• Journal of Intelligence and Information Systems
• /
• v.29 no.4
• /
• pp.15-30
• /
• 2023

Recently, various services using artificial intelligence(AI) are emerging in the media field as well However, most of the video editing, which involves finding an editing point and attaching the video, is carried out in a passive manner, requiring a lot of time and human resources. Therefore, this study proposes a methodology that can detect the edit points of video according to whether person in video are spoken by using Video Swin Transformer. First, facial keypoints are detected through face alignment. To this end, the proposed structure first detects facial keypoints through face alignment. Through this process, the temporal and spatial changes of the face are reflected from the input video data. And, through the Video Swin Transformer-based model proposed in this study, the behavior of the person in the video is classified. Specifically, after combining the feature map generated through Video Swin Transformer from video data and the facial keypoints detected through Face Alignment, utterance is classified through convolution layers. In conclusion, the performance of the image editing point detection model using facial keypoints proposed in this paper improved from 87.46% to 89.17% compared to the model without facial keypoints.

• Journal of Intelligence and Information Systems
• /
• v.29 no.1
• /
• pp.143-174
• /
• 2023

With the development of deep learning technologies, Artificial Intelligence powered Optical Character Recognition (AI-OCR) has evolved to read multiple languages from various forms of images accurately. For the financial industry, where a large number of diverse documents are processed through manpower, the potential for using AI-OCR is great. In this study, we present a configuration and a design of an AI-OCR modality for use in the financial industry and discuss the platform construction with application cases. Since the use of financial domain data is prohibited under the Personal Information Protection Act, we developed a deep learning-based data generation approach and used it to train the AI-OCR models. The AI-OCR models are trained for image preprocessing, text recognition, and language processing and are configured as a microservice architected platform to process a broad variety of documents. We have demonstrated the AI-OCR platform by applying it to financial domain tasks of document sorting, document verification, and typing assistance The demonstrations confirm the increasing work efficiency and conveniences.

• Journal of Intelligence and Information Systems
• /
• v.29 no.2
• /
• pp.57-84
• /
• 2023

With the development of information and communications technology (ICT) and big data technology, anyone can easily obtain and utilize vast amounts of data through the Internet. Therefore, the capability of selecting high-quality data from a large amount of information is becoming more important than the capability of just collecting them. This trend continues in academia; literature reviews, such as systematic and non-systematic reviews, have been conducted in various research fields to construct a healthy knowledge structure by selecting high-quality research from accumulated research materials. Meanwhile, after the COVID-19 pandemic, remote healthcare services, which have not been agreed upon, are allowed to a limited extent, and new healthcare services such as health recommender systems (HRS) equipped with artificial intelligence (AI) and big data technologies are in the spotlight. Although, in practice, HRS are considered one of the most important technologies to lead the future healthcare industry, literature review on HRS is relatively rare compared to other fields. In addition, although HRS are fields of convergence with a strong interdisciplinary nature, prior literature review studies have mainly applied either systematic or non-systematic review methods; hence, there are limitations in analyzing interactions or dynamic relationships with other research fields. Therefore, in this study, the overall network structure of HRS and surrounding research fields were identified using citation network analysis (CNA). Additionally, in this process, in order to address the problem that the latest papers are underestimated in their citation relationships, the GraphSAGE algorithm was applied. As a result, this study identified 'recommender system', 'wireless & IoT', 'computer vision', and 'text mining' as increasingly important research fields related to HRS research, and confirmed that 'personalization' and 'privacy' are emerging issues in HRS research. The study findings would provide both academic and practical insights into identifying the structure of the HRS research community, examining related research trends, and designing future HRS research directions.

• Journal of Broadcast Engineering
• /
• v.23 no.3
• /
• pp.351-360
• /
• 2018

Human emotion recognition is a research topic that is receiving continuous attention in computer vision and artificial intelligence domains. This paper proposes a method for classifying human emotions through multiple neural networks based on multi-modal signals which consist of image, landmark, and audio in a wild environment. The proposed method has the following features. First, the learning performance of the image-based network is greatly improved by employing both multi-task learning and semi-supervised learning using the spatio-temporal characteristic of videos. Second, a model for converting 1-dimensional (1D) landmark information of face into two-dimensional (2D) images, is newly proposed, and a CNN-LSTM network based on the model is proposed for better emotion recognition. Third, based on an observation that audio signals are often very effective for specific emotions, we propose an audio deep learning mechanism robust to the specific emotions. Finally, so-called emotion adaptive fusion is applied to enable synergy of multiple networks. The proposed network improves emotion classification performance by appropriately integrating existing supervised learning and semi-supervised learning networks. In the fifth attempt on the given test set in the EmotiW2017 challenge, the proposed method achieved a classification accuracy of 57.12%.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.308-317
• /
• 2022

Researches are underway to efficiently reduce the size of video data transmitted and stored in the image analysis process using deep learning-based machine vision technology. MPEG (Moving Picture Expert Group) has newly established a standardization project called VCM (Video Coding for Machine) and is conducting research on video encoding for machines rather than video encoding for humans. We are researching a multitask that performs various tasks with one image input. The proposed pipeline does not perform all object detection of each task that should precede object detection, but precedes it only once and uses the result as an input for each task. In this paper, we propose a pipeline for efficient multitasking and perform comparative experiments on compression efficiency, execution time, and result accuracy of the input image to check the efficiency. As a result of the experiment, the capacity of the input image decreased by more than 97.5%, while the accuracy of the result decreased slightly, confirming the possibility of efficient multitasking.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.10
• /
• pp.409-418
• /
• 2022

In this paper, we propose SAAnnot-C3Pap, a semi-automatic annotation method for obtaining ground truth of a player's posture. In order to obtain ground truth about the two-dimensional joint position in the existing music domain, openpose, a two-dimensional posture estimation method, was used or manually labeled. However, automatic annotation methods such as the existing openpose have the disadvantages of showing inaccurate results even though they are fast. Therefore, this paper proposes SAAnnot-C3Pap, a semi-automated annotation method that is a compromise between the two. The proposed approach consists of three main steps: extracting postures using openpose, correcting the parts with errors among the extracted parts using supervisely, and then analyzing the results of openpose and supervisely. Perform the synchronization process. Through the proposed method, it was possible to correct the incorrect 2D joint position detection result that occurred in the openpose, solve the problem of detecting two or more people, and obtain the ground truth in the playing posture. In the experiment, we compare and analyze the results of the semi-automated annotation method openpose and the SAAnnot-C3Pap proposed in this paper. As a result of comparison, the proposed method showed improvement of posture information incorrectly collected through openpose.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.50 no.6
• /
• pp.42-57
• /
• 2022

This study aims to create a basic model for classifying the activity photos that urban park users shared on social media using Deep Learning through Artificial Intelligence. Regarding the social media data, photos related to urban parks were collected through a Naver search, were collected, and used for the classification model. Based on the indicators of Naturalness, Potential Attraction, and Activity, which can be used to evaluate the characteristics of urban parks, 21 classification categories were created. Urban park photos shared on Naver were collected by category, and annotated datasets were created. A custom CNN model and a transfer learning model utilizing a CNN pre-trained on the collected photo datasets were designed and subsequently analyzed. As a result of the study, the Xception transfer learning model, which demonstrated the best performance, was selected as the urban park user activity image classification model and evaluated through several evaluation indicators. This study is meaningful in that it has built AI as an index that can evaluate the characteristics of urban parks by using user-shared photos on social media. The classification model using Deep Learning mitigates the limitations of manual classification, and it can efficiently classify large amounts of urban park photos. So, it can be said to be a useful method that can be used for the monitoring and management of city parks in the future.

• Korean Journal of Radiology
• /
• v.24 no.6
• /
• pp.541-552
• /
• 2023

Objective: Detection of pneumoperitoneum using abdominal radiography, particularly in the supine position, is often challenging. This study aimed to develop and externally validate a deep learning model for the detection of pneumoperitoneum using supine and erect abdominal radiography. Materials and Methods: A model that can utilize "pneumoperitoneum" and "non-pneumoperitoneum" classes was developed through knowledge distillation. To train the proposed model with limited training data and weak labels, it was trained using a recently proposed semi-supervised learning method called distillation for self-supervised and self-train learning (DISTL), which leverages the Vision Transformer. The proposed model was first pre-trained with chest radiographs to utilize common knowledge between modalities, fine-tuned, and self-trained on labeled and unlabeled abdominal radiographs. The proposed model was trained using data from supine and erect abdominal radiographs. In total, 191212 chest radiographs (CheXpert data) were used for pre-training, and 5518 labeled and 16671 unlabeled abdominal radiographs were used for fine-tuning and self-supervised learning, respectively. The proposed model was internally validated on 389 abdominal radiographs and externally validated on 475 and 798 abdominal radiographs from the two institutions. We evaluated the performance in diagnosing pneumoperitoneum using the area under the receiver operating characteristic curve (AUC) and compared it with that of radiologists. Results: In the internal validation, the proposed model had an AUC, sensitivity, and specificity of 0.881, 85.4%, and 73.3% and 0.968, 91.1, and 95.0 for supine and erect positions, respectively. In the external validation at the two institutions, the AUCs were 0.835 and 0.852 for the supine position and 0.909 and 0.944 for the erect position. In the reader study, the readers' performances improved with the assistance of the proposed model. Conclusion: The proposed model trained with the DISTL method can accurately detect pneumoperitoneum on abdominal radiography in both the supine and erect positions.

• Journal of Intelligence and Information Systems
• /
• v.28 no.1
• /
• pp.89-106
• /
• 2022

Recently, the amount of video data collected from smartphones, CCTVs, black boxes, and high-definition cameras has increased rapidly. According to the increasing video data, the requirements for analysis and utilization are increasing. Due to the lack of skilled manpower to analyze videos in many industries, machine learning and artificial intelligence are actively used to assist manpower. In this situation, the demand for various computer vision technologies such as object detection and tracking, action detection, emotion detection, and Re-ID also increased rapidly. However, the object detection and tracking technology has many difficulties that degrade performance, such as re-appearance after the object's departure from the video recording location, and occlusion. Accordingly, action and emotion detection models based on object detection and tracking models also have difficulties in extracting data for each object. In addition, deep learning architectures consist of various models suffer from performance degradation due to bottlenects and lack of optimization. In this study, we propose an video analysis system consists of YOLOv5 based DeepSORT object tracking model, SlowFast based action recognition model, Torchreid based Re-ID model, and AWS Rekognition which is emotion recognition service. Proposed model uses single-linkage hierarchical clustering based Re-ID and some processing method which maximize hardware throughput. It has higher accuracy than the performance of the re-identification model using simple metrics, near real-time processing performance, and prevents tracking failure due to object departure and re-emergence, occlusion, etc. By continuously linking the action and facial emotion detection results of each object to the same object, it is possible to efficiently analyze videos. The re-identification model extracts a feature vector from the bounding box of object image detected by the object tracking model for each frame, and applies the single-linkage hierarchical clustering from the past frame using the extracted feature vectors to identify the same object that failed to track. Through the above process, it is possible to re-track the same object that has failed to tracking in the case of re-appearance or occlusion after leaving the video location. As a result, action and facial emotion detection results of the newly recognized object due to the tracking fails can be linked to those of the object that appeared in the past. On the other hand, as a way to improve processing performance, we introduce Bounding Box Queue by Object and Feature Queue method that can reduce RAM memory requirements while maximizing GPU memory throughput. Also we introduce the IoF(Intersection over Face) algorithm that allows facial emotion recognized through AWS Rekognition to be linked with object tracking information. The academic significance of this study is that the two-stage re-identification model can have real-time performance even in a high-cost environment that performs action and facial emotion detection according to processing techniques without reducing the accuracy by using simple metrics to achieve real-time performance. The practical implication of this study is that in various industrial fields that require action and facial emotion detection but have many difficulties due to the fails in object tracking can analyze videos effectively through proposed model. Proposed model which has high accuracy of retrace and processing performance can be used in various fields such as intelligent monitoring, observation services and behavioral or psychological analysis services where the integration of tracking information and extracted metadata creates greate industrial and business value. In the future, in order to measure the object tracking performance more precisely, there is a need to conduct an experiment using the MOT Challenge dataset, which is data used by many international conferences. We will investigate the problem that the IoF algorithm cannot solve to develop an additional complementary algorithm. In addition, we plan to conduct additional research to apply this model to various fields' dataset related to intelligent video analysis.

• Journal of Intelligence and Information Systems
• /
• v.29 no.2
• /
• pp.171-187
• /
• 2023

When a person's face is recognized through a recording device such as a low-pixel surveillance camera, it is difficult to capture the face due to low image quality. In situations where it is difficult to recognize a person's face, problems such as not being able to identify a criminal suspect or a missing person may occur. Existing studies on face recognition used refined datasets, so the performance could not be measured in various environments. Therefore, to solve the problem of poor face recognition performance in low-quality images, this paper proposes a method to generate high-quality images by performing image quality improvement on low-quality facial images considering various environments, and then improve the performance of facial feature point detection. To confirm the practical applicability of the proposed architecture, an experiment was conducted by selecting a data set in which people appear relatively small in the entire image. In addition, by choosing a facial image dataset considering the mask-wearing situation, the possibility of expanding to real problems was explored. As a result of measuring the performance of the feature point detection model by improving the image quality of the face image, it was confirmed that the face detection after improvement was enhanced by an average of 3.47 times in the case of images without a mask and 9.92 times in the case of wearing a mask. It was confirmed that the RMSE for facial feature points decreased by an average of 8.49 times when wearing a mask and by an average of 2.02 times when not wearing a mask. Therefore, it was possible to verify the applicability of the proposed method by increasing the recognition rate for facial images captured in low quality through image quality improvement.