• The Journal of the Acoustical Society of Korea
• /
• v.37 no.6
• /
• pp.506-511
• /
• 2018

This paper proposes an GMM(Gaussian Mixture Model)-DNN(Deep Neural Network) hybrid-based feature compensation method for effective speech recognition in noisy environments. In the proposed algorithm, the posterior probability for the conventional GMM-based feature compensation method is calculated using DNN. The experimental results using the Aurora 2.0 framework and database demonstrate that the proposed GMM-DNN hybrid-based feature compensation method shows more effective in Known and Unknown noisy environments compared to the GMM-based method. In particular, the experiments of the Unknown environments show 9.13 % of relative improvement in the average of WER (Word Error Rate) and considerable improvements in lower SNR (Signal to Noise Ratio) conditions such as 0 and 5 dB SNR.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.9
• /
• pp.1-6
• /
• 2019

One method for image recognition is template matching. In conventional template matching, the block matching algorithm (BMA) is performed while changing the two-dimensional translational displacement of the template within a given matching image. The template size and shape do not change during the BMA. Since only two-dimensional translational displacement is considered, the success rate decreases if the size and direction of the object do not match in the template and the matching image. In this paper, a variable is added to adjust the two-dimensional direction and size of the template, and the optimal value of the variable is automatically calculated in the block corresponding to each two-dimensional translational displacement. Using the calculated optimal value, the template is automatically transformed into an optimal template for each block. The matching error value of each block is then calculated based on the automatically deformed template. Therefore, a more stable result can be obtained for the difference in direction and size. For ease of use, this study focuses on designing the algorithm in a closed form that does not require additional information beyond the template image, such as distance information.

• Journal of IKEEE
• /
• v.23 no.3
• /
• pp.777-784
• /
• 2019

A pre-fire awareness and automatic notification system are required because it is possible to minimize the damage if the fire situation is precisely detected after a fire occurs in a place where people are unusual or in a mountainous area. In this study, we developed a RaspberryPi-based fire recognition system using Faster-recurrent convolutional neural network (F-RCNN) and single shot multibox detector (SSD) and demonstrated a fire alarm system that works with power line communication. Image recognition was performed with a pie camera of RaspberryPi, and the detected fire image was transmitted to a monitoring PC through an inductive power line communication network. The frame rate per second (fps) for each learning model was 0.05 fps for Faster-RCNN and 1.4 fps for SSD. SSD was 28 times faster than F-RCNN.

• Journal of the Korean Dietetic Association
• /
• v.27 no.1
• /
• pp.15-25
• /
• 2021

This study sought to analyze the need for nutritional education to enhance the selection of desirable foods by adolescents. A total of 480 high school students in Incheon were surveyed and their responses were analyzed for the recognition of nutrition labeling and their consumption of processed foods and snacks. Almost all the students (93.1%) recognized nutrition labeling, but 54.6% rarely checked the nutrition labeling. The nutrients recognized as important in the nutrition label of processed foods were total fat/saturated fat/trans fat (3.75 out of 5), calories (3.68), and sodium (3.67) in that order. The recognition of the importance of calories was significantly higher in female students compared to male students (3.78 vs. 3.58, P<0.05). The information identified as important were the date of manufacture and expiration (4.21 out of 5), price (4.14), and the nutrition label (3.15). The preference for processed foods was highest in the order of beverages (4.03 out of 5), noodles (4.02), and frozen desserts (3.97), and the preference of females for processed foods was significantly higher than males (P<0.001). The time when snacks were most frequently consumed were before going to the academy (21.0%), after school (19.4%), and after attending private institutes (15.0%). The main reasons for using processed foods as snacks were taste (44.4%), 'no time to eat a meal' (26.4%), and low price (17.7%). In conclusion, while students mostly recognized nutrition labeling, the actual usage rate was low, and snacks were selected and consumed instead of meals for reasons such as attending academic institutes after school. Thus, to encourage the choice of proper processed foods and snacks for adolescents, practical nutrition education using nutrition labeling is needed.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.1
• /
• pp.93-101
• /
• 2023

In this paper, we propose a method of creating a 3D bounding box for an object using a vanishing point to increase the accuracy of object recognition in an image when recognizing an traffic object using a video camera. Recently, when vehicles captured by a traffic video camera is to be detected using artificial intelligence, this 3D bounding box generation algorithm is applied. The vertical vanishing point (VP1) and horizontal vanishing point (VP2) are derived by analyzing the camera installation angle and the direction of the image captured by the camera, and based on this, the moving object in the video subject to analysis is specified. If this algorithm is applied, it is easy to detect object information such as the location, type, and size of the detected object, and when applied to a moving type such as a car, it is tracked to determine the location, coordinates, movement speed, and direction of each object by tracking it. Able to know. As a result of application to actual roads, tracking improved by 10%, in particular, the recognition rate and tracking of shaded areas (extremely small vehicle parts hidden by large cars) improved by 100%, and traffic data analysis accuracy was improved.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.19 no.1
• /
• pp.165-172
• /
• 2024

In this paper, we designed a smart home system that can be used intuitively and easily in everyday life, such as sending text messages to users, providing various information and scheduling using smart AI, and providing lighting and atmosphere suitable for the atmosphere in situations such as listening to music using neopixels, as well as using ESP32, RFID, and Google Cloude Console using raspberry pie. As a result of the experiment, it was confirmed that security characters were normally sent to users when RFID was recognized on ESP32 connected to Wi-Fi even if the power was reconnected, and smart AI using Neopixel lighting, Raspberry Pie, and voice recognition, which calculated frequency, also changed the recognition rate over distance, but it worked.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.39C no.10
• /
• pp.909-919
• /
• 2014

License plate (LP) detection is the most imperative part of an automatic LP recognition (LPR) system. Typical LPR contains two steps, namely LP detection (LPD) and character recognition. In this paper, we propose an efficient Vehicle-to-LP detection framework which combines with an adaptive GMM (Gaussian Mixture Model) and a cascade of boosted classifiers to make a faster vehicle LP detector. To develop a background model by using a GMM is possible in the circumstance of a fixed camera and extracts the motions using background subtraction. Firstly, an adaptive GMM is used to find the region of interest (ROI) on which motion detectors are running to detect the vehicle area as blobs ROIs. Secondly, a cascade of boosted classifiers is executed on the blobs ROIs to detect a LP. The experimental results on our test video with the resolution of $720{\times}576$ show that the LPD rate of the proposed system is 99.14% and the average computational time is approximately 42ms.

• Journal of Computing Science and Engineering
• /
• v.2 no.1
• /
• pp.1-25
• /
• 2008

Spoken dialog tasks incur many errors including speech recognition errors, understanding errors, and even dialog management errors. These errors create a big gap between the user's intention and the system's understanding, which eventually results in a misinterpretation. To fill in the gap, people in human-to-human dialogs try to clarify the major causes of the misunderstanding to selectively correct them. This paper presents a method of clarification techniques to human-to-machine spoken dialog systems. We viewed the clarification dialog as a two-step problem-Belief confirmation and Clarification strategy establishment. To confirm the belief, we organized the clarification process into three systematic phases. In the belief confirmation phase, we consider the overall dialog system's processes including speech recognition, language understanding and semantic slot and value pairs for clarification dialog management. A clarification expert is developed for establishing clarification dialog strategy. In addition, we proposed a new design of plugging clarification dialog module in a given expert based dialog system. The experiment results demonstrate that the error verifiers effectively catch the word and utterance-level semantic errors and the clarification experts actually increase the dialog success rate and the dialog efficiency.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.10
• /
• pp.1991-2000
• /
• 1994

In this paper, an alagorithm with six types of classification is suggested for the recognition system of hand-written Korean characters. After thinning process and truncating process for noise redection. The input images are used generalized by $64\times64$ size. The six type classification is composed of preliminary and secondary classification process by using the learning algoritm of multi-layer perceptron. Subblock Hough transform is used as local feature and sampling Hough transform is used as global feature. Experiment is conducted for 1800 characters which is written 31 times per each type by 10 persons. The 90% recognition rate is resulted by the preliminary classification of detection the final consonant and by the secondary classification of detecting the vowels.

• Journal of Institute of Control, Robotics and Systems
• /
• v.19 no.2
• /
• pp.119-124
• /
• 2013

Recent increasing demand on the indoor localization requires more advanced and hybrid technology. This paper proposes an application of the hybrid indoor localization method based on a position-coded pattern that can be used with other existing indoor localization techniques such as vision, beacon, or landmark technique. To reduce the pattern-recognition error rate, the error detection and correction algorithm was applied based on Hamming code. The indoor localization experiments based on the proposed algorithm were performed by using a QCIF-grade CMOS sensor and a position-coded pattern with an area of $1.7{\times}1.7mm^2$. The experiments have shown that the position recognition error ratio was less than 0.9 % with 0.4 mm localization accuracy. The results suggest that the proposed method could be feasibly applied for the localization of the indoor mobile service robots.