• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.10
• /
• pp.323-328
• /
• 2020

Recently, new approaches that significantly improve performance in object detection and recognition using deep learning technology have been proposed quickly. Of the various techniques for object detection, especially facial object detection (Faster R-CNN, R-CNN, YOLO, SSD, etc), SSD is superior in accuracy and speed to other techniques. At the same time, multiple object detection networks are also readily available. In this paper, among object detection networks, Mobilenet v2 network is used, models combined with SSDs are trained, and methods for detecting objects at a rate of four times or more than conventional performance are proposed using TensorRT engine, and the performance is verified through experiments. Facial object detector was created as an application to verify the performance of the proposed method, and its behavior and performance were tested in various situations.

• Journal of KIISE:Information Networking
• /
• v.30 no.4
• /
• pp.504-512
• /
• 2003

In this paper, we propose a new adaptive dissolve detection method based on the analysis of a dissolve modeling error which is the difference between an ideally modeled dissolve curve with no correlation and an actual dissolve curve including a correlation. The proposed dissolve detection method consists of two steps. First, candidate dissolve regions are extracted using the characteristics of a downward convex parabola, then each candidate region is verified based oil the dissolve modeling error. If the dissolve modeling error for a candidate region is less than a threshold defined by the target modeling error with a target correlation, the candidate region is determined as a resolve region with a lower correlation than the target correlation. The threshold is adaptively determined based on the variances between the candidate regions and the target correlation. By considering the correlation between neighbor scenes, the proposed method is able to be a semantic scene-change detector. The proposed method was tested on various types of data and its performance proved to be more accurate and reliable regardless of variation of variance of test sequences when compared with other commonly use methods.

• Journal of IKEEE
• /
• v.22 no.4
• /
• pp.1079-1087
• /
• 2018

Recently, computer vision field based deep learning artificial intelligence has become a hot topic among various image analysis boundaries. In this study, flames are detected in fire images using the Faster R-CNN algorithm, which is used to detect objects within the image, among various image recognition algorithms based on deep learning. In order to improve fire detection accuracy through a small amount of data sets in the learning process, we use image augmentation techniques, and learn image augmentation by dividing into 6 types and compare accuracy, precision and detection rate. As a result, the detection rate increases as the type of image augmentation increases. However, as with the general accuracy and detection rate of other object detection models, the false detection rate is also increased from 10% to 30%.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.4
• /
• pp.373-379
• /
• 2019

In this study, image-based flock monitoring and density evaluation were conducted for broiler chicks welfare. Image data were captured by using a mono camera and region of broiler chicks in the image was detected using converting to HSV color model, thresholding, and clustering with filtering. The results show that region detection was performed with 5% relative error and 0.81 IoU on average. The detected region was corrected to the actual region by projection into ground using coordinate transformation between camera and real-world. The flock density of broiler chicks was estimated using the corrected actual region, and it was observed with an average of 80%. The developed algorithm can be applied to the broiler chicks house through enhancing accuracy of region detection and low-cost system configuration.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.4
• /
• pp.19-30
• /
• 2009

In this paper, We propose a method with which we can detect facial components and face in input image. We use eye map and mouth map to detect facial components using eyes and mouth. First, We find out eye zone, and second, We find out color value distribution of skin region using the color around the eye zone. Skin region have characteristic distribution in YCbCr color space. By using it, we separate the skin region and background area. We find out the color value distribution of the extracted skin region and extract around the region. Then, detect mouth using mouthmap from extracted skin region. Proposed method is better than traditional method the reason for it comes good result with accurate mouth region.

• Journal of KIISE:Software and Applications
• /
• v.36 no.7
• /
• pp.548-559
• /
• 2009

The cascade face detector learned by Adaboost algorithm, which was proposed by Viola and Jones, is state of the art face detector due to its great speed and accuracy. In spite of its great performance, it still suffers from false alarms, and more computation is required to reduce them. In this paper, we want to reduce false alarms with less computation using facial color. Using facial color information, proposed face detection model scans sub-window efficiently and adapts a fast face/non-face classifier at the first stage of cascade face detector. This makes face detection faster and reduces false alarms. For facial color filtering, we define a facial color membership function, and facial color filtering image is obtained using that. An integral image is calculated from facial color filtering image. Using this integral image, its density of subwindow could be obtained very fast. The proposed scanning method skips over sub-windows that do not contain possible faces based on this density. And the face/non-face classifier at the first stage of cascade detector rejects a non-face quickly. By experiment, we show that the proposed face detection model reduces false alarms and is faster than the original cascade face detector.

• KIPS Transactions on Software and Data Engineering
• /
• v.1 no.3
• /
• pp.177-186
• /
• 2012

In this paper, we propose a real-time approach for detecting and tracking a hand region by analyzing depth images. We build a hand model in advance. The model has the shape information of a hand. The detecting process extracts out moving areas in an image, which are possibly caused by moving a hand in front of a camera. The moving areas can be identified by analyzing accumulated difference images and applying the region growing technique. The extracted moving areas are compared against a hand model to get justified as a hand region. The tracking process keeps the track of center points of hand regions of successive frames. For this purpose, it involves three steps. The first step is to determine a seed point that is the closest point to the center point of a previous frame. The second step is to perform region growing to form a candidate region of a hand. The third step is to determine the center point of a hand to be tracked. This point is searched by the mean-shift algorithm within a confined area whose size varies adaptively according to the depth information. To verify the effectiveness of our approach, we have evaluated the performance of our approach while changing the shape and position of a hand as well as the velocity of hand movement.

• Journal of Institute of Control, Robotics and Systems
• /
• v.7 no.9
• /
• pp.749-757
• /
• 2001

We present a simple artificial landmark model and robust landmark tracking algorithm for mobile robot localization. The landmark model, consisting of symmetric and repetitive color patches, produces color histograms that are invariant under the geometric and photometric distortions. A stochastic approach based on the CONDENSATION tracks the landmark model robustly even under the varying illumination conditions. After the landmark detection, relative position of the mobile robot to the landmark is calculated. Experimental results show that the proposed landmark model is effective and can be detected and tracked in a clustered scene robustly. With the tracked single landmark, we extract geometrical information than achieve accurate localization.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.40 no.12
• /
• pp.39-45
• /
• 2003

Design verification of VHDL models is getting difficult and has become a critical and time-consuming process in hardware design. Recent]y the methods using Bayesian estimation and stopping rule have been introduced to verify behavioral models and to reduce clock cycles. This paper presents two strategies to reduce clock cycles when using stopping rule in a VHDL model verification. The first method is that a semi-random variable is defined and the data that stay in the range of semi-random variable are skipped when stopping rule is running. The second one is to keep the old values of parameters when phases of stopping rule are changed. 12 VHDL models are examined to observe the effectiveness of strategies, and the simulation results show that more than about 25% of clock cycles is reduced by using the two proposed strategies with 0.6% losses of branch coverage rate.

• Journal of KIISE:Software and Applications
• /
• v.36 no.1
• /
• pp.82-91
• /
• 2009

Sign language spotting is the task of detecting and recognizing the signs in a signed utterance. The difficulty of sign language spotting is that the occurrences of signs vary in both motion and shape. Moreover, the signs appear within a continuous gesture stream, interspersed with transitional movements between signs in a vocabulary and non-sign patterns(which include out-of-vocabulary signs, epentheses, and other movements that do not correspond to signs). In this paper, a novel method for designing a threshold model in a conditional random field(CRF) model is proposed. The proposed model performs an adaptive threshold for distinguishing between signs in the vocabulary and non-sign patterns. A hand appearance-based sign verification method, a short-sign detector, and a subsign reasoning method are included to further improve sign language spotting accuracy. Experimental results show that the proposed method can detect signs from continuous data with an 88% spotting rate and can recognize signs from isolated data with a 94% recognition rate, versus 74% and 90% respectively for CRFs without a threshold model, short-sign detector, subsign reasoning, and hand appearance-based sign verification.