• Journal of Internet Computing and Services
• /
• v.20 no.4
• /
• pp.91-102
• /
• 2019

In autonomous driving systems, the ability to classify pedestrians in images captured by cameras is very important for pedestrian safety. In the past, after extracting features of pedestrians with HOG(Histogram of Oriented Gradients) or SIFT(Scale-Invariant Feature Transform), people classified them using SVM(Support Vector Machine). However, extracting pedestrian characteristics in such a handcrafted manner has many limitations. Therefore, this paper proposes a method to classify pedestrians reliably and effectively using CNN's(Convolutional Neural Network) deep features and transfer learning. We have experimented with both the fixed feature extractor and the fine-tuning methods, which are two representative transfer learning techniques. Particularly, in the fine-tuning method, we have added a new scheme, called M-Fine(Modified Fine-tuning), which divideslayers into transferred parts and non-transferred parts in three different sizes, and adjusts weights only for layers belonging to non-transferred parts. Experiments on INRIA Person data set with five CNN models(VGGNet, DenseNet, Inception V3, Xception, and MobileNet) showed that CNN's deep features perform better than handcrafted features such as HOG and SIFT, and that the accuracy of Xception (threshold = 0.5) isthe highest at 99.61%. MobileNet, which achieved similar performance to Xception and learned 80% fewer parameters, was the best in terms of efficiency. Among the three transfer learning schemes tested above, the performance of the fine-tuning method was the best. The performance of the M-Fine method was comparable to or slightly lower than that of the fine-tuningmethod, but higher than that of the fixed feature extractor method.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.17 no.6
• /
• pp.173-184
• /
• 2018

As the portable traffic information collection system using probe vehicles spreads, it is becoming possible to collect road hazard information such as portholes, falling objects, and road surface freezing using in-vehicle sensors in addition to existing traffic information. In this study, we developed a integration and decision algorithm that integrates time and space in real time when multiple probe vehicles detect events such as road hazard information based on GPS coordinates. The core function of the algorithm is to determine whether the road hazard information generated at a specific point is the same point from the result of detecting multiple GPS probes with different GPS coordinates, Generating the data, (3) continuously determining whether the generated event data is valid, and (4) ending the event when the road hazard situation ends. For this purpose, the road risk information collected by the probe vehicle was processed in real time to achieve the conditional probability, and the validity of the event was verified by continuously updating the road risk information collected by the probe vehicle. It is considered that the developed hybrid processing algorithm can be applied to probe-based traffic information collection and event information processing such as C-ITS and autonomous driving car in the future.

• Journal of Cadastre & Land InformatiX
• /
• v.51 no.1
• /
• pp.53-65
• /
• 2021

C-ITS(Cooperative Intelligent Transport System) that pursues traffic safety and convenience uses various sensors to generate traffic information. Therefore, it is necessary to improve the sensor-related technology to increase the efficiency and reliability of the traffic information. Recently, the role of CCTV in collecting video information has become more important due to advances in AI(Artificial Intelligence) technology. In this study, we propose to identify and track dynamic objects(vehicles, people, etc.) in CCTV images, and to analyze and provide information about them in various environments. To this end, we conducted identification and tracking of dynamic objects using the Yolov4 and Deepsort algorithms, establishment of real-time multi-user support servers based on Kafka, defining transformation matrices between images and spatial coordinate systems, and map-based dynamic object visualization. In addition, a positional consistency evaluation was performed to confirm its usefulness. Through the proposed scheme, we confirmed that CCTVs can serve as important sensors to provide relevant information by analyzing road conditions in real time in terms of road infrastructure beyond a simple monitoring role.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.6
• /
• pp.587-598
• /
• 2021

This study aims to develop a multi-camera based MMS (Mobile Mapping System) technology for building a HD (High Definition) map for autonomous driving and for quick update. To replace expensive lidar sensors and reduce long processing times, we intend to develop a low-cost and efficient MMS by applying multiple cameras and real-time data pre-processing. To this end, multi-camera storage technology development, multi-camera time synchronization technology development, and MMS prototype development were performed. We developed a storage module for real-time JPG compression of high-speed images acquired from multiple cameras, and developed an event signal and GNSS (Global Navigation Satellite System) time server-based synchronization method to record the exposure time multiple images taken in real time. And based on the requirements of each sector, MMS was designed and prototypes were produced. Finally, to verify the performance of the manufactured multi-camera-based MMS, data were acquired from an actual 1,000 km road and quantitative evaluation was performed. As a result of the evaluation, the time synchronization performance was less than 1/1000 second, and the position accuracy of the point cloud obtained through SFM (Structure from Motion) image processing was around 5 cm. Through the evaluation results, it was found that the multi-camera based MMS technology developed in this study showed the performance that satisfies the criteria for building a HD map.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.20 no.6
• /
• pp.37-46
• /
• 2021

Government agencies, such as police and local governments, strive to prevent traffic hazards and create a comfortable road environment by pormoting transportation and road facilities. To this end, roads and transportation facilities are enhanced and adjusted, and improvement projects in areas with frequent traffic accidents are carried out. Usually, improvement projects in areas with frequent traffic accidents vary by projects and region. Moreover, these projects are carried out under the supervision of a person in charge and related parties. Hence, civil complaints and subjectivity are reflected in deriving priorities for the improvement projects, limiting the efficiency of the project. To this end, a study was conducted to estimate the direction of improvement of the project target site. This study comprehensively considered road, traffic, and accident conditions of representative projects with high effectiveness in handling traffic accidents. The results of the study state that the accuracy of estimating the improvement project was around 88%. In addition, the study found that there was a strong relationship between traffic volume, accident rate, and accident severity in estimating the improvement direction.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.3
• /
• pp.239-247
• /
• 2022

Cities are becoming more complex due to rapid industrialization and population growth in modern times. In particular, urban areas are rapidly changing due to housing site development, reconstruction, and demolition. Thus accurate road information is necessary for various purposes, such as High Definition Map for autonomous car driving. In the case of the Republic of Korea, accurate spatial information can be generated by making a map through the existing map production process. However, targeting a large area is limited due to time and money. Road, one of the map elements, is a hub and essential means of transportation that provides many different resources for human civilization. Therefore, it is essential to update road information accurately and quickly. This study uses Semantic Segmentation algorithms Such as LinkNet, D-LinkNet, and NL-LinkNet to extract roads from drone images and then apply hyperparameter optimization to models with the highest performance. As a result, the LinkNet model using pre-trained ResNet-34 as the encoder achieved 85.125 mIoU. Subsequent studies should focus on comparing the results of this study with those of studies using state-of-the-art object detection algorithms or semi-supervised learning-based Semantic Segmentation techniques. The results of this study can be applied to improve the speed of the existing map update process.

• Journal of IKEEE
• /
• v.26 no.2
• /
• pp.141-149
• /
• 2022

With the development of artificial intelligence and unmanned technologies, the remote surveillance/autonomous driving systems have been actively researched. For an effective performance analysis of the developed remote control system, it is important to record the data of it in real time. In addition, in order to analyze the performance between the control system and the remote system, the recorded data from them should be synchronized with time. In this paper we proposed a novel time synchronization method for the remote control system. The proposed remote control system satisfies the time difference of the recorded data within 1 ms, and we can reduce the time difference by using a CPU shielding and affinity setting. The performance of the proposed method was proved through various network data storage experiments. And the experiments confirmed that the proposed method can be applied to recording devices of unmanned ground vehicles and control vehicles. The proposed method will be used as a method for analyzing network data of UGV-R (Unmanned Ground Vehicle - Reconnaissance).

• Journal of Convergence for Information Technology
• /
• v.11 no.4
• /
• pp.83-101
• /
• 2021

This study is to investigate, using AHP, what technologies should be applied first to Korean SMEs in order to respond to the 4th industrial revolution and change to a smart enterprise. To this end, technologies related to the 4th industrial revolution and smart factory are synthesized, and the classification criteria of Dae-Hoon Kim et al. (2019) are applied, but additional opinions of experts are collected and related technologies are converted to artificial intelligence (AI), Big Data, and Cloud Computing. As a base technology, mobile, Internet of Things (IoT), block chain as hyper-connected technology, unmanned transportation (autonomous driving), robot, 3D printing, drone as a convergence technology, smart manufacturing and logistics, smart healthcare, smart transportation and smart finance were classified as smart industrial technologies. As a result of confirming the priorities for technical use by AHP analysis and calculating the total weight, manufacturing companies have a high ranking in mobile, artificial intelligence (AI), big data, and robots, while service companies are in big data and robots, artificial intelligence (AI), and smart healthcare are ranked high, and in all companies, it is in the order of big data, artificial intelligence (AI), robot, and mobile. Through this study, it was clearly identified which technologies should be applied first in order to respond to the 4th industrial revolution and change to a smart company.

• Journal of the Korean Society of Industry Convergence
• /
• v.24 no.6_2
• /
• pp.729-735
• /
• 2021

Recently Artificial Intelligence(AI) has been developed and used in various fields. Especially AI recognition technology can perceive and distinguish images so it should plays a significant role in quality inspection process. For stability of autonomous driving technology, semiconductors inside automobiles must be protected from external electromagnetic wave(EM wave). As a shield film, a thin polymeric material with hole shaped micro-patterns created by a laser processing could be used for the protection. The shielding efficiency of the film can be increased by the hole structure with appropriate pitch and size. However, since the sensitivity of micro-machining for some parameters, the shape of every single hole can not be same, even it is possible to make defective patterns during process. And it is absolutely time consuming way to inspect all patterns by just using optical microscope. In this paper, we introduce a AI inspection system which is based on web site AI tool. And we evaluate the usefulness of AI model by calculate Area Under ROC curve(Receiver Operating Characteristics). The AI system can classify the micro-patterns into normal or abnormal ones displaying the text of the result on real-time images and save them as image files respectively. Furthermore, pressing the running button, the Hardware of robot arm with two Arduino motors move the film on the optical microscopy stage in order for raster scanning. So this AI system can inspect the entire micro-patterns of a film automatically. If our system could collect much more identified data, it is believed that this system should be a more precise and accurate process for the efficiency of the AI inspection. Also this one could be applied to image-based inspection process of other products.

• Clean Technology
• /
• v.28 no.3
• /
• pp.227-231
• /
• 2022

The demand for high-voltage power devices is rising in various industries, but especially in the transportation industry due to autonomous driving and electric vehicles. IGBT module parts of 3.3 kV or more are used in the power propulsion control device of electric vehicles, and the procurement of these parts for new construction and maintenance is increasing every year. In addition, research to optimize high-voltage IGBT parts is urgently required to overcome their very high technology entry barrier. For the development of high-voltage IGBT devices over 3.3 kV, the resistivity range setting of the wafer and the optimal conditions for major unit processes are important variables. Among the manufacturing processes to secure the optimal junction depth, the optimization of the diffusion process, which is one step of the unit process, was examined. In the diffusion process, the type of gas injected, the injection time, and the injection temperature are the main variables. In this study, the range of wafer resistance (Ω cm) was set for the development of high voltage IGBT devices through unit process simulation. Additionally, the well drive in (WDR) condition optimization of the diffusion process according to temperature was studied. The junction depth was 7.4 to7.5 ㎛ for a ring pattern width of 23.5 to25.87 ㎛, which can be optimized for supporting 3.3 kV high voltage power devices.