• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.751-770
• /
• 2019

Action recognition is an essential task in computer vision due to the variety of prospective applications, such as security surveillance, machine learning, and human-computer interaction. The availability of more video data than ever before and the lofty performance of deep convolutional neural networks also make it essential for action recognition in video. Unfortunately, limited crafted video features and the scarcity of benchmark datasets make it challenging to address the multi-person action recognition task in video data. In this work, we propose a deep convolutional neural network-based Effective Hybrid Learning (EHL) framework for two-person interaction classification in video data. Our approach exploits a pre-trained network model (the VGG16 from the University of Oxford Visual Geometry Group) and extends the Faster R-CNN (region-based convolutional neural network a state-of-the-art detector for image classification). We broaden a semi-supervised learning method combined with an active learning method to improve overall performance. Numerous types of two-person interactions exist in the real world, which makes this a challenging task. In our experiment, we consider a limited number of actions, such as hugging, fighting, linking arms, talking, and kidnapping in two environment such simple and complex. We show that our trained model with an active semi-supervised learning architecture gradually improves the performance. In a simple environment using an Intelligent Technology Laboratory (ITLab) dataset from Inha University, performance increased to 95.6% accuracy, and in a complex environment, performance reached 81% accuracy. Our method reduces data-labeling time, compared to supervised learning methods, for the ITLab dataset. We also conduct extensive experiment on Human Action Recognition benchmarks such as UT-Interaction dataset, HMDB51 dataset and obtain better performance than state-of-the-art approaches.

• Journal of IKEEE
• /
• v.24 no.4
• /
• pp.1141-1147
• /
• 2020

The hornet species are so similar in shape that they are difficult for non-experts to classify, and because the size of the objects is small and move fast, it is more difficult to detect and classify the species in real time. In this paper, we developed a system that classifies hornets species in real time based on a deep learning algorithm using a boundary box. In order to minimize the background area included in the bounding box when labeling the training image, we propose a method of selecting only the head and body of the hornet. It also experimentally compares existing boundary box-based object recognition algorithms to find the best algorithms that can detect wasps in real time and classify their species. As a result of the experiment, when the mish function was applied as the activation function of the convolution layer and the hornet images were tested using the YOLOv4 model with the Spatial Attention Module (SAM) applied before the object detection block, the average precision was 97.89% and the average recall was 98.69%.

• Journal of Auto-vehicle Safety Association
• /
• v.13 no.2
• /
• pp.35-41
• /
• 2021

In this study, improved path tracking control algorithm based on pure pursuit algorithm is newly proposed by using improved lane detection algorithm through real time post-processing with interpolation methodology. Since the original pure pursuit works well only at speeds below 20 km/h, the look-ahead distance is implemented as a sigmoid function to work well at an average speed of 45 km/h to improve tracking performance. In addition, a smoothing filter was added to reduce the steering angle vibration of the original algorithm, and the stability of the steering angle was improved. The post-processing algorithm presented has implemented more robust lane recognition system using real-time pre/post processing method with deep learning and estimated interpolation. Real time processing is more cost-effective than the method using lots of computing resources and building abundant datasets for improving the performance of deep learning networks. Therefore, this paper also presents improved lane detection performance by using the final results with naive computer vision codes and pre/post processing. Firstly, the pre-processing was newly designed for real-time processing and robust recognition performance of augmentation. Secondly, the post-processing was designed to detect lanes by receiving the segmentation results based on the estimated interpolation in consideration of the properties of the continuous lanes. Consequently, experimental results by utilizing driving guidance line information from processing parts show that the improved lane detection algorithm is effective to minimize the lateral offset error in the diverse maneuvering roads.

• Journal of Internet Computing and Services
• /
• v.22 no.3
• /
• pp.17-26
• /
• 2021

Recently, COVID-19 has spread and time to stay at home has been increasing in accordance with quarantine guidelines of the government such as recommendations to refrain from going out. As a result, the number of single-person households staying at home is also increasingsingle-person households are less likely to be notified to the outside world in times of emergency than multi-person households. This study collects various situations occurring in the home with lidar, image, and voice sensors and analyzes the data according to the sensors through their respective algorithms. Using this method, we analyzed abnormal patterns such as emergency situations and conducted research to detect abnormal signs in humans. Artificial intelligence algorithms that detect abnormalities in people by each sensor were studied and the accuracy of anomaly detection was measured according to the sensor. Furthermore, this work proposes a fusion method that complements the pros and cons between sensors by experimenting with the detectability of sensors for various situations.

• Fashion & Textile Research Journal
• /
• v.22 no.6
• /
• pp.716-726
• /
• 2020

This study presents methods and alternative examples for fashion brands to effectively use video-based communication channels to form brand identity that analyzes the definition, status and type of YouTube channel fashion films as well as enables the ability to derive brand identity characteristics. Literature studies focused on Christian Dior's official website and related previous studies. The temporal range of the case studies was from October 7, 2010, the date when the first fashion film was uploaded to current Christian Dior YouTube to July 17, 2020 (the survey date), and there are a total of 550 subjects for quantitative analysis. The succession of the couture spirit means that Christian Dior's craftsmanship was created and passed down by Musée Christian Dior to act as a contemporary key element of brand identity. The iconic expression of femininity is Dior's core design philosophy that began when the woman image of a new era was presented through a new look, and Dior's femininity means a woman that reflects the character of the times as is interpreted as her own personality from the perspective of modernism through the creative directors of future generations. The brand's core identity code 'Miss Dior' expresses the brand's vision and eternity through perfume as well as targets Z generation male consumers through an emotional approach based on forms that used emotional images such as movie-type films.

• Journal of the Korean Society of Industry Convergence
• /
• v.25 no.2_1
• /
• pp.177-184
• /
• 2022

The automatic defect sorting function of machinery parts is being introduced to the automation of the manufacturing process. In the final stage of automation of the manufacturing process, it is necessary to apply computer vision rather than human visual judgment to determine whether there is a defect. In this paper, we introduce a deep learning method to improve the classification performance of typical mechanical parts, such as welding parts, galvanized round plugs, and electro galvanized nuts, based on the results of experiments. In the case of poor welding, the method to further increase the depth of layer of the basic deep learning model was effective, and in the case of a circular plug, the surrounding data outside the defective target area affected it, so it could be solved through an appropriate pre-processing technique. Finally, in the case of a nut plated with zinc, since it receives data from multiple cameras due to its three-dimensional structure, it is greatly affected by lighting and has a problem in that it also affects the background image. To solve this problem, methods such as two-dimensional connectivity were applied in the object segmentation preprocessing process. Although the experiments suggested that the proposed methods are effective, most of the provided good/defective images data sets are relatively small, which may cause a learning balance problem of the deep learning model, so we plan to secure more data in the future.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.3
• /
• pp.507-514
• /
• 2022

Mobile manipulators are getting lime light in the field of home automation due to their mobility and manipulation capabilities. In this paper, we developed a small size manipulator system that can be mounted on a mobile robot as a preliminary study to develop a mobile manipulator. The developed manipulator has four degree-of-freedom. At the end-effector of manipulator, there are a camera and a gripper to recognize and manipulate the object. One of four degree-of-freedom is linear motion in vertical direction for better interaction with human hands which are located higher than the mobile manipulator. The developed manipulator was designed to dispose the four actuators close to the base of the manipulator to reduce rotational inertia of the manipulator, which improves stability of manipulation and reduces the risk of rollover. The developed manipulator repeatedly performed a pick and place task and successfully manipulate the object within the workspace of manipulator.

• Journal of Korean Neurosurgical Society
• /
• v.65 no.4
• /
• pp.572-581
• /
• 2022

Objective : Compared to microscopes, exoscopes have advantages in field-depth, ergonomics, and educational value. Exoscopes are especially well-poised for adaptation into fluorescence-guided surgery (FGS) due to their excitation source, light path, and image processing capabilities. We evaluated the feasibility of near-infrared FGS using a 3-dimensional (3D), 4 K exoscope with near-infrared fluorescence imaging capability. We then compared it to the most sensitive, commercially-available near-infrared exoscope system (3D and 960 p). In-vitro and intraoperative comparisons were performed. Methods : Serial dilutions of indocyanine-green (1-2000 ㎍/mL) were imaged with the 3D, 4 K Olympus Orbeye (system 1) and the 3D, 960 p VisionSense Iridium (system 2). Near-infrared sensitivity was calculated using signal-to-background ratios (SBRs). In addition, three patients with brain tumors were administered indocyanine-green and imaged with system 1, with two also imaged with system 2 for comparison. Results : Systems 1 and 2 detected near-infrared fluorescence from indocyanine green concentrations of >250 ㎍/L and >31.3 ㎍/L, respectively. Intraoperatively, system 1 visualized strong near-infrared fluorescence from two, strongly gadolinium-enhancing meningiomas (SBR=2.4, 1.7). The high-resolution, bright images were sufficient for the surgeon to appreciate the underlying anatomy in the near-infrared mode. However, system 1 was not able to visualize fluorescence from a weakly-enhancing intraparenchymal metastasis. In contrast, system 2 successfully visualized both the meningioma and the metastasis but lacked high resolution stereopsis. Conclusion : Three-dimensional exoscope systems provide an alternative visualization platform for both standard microsurgery and near-infrared fluorescent guided surgery. However, when tumor fluorescence is weak (i.e., low fluorophore uptake, deep tumors), highly sensitive near-infrared visualization systems may be required.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.17 no.1
• /
• pp.99-104
• /
• 2022

Artificial intelligence is getting a crucial part of our lives with its incredible benefits. Machines outperform humans in recognizing objects in images, particularly in classifying people into correct age and gender groups. In this respect, age and gender classification has been one of the hot topics among computer vision researchers in recent decades. Deployment of deep Convolutional Neural Network(: CNN) models achieved state-of-the-art performance. However, the most of CNN based architectures are very complex with several dozens of training parameters so they require much computation time and resources. For this reason, we propose a new CNN-based classification algorithm with significantly fewer training parameters and training time compared to the existing methods. Despite its less complexity, our model shows better accuracy of age and gender classification on the UTKFace dataset.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.10
• /
• pp.1477-1482
• /
• 2022

Haze is a local weather phenomenon in which very small droplets float in the atmosphere, and the amount and characteristics of haze may vary depending on the region. In particular, these haze reduce visibility, which can cause air traffic interference and vehicle traffic accidents, and degrade the quality of security CCTVs and so on. Therefore, in the past 10 years, research on haze removal has been actively conducted to reduce damage caused by haze. In this study, local haze removal is performed by weight generation using a haziness degree evaluator to adaptively respond to haze-free, homogeneous haze, and non-homogeneous haze cases. And the proposed method improves the limitations of the existing static haze removal method, which assumes that there is haze in the input image and removes the haze. We also demonstrate the superiority of the proposed method through quantitative and qualitative performance evaluations with benchmark algorithms.