object interaction(HOI) detection requires both object detection and interaction recognition, and requires a large amount of data to learn a detection model. Current opened dataset is insufficient in scale for training model enough. In this paper, we propose an easy and effective data augmentation method called Simple Quattro Augmentation(SQA) and Random Quattro Augmentation(RQA) for human-object interaction detection. We show that our proposed method can be easily integrated into State-of-the-Art HOI detection models with HICO-DET dataset.

In this paper, is propose to the object(human) dection method using geometrical structures and projection histograms in the image. The problem of existing methods for objects tracking of background subtracted is resulted from uncertainty at background unfixed. In this paper, two methods are applied to solve problem. This problems are proved by method. This problems is demonstrated by using this methods and applied to the train railroad crossing. Therefore, this paper aims that this contributes to improve the accident of the train railroad crossing.

Effective border security is crucial in managing and mitigating firearm-related threats. While prior research has focused on firearm detection, it lacks contextual analysis. This paper advances firearm-related incident assessment by integrating pose estimation to improve gun violence detection. Our novel approach extracts body and firearm pose graphs and employs Graph Attention Networks(GAT) for graph analysis to accurately identify gun violence incidents. By recognizing associated actions, our system provides greater situational awareness beyond mere firearm detection. Utilizing Graph-LSTM, we capture spatial and temporal information. As a result, our proposed algorithm is lighter and more accurate than the CNN-LSTM model used as a baseline, achieving test F1-scores of 82.04 % on our collected data.

컴퓨터 비전 연구에서 2차원 인체 자세는 매우 광범위한 연구 방향으로 특히 자세 추적과 행동 인식에서 유의미한 분야다. 인체 자세 표적 획득은 이미지에서 인체 목표를 정확히 찾는 방법을 연구하는 것이 핵심이며 인체 자세 인식은 인공지능(AI)에 적용하는 한편 일상생활에 활용되고 있어서 매우 중요한 연구의의가 있다. 인체 자세 인식 효과의 우수성의 기준은 인식 과정의 성공률과 정확도에 의해 결정된다. 본 연구의 인체 자세 인식에서는 딥러닝 전용 데이터셋인 MS COCO를 기반하여 인체를 17개의 키 포인트로 구분하였다. 다음으로 주요 특징에 대한 세분화 마스크(segmentation mask) 방법을 사용하여 인식률을 개선하였다. 최종적으로 신경망 모델을 설계하고 간단한 단계별 학습부터 효율적인 학습에 이르기까지 많은 수의 표본을 학습시키는 알고리즘을 제안하여 정확도를 향상할 수 있었다.

In computer vision research, the two-dimensional human pose is a very extensive research direction, especially in pose tracking and behavior recognition, which has very important research significance. The acquisition of human pose targets, which is essentially the study of how to accurately identify human targets from pictures, is of great research significance and has been a hot research topic of great interest in recent years. Human pose recognition is used in artificial intelligence on the one hand and in daily life on the other. The excellent effect of pose recognition is mainly determined by the success rate and the accuracy of the recognition process, so it reflects the importance of human pose recognition in terms of recognition rate. In this human body gesture recognition, the human body is divided into 17 key points for labeling. Not only that but also the key points are segmented to ensure the accuracy of the labeling information. In the recognition design, use the comprehensive data set MS COCO for deep learning to design a neural network model to train a large number of samples, from simple step-by-step to efficient training, so that a good accuracy rate can be obtained.

Human Action Behavior (HAB) systems critically depend on the reliability of human candidate generation at the initial stage, as detection errors directly propagate to subsequent behavior analysis. Conventional object detection approaches, such as YOLO, often suffer from false detections and missed detections (False Negative, FN) under real-world conditions including occlusion, illumination changes, and background clutter. To address these limitations, this study investigates a pose estimation-based approach that leverages structural constraints of the human body and compares it with object detection under identical conditions. A large-scale HAB dataset consisting of 389,040 images was constructed from multiple open-source sources. From the same dataset, both bounding box and keypoint annotations were generated to ensure a fair comparison. Models were trained under identical settings and evaluated using AP@50 and FN metrics. Experimental results show that all models achieved high accuracy (mAP@50 ≥ 0.994), confirming dataset reliability. However, pose estimation demonstrated superior performance in reducing FN. In particular, the YOLOv8-based pose model with CBAM reduced FN by approximately 6% compared to the baseline. Qualitative analysis further revealed that pose estimation maintains robustness in challenging scenarios such as occlusion, body deformation, and crowded environments. These results indicate that pose-based human candidate generation, combined with attention mechanisms, can significantly enhance the operational reliability of HAB systems.

최근 사람형(Anthropomorphic)로봇손의 사물조작 지능을 개발하기 위하여 행동복제(Behavior Cloning) Deep Reinforcement Learning(DRL) 연구가 진행중이다. 자유도(Degree of Freedom, DOF)가 높은 사람형 로봇손의 학습 문제점을 개선하기 위하여, 행동 복제를 통한 Human Demonstration Augmented(DA)강화 학습을 통하여 사람처럼 사물을 조작하는 지능을 학습시킬 수 있다. 그러나 사물 조작에 있어, 의미 있는 파지를 위해서는 사물의 특정 부위를 인식하고 파지하는 방법이 필수적이다. 본 연구에서는 딥러닝 YOLO기술을 적용하여 사물의 특정 부위를 인식하고, DA-DRL을 적용하여, 사물의 특정 부분을 파지하는 딥러닝 학습 기술을 제안하고, 2 종 사물(망치 및 칼)의 손잡이 부분을 인식하고 파지하여 검증한다. 본 연구에서 제안하는 학습방법은 사람과 상호작용하거나 도구를 용도에 맞게 사용해야하는 분야에서 유용할 것이다.