• Journal of the Semiconductor & Display Technology
• /
• v.19 no.2
• /
• pp.60-67
• /
• 2020

Recently, artificial intelligence related technologies including machine learning are being applied to various fields, and the demand is also increasing. In particular, with the development of AR, VR, and MR technologies related to image processing, the utilization of computer vision based on deep learning has increased. The algorithms for object recognition and detection based on deep learning required for image processing are diversified and advanced. Accordingly, problems that were difficult to solve with the existing methodology were solved more simply and easily by using deep learning. This paper introduces various deep learning-based object recognition and extraction algorithms used to detect and recognize various objects in an image and analyzes the technologies that attract attention.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.49 no.10
• /
• pp.883-892
• /
• 2021

This paper provides a process for developing deep learning-based aerial object detection models that can run in realtime on onboard. To improve object detection performance, we pre-process and augment the training data in the training stage. In addition, we perform transfer learning and apply a weighted cross-entropy method to reduce the variations of detection performance for each class. To improve the inference speed, we have generated inference acceleration engines with quantization. Then, we analyze the real-time performance and detection performance on custom aerial image dataset to verify generalization.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.2
• /
• pp.20-25
• /
• 2021

The target-object classification method was implemented using a deep-learning-based detection model in real-time images. The object detection model was a deep-learning-based detection model that allowed extensive data collection and machine learning processes to classify similar target-objects. The recognition model was implemented by changing the processing structure of the detection model and combining developed the vision-processing module. To classify the target-objects, the identity and similarity were defined and applied to the detection model. The use of the recognition model in industry was also considered by verifying the effectiveness of the recognition model using the real-time images of an actual soccer game. The detection model and the newly constructed recognition model were compared and verified using real-time images. Furthermore, research was conducted to optimize the recognition model in a real-time environment.

• Smart Structures and Systems
• /
• v.24 no.6
• /
• pp.745-757
• /
• 2019

Road surface deteriorations such as potholes have caused motorists heavy monetary damages every year. However, effective road condition monitoring has been a continuing challenge to road owners. Depth cameras have a small field of view and can be easily affected by vehicle bouncing. Traditional image processing methods based on algorithms such as segmentation cannot adapt to varying environmental and camera scenarios. In recent years, novel object detection methods based on deep learning algorithms have produced good results in detecting typical objects, such as faces, vehicles, structures and more, even in scenarios with changing object distances, camera angles, lighting conditions, etc. Therefore, in this study, a Deep Learning Pothole Detector (DLPD) based on the deep region-based convolutional neural network is proposed for autonomous detection of potholes from images. About 900 images with potholes and road surface conditions are collected and divided into training and testing data. Parameters of the network in the DLPD are calibrated based on sensitivity tests. Then, the calibrated DLPD is trained by the training data and applied to the 215 testing images to evaluate its performance. It is demonstrated that potholes can be automatically detected with high average precision over 93%. Potholes can be differentiated from manholes by training and applying a manhole-pothole classifier which is constructed using the convolutional neural network layers in DLPD. Repeated detection of the same potholes can be prevented through feature matching of the newly detected pothole with previously detected potholes within a small region.

• Journal of Korea Multimedia Society
• /
• v.20 no.5
• /
• pp.769-781
• /
• 2017

Deep learning networks like Convolutional Neural Networks (CNNs) show successful performances in many computer vision applications such as image classification, object detection, and so on. For implementation of deep learning networks in embedded system with limited processing power and memory, deep learning network may need to be simplified. However, simplified deep learning network cannot learn every possible scene. One realistic strategy for embedded deep learning network is to construct a simplified deep learning network model optimized for the scene images of the installation place. Then, automatic training will be necessitated for commercialization. In this paper, as an intermediate step toward automatic training under fisheye camera environments, we study more precise human localization in fisheye images, and propose an accurate human localization method, Automatic Ground-Truth Labelling Method (AGTLM). AGTLM first localizes candidate human object bounding boxes by utilizing GoogLeNet-LSTM approach, and after reassurance process by GoogLeNet-based CNN network, finally refines them more correctly and precisely(tightly) by applying saliency object detection technique. The performance improvement of the proposed human localization method, AGTLM with respect to accuracy and tightness is shown through several experiments.

• Journal of the Institute of Convergence Signal Processing
• /
• v.22 no.1
• /
• pp.30-37
• /
• 2021

Accurate 3D object tracking with camera images is a key enabling technology for augmented reality applications. Motivated by the impressive success of convolutional neural networks (CNNs) in computer vision tasks such as image classification, object detection, image segmentation, recent studies for 3D object tracking have focused on leveraging deep learning. In this paper, we review deep learning approaches for 3D object tracking. We describe key methods in this field and discuss potential future research directions.

• Journal of the Semiconductor & Display Technology
• /
• v.20 no.3
• /
• pp.140-145
• /
• 2021

This paper deals with the development of a deep-learning-based runway cleaning robot using an optical camera. A suitable model to realize real-time object detection was investigated, and the differences between the selected YOLOv3 and other deep learning models were analyzed. In order to check whether the proposed system is applicable to the actual runway, an experiment was conducted by making a prototype of the robot and a runway model. As a result, it was confirmed that the robot was well developed because the detection rate of FOD (Foreign Object Debris) and cracks was high, and the collection of foreign substances was carried out smoothly.

• Journal of Broadcast Engineering
• /
• v.25 no.7
• /
• pp.1081-1094
• /
• 2020

With the recent development of deep learning, most computer vision-related tasks are being solved with deep learning-based network technologies such as CNN and RNN. Computer vision tasks such as object detection or object segmentation use intermediate features extracted from the same backbone such as Resnet or FPN for training and inference for object detection and segmentation. In this paper, an experiment was conducted to find out the compression efficiency and the effect of encoding on task inference performance when the features extracted in the intermediate stage of CNN are encoded. The feature map that combines the features of 256 channels into one image and the original image were encoded in HEVC to compare and analyze the inference performance for object detection and segmentation. Since the intermediate feature map encodes the five levels of feature maps (P2 to P6), the image size and resolution are increased compared to the original image. However, when the degree of compression is weakened, the use of feature maps yields similar or better inference results to the inference performance of the original image.

• Journal of Korea Multimedia Society
• /
• v.25 no.10
• /
• pp.1433-1447
• /
• 2022

Recently, as the demand for physical security technology to prevent leakage of technical and business information of companies and public institutions increases, the high tech companies are operating X-ray security checkpoints at building entrances to protect their intellectual property and technology. X-ray security checkpoints are operated to detect cameras and storage media that may store or leak important technologies in the bags of people entering and leaving the building. In this study, we propose an X-ray security checkpoint system that automatically detects a storage medium in an X-ray image using a deep learning based object detection method. The proposed system consists of an edge computing unit and a cloud-computing unit. We employ the RetinaNet for automatic storage media detection in the X-ray security checkpoint images. The proposed approach achieved mAP of 95.92% on private dataset.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.1
• /
• pp.1-8
• /
• 2022

In this paper, we propose a fire detection system based on CCTV images using an object tracking technology with YOLOv4 model capable of real-time object detection and a DeepSORT algorithm. The fire detection model was learned from 10800 pieces of learning data and verified through 1,000 separate test sets. Subsequently, the fire detection rate in a single image and fire detection maintenance performance in the image were increased by tracking the detected fire area through the DeepSORT algorithm. It is verified that a fire detection rate for one frame in video data or single image could be detected in real time within 0.1 second. In this paper, our AI fire detection system is more stable and faster than the existing fire accident detection system.