• Journal of Korea Multimedia Society
• /
• v.22 no.3
• /
• pp.349-356
• /
• 2019

When a prescription change occurs in the hospital depending on a patient's improvement status, pharmacists directly classify manually returned pills which are not taken by a patient. There are hundreds of kinds of pills to classify. Because it is manual, mistakes can occur and which can lead to medical accidents. In this study, we have compared YOLO, Faster R-CNN and RetinaNet to classify and detect pills. The data consisted of 10 classes and used 100 images per class. To evaluate the performance of each model, we used cross-validation. As a result, the YOLO Model had sensitivity of 91.05%, FPs/image of 0.0507. The Faster R-CNN's sensitivity was 99.6% and FPs/image was 0.0089. The RetinaNet showed sensitivity of 98.31% and FPs/image of 0.0119. Faster RCNN showed the best performance among these three models tested. Thus, the most appropriate model for classifying pills among the three models is the Faster R-CNN with the most accurate detection and classification results and a low FP/image.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.16 no.2
• /
• pp.36-54
• /
• 2017

The purpose of this paper is to develop a multi-sensor fusion-based traffic information acquisition system with robust to environmental changes. it combines the characteristics of each sensor and is more robust to the environmental changes than the video detector. Moreover, it is not affected by the time of day and night, and has less maintenance cost than the inductive-loop traffic detector. This is accomplished by synthesizing object tracking informations based on a radar, vehicle classification informations based on a video detector and reliable object detections of a infrared range finder. To prove the effectiveness of the proposed system, I conducted experiments for 6 hours over 5 days of the daytime and early evening on the pedestrian - accessible road. According to the experimental results, it has 88.7% classification accuracy and 95.5% vehicle detection rate. If the parameters of this system is optimized to adapt to the experimental environment changes, it is expected that it will contribute to the advancement of ITS.

• Journal of Broadcast Engineering
• /
• v.25 no.4
• /
• pp.576-586
• /
• 2020

Analyzing baseball games using sensors such as radars or riders is expensive. In this paper, we propose an algorithm to detect pitch shots and hit shots using baseball video and to generate ball trajectories within hit shots using camera movement. After the pitch shot and the hit shot detection using object detection and optical flow, we generate the transformation relationship between frames and ball locations in the frame, and calculates the ball trajectory. The performance of the proposed method is evaluated for three KBO baseball video sequences, and the detection accuracy and detection rate of pitch shot and hit shot were within 89-95 [%], and the average error for shot range was 13.6[m], The direction error was 7.5° and foul classification accuracy was 98.6%.

• The Journal of Bigdata
• /
• v.9 no.1
• /
• pp.89-100
• /
• 2024

In the field of computer vision, models such as You Look Only Once (YOLO) and ResNet are widely used due to their real-time performance and high accuracy. However, to apply these models in real-world environments, factors such as runtime compatibility, memory usage, computing resources, and real-time conditions must be considered. This study compares the characteristics of three deep model runtimes: ONNX Runtime, TensorRT, and OpenCV DNN, and analyzes their performance on two models. The aim of this paper is to provide criteria for runtime selection for practical applications. The experiments compare runtimes based on the evaluation metrics of time, memory usage, and accuracy for vehicle license plate recognition and classification tasks. The experimental results show that ONNX Runtime excels in complex object detection performance, OpenCV DNN is suitable for environments with limited memory, and TensorRT offers superior execution speed for complex models.

• Journal of Ocean Engineering and Technology
• /
• v.34 no.4
• /
• pp.277-284
• /
• 2020

Underwater acoustics, which is the study of phenomena related to sound waves in water, has been applied mainly in research on the use of sound navigation and range (SONAR) systems for communication, target detection, investigation of marine resources and environments, and noise measurement and analysis. The main objective of underwater acoustic remote sensing is to obtain information on a target object indirectly by using acoustic data. Presently, various types of machine learning techniques are being widely used to extract information from acoustic data. The machine learning techniques typically used in underwater acoustics and their applications in passive SONAR systems were reviewed in the first two parts of this work (Yang et al., 2020a; Yang et al., 2020b). As a follow-up, this paper reviews machine learning applications in SONAR signal processing with a focus on active target detection and classification.

• IEIE Transactions on Smart Processing and Computing
• /
• v.5 no.5
• /
• pp.327-330
• /
• 2016

Currently, the world's elderly population continues to grow at a dramatic rate. As the number of senior citizens increases, detection of someone falling has attracted increasing attention for visual surveillance systems. This paper presents a novel fall-detection algorithm using motion estimation and an integrated spatiotemporal energy map of the object region. The proposed method first extracts a human region using a background subtraction method. Next, we applied an optical flow algorithm to estimate motion vectors, and an energy map is generated by accumulating the detected human region for a certain period of time. We can then detect a fall using k-nearest neighbor (kNN) classification with the previously estimated motion information and energy map. The experimental results show that the proposed algorithm can effectively detect someone falling in any direction, including at an angle parallel to the camera's optical axis.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2007.11a
• /
• pp.754-759
• /
• 2007

This study exploits the data discriminating capability of silhouette statistics, which combines wavelet-based vertical energy threshold technique for the purpose of extracting damage-sensitive features and clustering signals of the same class. This threshold technique allows to first obtain a suitable subset of the extracted or modified features of our data, i.e., good predictor sets should contain features that are strongly correlated to the characteristics of the data without considering the classification method used, although each of these features should be as uncorrelated with each other as possible. The silhouette statistics have been used to assess the quality of clustering by measuring how well an object is assigned to its corresponding cluster. We use this concept for the discriminant power function used in this paper. The simulation results of damage detection in a truss structure show that the approach proposed in this study can be successfully applied for locating both open- and breathing-type damage even in the presence of a considerable amount of process and measurement noise.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10a
• /
• pp.430-433
• /
• 1996

The shape of an object plays a very important role in pattern analysis and classification. Roughly, the researches on this topic can be classified into three fields, i.e. (i) edge detection, (ii) dominant points extraction, and (iii) shape recognition and classification. Many works have been done in these three fields. However, it is very seldom to see the research that discusses the connection relationship of objects. This problem is very important in robot assembly systems. Therefore, here we focus on this problem and discuss how to recover the connection relationship of planar objects. Our method is based on the partial curve identification algorithm. The experiment results show the efficiency and validity of this method.

• Journal of the korean academy of Pediatric Dentistry
• /
• v.51 no.1
• /
• pp.1-10
• /
• 2024

Deep learning algorithms are becoming more prevalent in dental research because they are utilized in everyday activities. However, dental researchers and clinicians find it challenging to interpret deep learning studies. This review aimed to provide an overview of the general concept of deep learning and current deep learning research in dental radiographic image analysis. In addition, the process of implementing deep learning research is described. Deep-learning-based algorithmic models perform well in classification, object detection, and segmentation tasks, making it possible to automatically diagnose oral lesions and anatomical structures. The deep learning model can enhance the decision-making process for researchers and clinicians. This review may be useful to dental researchers who are currently evaluating and assessing deep learning studies in the field of dentistry.

• Journal of Broadcast Engineering
• /
• v.13 no.6
• /
• pp.875-882
• /
• 2008

Ear detection in an image processing is the one of the important area in biometrics. In this paper we propose a human ear detection algorithm with side face images. First, we search a face candidate area in an input image by using skin-color model and try to find an ear area based on Haar-like feature. Then, to verity whether it is the ear area or not, we use the template which is excellent object classification compare to recognize the characters in the plate. In this experiment, the proposed method showed that the processing speed is improved by 60% than previous works and the detection success rate is 92%.