• 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.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2020.07a
• /
• pp.285-286
• /
• 2020

전 세계적으로 유행하는 COVID-19(코로나19)로 인해 사람들은 대면 접촉을 피하게 되었고, 전염성이 높은 이유로 마스크의 착용이 의무화되고 있고, 이를 검사하는 업무가 증가하고 있다. 그래서, 인공지능 기술을 통해 업무를 도와줄 수 있는 출입자 통계와 출입자 마스크 착용 검사를 할 수 있는 시스템이 필요하다. 이를 위해 본 논문에서는 딥러닝 알고리즘을 활용한 출입자 통계와 마스크 착용 판별 시스템을 제시한다. 또한, 실시간 영상인식에 많이 활용되고 있는 YOLO-v3와 YOLO-v4, YOLO-Tiny 알고리즘을 데스크탑 PC와 Nvidia사의 Jetson Nano에 적용하여 알고리즘별 성능 비교를 통해 적합한 방법을 찾고 적용하였다.

• Korean Journal of Remote Sensing
• /
• v.38 no.3
• /
• pp.237-250
• /
• 2022

In this study, we propose a method to estimate the biomass of invertebrate grazers from the videos with underwater drones by using a multi-object tracking model based on deep learning. In order to detect invertebrate grazers by classes, we used YOLOv5 (You Only Look Once version 5). For biomass estimation we used DeepSORT (Deep Simple Online and real-time tracking). The performance of each model was evaluated on a workstation with a GPU accelerator. YOLOv5 averaged 0.9 or more mean Average Precision (mAP), and we confirmed it shows about 59 fps at 4 k resolution when using YOLOv5s model and DeepSORT algorithm. Applying the proposed method in the field, there was a tendency to be overestimated by about 28%, but it was confirmed that the level of error was low compared to the biomass estimation using object detection model only. A follow-up study is needed to improve the accuracy for the cases where frame images go out of focus continuously or underwater drones turn rapidly. However,should these issues be improved, it can be utilized in the production of decision support data in the field of invertebrate grazers control and monitoring in the future.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.125-132
• /
• 2023

Roundabouts have strengths in traffic flow and safety but can present difficulties for inexperienced drivers. Demand to acquire and analyze drone images has increased to enhance a traffic environment allowing drivers to deal with roundabouts easily. In this paper, we propose a roundabout traffic analysis system that detects, tracks, and analyzes vehicles using a deep learning-based object detection model (YOLOv7) in drone images. About 3600 images for object detection model learning and testing were extracted and labeled from 1 hour of drone video. Through training diverse conditions and evaluating the performance of object detection models, we achieved an average precision (AP) of up to 97.2%. In addition, we utilized SORT (Simple Online and Realtime Tracking) and OC-SORT (Observation-Centric SORT), a real-time object tracking algorithm, which resulted in an average MOTA (Multiple Object Tracking Accuracy) of up to 89.2%. By implementing a method for measuring roundabout entry speed, we achieved an accuracy of 94.5%.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.6
• /
• pp.505-514
• /
• 2020

Since a large amount of training data is typically needed to train Deep Neural Networks (DNNs), a parallel training approach is required to train the DNNs. The Stochastic Gradient Descent (SGD) algorithm is one of the most widely used methods to train the DNNs. However, since the SGD is an inherently sequential process, it requires some sort of approximation schemes to parallelize the SGD algorithm. In this paper, we review various efforts on parallelizing the SGD algorithm, and analyze the computational overhead, communication overhead, and the effects of the approximations.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.5
• /
• pp.163-169
• /
• 2021

In this paper, we implemented a deep learning-based automatic object tracking and handy motion control drone system and analyzed the performance of the proposed system. The drone system automatically detects and tracks targets by analyzing images obtained from the drone's camera using deep learning algorithms, consisting of the YOLO, the MobileNet, and the deepSORT. Such deep learning-based detection and tracking algorithms have both higher target detection accuracy and processing speed than the conventional color-based algorithm, the CAMShift. In addition, in order to facilitate the drone control by hand from the ground control station, we classified handy motions and generated flight control commands through motion recognition using the YOLO algorithm. It was confirmed that such a deep learning-based target tracking and drone handy motion control system stably track the target and can easily control the drone.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.4
• /
• pp.1066-1079
• /
• 2023

Multi-target tracking based on the detector is a very hot and important research topic in target tracking. It mainly includes two closely related processes, namely target detection and target tracking. Where target detection is responsible for detecting the exact position of the target, while target tracking monitors the temporal and spatial changes of the target. With the improvement of the detector, the tracking performance has reached a new level. The problem that always exists in the research of target tracking is the problem that occurs again after the target is occluded during tracking. Based on this question, this paper proposes a DeepSORT model based on SIFT features to improve ship tracking. Unlike previous feature extraction networks, SIFT algorithm does not require the characteristics of pre-training learning objectives and can be used in ship tracking quickly. At the same time, we improve and test the matching method of our model to find a balance between tracking accuracy and tracking speed. Experiments show that the model can get more ideal results.

• Smart Media Journal
• /
• v.11 no.8
• /
• pp.84-92
• /
• 2022

Precision livestock monitoring promises greater management efficiency for farmers and higher welfare standards for animals. Recent studies on video-based animal activity recognition and tracking have shown promising solutions for understanding animal behavior. To achieve that, surveillance cameras are installed diagonally above the barn in a typical cattle farm setup to monitor animals constantly. Under these circumstances, tracking individuals requires addressing challenges such as occlusion and visual appearance, which are the main reasons for track breakage and increased misidentification of animals. This paper presents a framework for multi-cattle tracking in closed barns with appearance and motion models. To overcome the above challenges, we modify the DeepSORT algorithm to achieve higher tracking accuracy by three contributions. First, we reduce the weight of appearance information. Second, we use an Ensemble Kalman Filter to predict the random motion information of cattle. Third, we propose a supplementary matching algorithm that compares the absolute cattle position in the barn to reassign lost tracks. The main idea of the matching algorithm assumes that the number of cattle is fixed in the barn, so the edge of the barn is where new trajectories are most likely to emerge. Experimental results are performed on our dataset collected on two cattle farms. Our algorithm achieves 70.37%, 77.39%, and 81.74% performance on HOTA, AssA, and IDF1, representing an improvement of 1.53%, 4.17%, and 0.96%, respectively, compared to the original method.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.1
• /
• pp.63-70
• /
• 2023

Due to the spread of high-definition black boxes and the introduction of mobile applications such as 'Smart Citizens Report' and 'Safety Report', the number of public interest reports for violations of Traffic Law has increased rapidly, resulting in shortage of police personnel to handle them. In this paper, we describe the development of a system that can automatically detect lane violations which account for the largest proportion of public interest reporting videos for violations of traffic laws, using deep learning algorithms. In this study, a method for recognizing a vehicle and a solid line object using a YOLO model and a Lanenet model, a method for tracking an object individually using a deep sort algorithm, and a method for detecting lane change violations by recognizing the overlapping range of a vehicle object's bounding box and a solid line object are described. Using this system, it is expected that the shortage of police personnel in charge will be resolved.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.3
• /
• pp.840-860
• /
• 2023

Aiming at the serious occlusion and slow tracking speed in pedestrian target tracking and recognition in complex scenes, a target tracking method based on improved YOLO v5 combined with Deep SORT is proposed. By merging the attention mechanism ECA-Net with the Neck part of the YOLO v5 network, using the CIoU loss function and the method of CIoU non-maximum value suppression, connecting the Deep SORT model using Shuffle Net V2 as the appearance feature extraction network to achieve lightweight and fast speed tracking and the purpose of improving tracking under occlusion. A large number of experiments show that the improved YOLO v5 increases the average precision by 1.3% compared with other algorithms. The improved tracking model, MOTA reaches 54.3% on the MOT17 pedestrian tracking data, and the tracking accuracy is 3.7% higher than the related algorithms and The model presented in this paper improves the FPS by nearly 5 on the fps indicator.