• ETRI Journal
• /
• v.36 no.5
• /
• pp.714-720
• /
• 2014

In this paper, we propose the use of data-driven probabilistic utterance-level decision logic to improve Weighted Finite State Transducer (WFST)-based endpoint detection. In general, endpoint detection is dealt with using two cascaded decision processes. The first process is frame-level speech/non-speech classification based on statistical hypothesis testing, and the second process is a heuristic-knowledge-based utterance-level speech boundary decision. To handle these two processes within a unified framework, we propose a WFST-based approach. However, a WFST-based approach has the same limitations as conventional approaches in that the utterance-level decision is based on heuristic knowledge and the decision parameters are tuned sequentially. Therefore, to obtain decision knowledge from a speech corpus and optimize the parameters at the same time, we propose the use of data-driven probabilistic utterance-level decision logic. The proposed method reduces the average detection failure rate by about 14% for various noisy-speech corpora collected for an endpoint detection evaluation.

• International Journal of Aeronautical and Space Sciences
• /
• v.9 no.2
• /
• pp.114-120
• /
• 2008

In the vision-based recovery phase, a terminal guidance for the blended-wing UAV requires visual information of high accuracy. This paper presents the light source target design and detection algorithm for vision-based UAV recovery. We propose a recovery target design with red and green LEDs. This frame provides the relative position between the target and the UAV. The target detection algorithm includes HSV-based segmentation, morphology, and blob processing. These techniques are employed to give efficient detection results in day and night net recovery operations. The performance of the proposed target design and detection algorithm are evaluated through ground-based experiments.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.3 s.309
• /
• pp.17-24
• /
• 2006

In this paper, an algorithm for anchor frame detection in news video is proposed, which consists of four steps. In the first step, the cumulative histogram method is used to detect shot boundaries in order to segment a news video into video shots. In the second step, skin color information is used to detect face regions in each shot boundary. In the third step, color information of upper body regions is used to extract anchor object, which produces candidate anchor frames. Then, from the candidate anchor frames, a graph-theoretic cluster analysis algorithm is utilized to classify the news video into anchor-person frames and non-anchor frames. Experiment results have shown the effectiveness of the proposed algorithm.

• Journal of Broadcast Engineering
• /
• v.2 no.1
• /
• pp.56-64
• /
• 1997

An HDTV frame has 4~6 times more pixels than a DTV frame. In order to encode the HDTV image in real time, parallel processing architectures have been widely used in many HDTV codec developments. That is, an HDTV Image is divided into several subbands and each subband is encoded in parallel using some DTV level encoders. In this paper, we adopt an HDTV codec architecture which divides an HDTV frame into 4 subbands and propose a new scene change detection algorithm using local variance. In addition, a new adaptive multichannel rate control scheme which allocate target bits adaptively to each subband of the HDTV image based on the activities of subband images is suggested in this paper. The activities of subband images are calculated at scene change detection part and reused at the adaptive rate control part. The simulation results show that the proposed scene change detection algorithm detects the scene change of HDTV video very accurately. Also the suggested adaptive multichannel rate control scheme shows better performance than the rate control method which allocates target bits equally to each subbands of the HDTV image.

• Journal of Astronomy and Space Sciences
• /
• v.41 no.3
• /
• pp.159-170
• /
• 2024

A method for resident space object detection in video stream processing using a set of matched filters has been proposed. Matched filters are constructed based on the connection between the Fourier spectrum shape of the difference frame and the magnitude of the linear velocity projection onto the observation plane. Experimental data were obtained using the mobile optical surveillance system for low-orbit space objects. The detection problem in testing mode was solved for raw video data with intensity signals from three satellites: KORONAS-FOTON, CUSAT 2/FALCON 9, and GENESIS-1. Difference frames of video data with the AQUA satellite pass were used to construct matched filters. The satellites were automatically detected at points where the difference in the value of their linear velocity projection and the reference satellite was close in value. An initial approximation of the satellites slant range vector and position vector has been obtained based on the values of linear velocity projection onto the frame plane. It has been established that the difference in the inclination angle between the detected satellite intensity signal Fourier image and the reference satellite mask corresponds to the difference in the inclinations of these objects. The proposed method allows for detecting and estimating the initial approximation of the slant range and position vector of artificial and natural space objects, such as satellites, debris, and asteroids.

• Journal of Biosystems Engineering
• /
• v.33 no.6
• /
• pp.446-452
• /
• 2008

Machine vision system was used for analyzing leaf color disorders of tomato plants in a greenhouse. From the day when a few leave of tomato plants had started to wither, a series of images were captured by 4 times during 14 days. Among several color image spaces, Saturation frame in HSI color space was adequate to eliminate a background and Hue frame was good to detect infected disease area and tomato fruits. The processed image ($G{\sqcup}b^*$ image) by OR operation between G frame in RGB color space and $b^*$ frame in $La^*b^*$ color space was useful for image segmentation of a plant canopy area. This study calculated a ratio of the infected area to the plant canopy and manually analyzed leaf color disorders through an image segmentation for Hue frame of a tomato plant image. For automatically analyzing plant leave disease, this study selected twenty-seven color patches on the calibration bars as the corresponding to leaf color disorders. These selected color patches could represent 97% of the infected area analyzed by the manual method. Using only ten color patches among twenty-seven ones could represent over 85% of the infected area. This paper showed a proposed machine vision system may be effective for evaluating various leaf color disorders of plants growing in a greenhouse.

• ETRI Journal
• /
• v.29 no.4
• /
• pp.411-420
• /
• 2007

The steady evolution of mapping technology is leading to an increasing availability of multi-sensory geo-spatial datasets, such as data acquired by single-head frame cameras, multi-head frame cameras, line cameras, and light detection and ranging systems, at a reasonable cost. The complementary nature of the data collected by these systems makes their integration to obtain a complete description of the object space. However, such integration is only possible after accurate co-registration of the collected data to a common reference frame. The registration can be carried out reliably through a triangulation procedure which considers the characteristics of the involved data. This paper introduces algorithms for a multi-primitive and multi-sensory triangulation environment, which is geared towards taking advantage of the complementary characteristics of spatial data available from the above mentioned sensors. The triangulation procedure ensures the alignment of involved data to a common reference frame. The devised methodologies are tested and proven efficient through experiments using real multi-sensory data.

• Journal of the Optical Society of Korea
• /
• v.17 no.6
• /
• pp.513-517
• /
• 2013

Recently, maskless lithography (ML) systems have become popular in digital manufacturing technologies. To achieve high-throughput manufacturing processes, digital micromirror devices (DMD) in ML systems must be driven to their operational limits, often in harsh conditions. We propose an instrument and algorithm to detect DMD malfunctions to ensure perfect mask image transfer to the photoresist in ML systems. DMD malfunctions are caused by either bad DMD pixels or data transfer errors. We detect bad DMD pixels with $20{\times}20$ pixel by white and black image tests. To analyze data transfer errors at high frame rates, we monitor changes in the frame rate of a target DMD pixel driven by the input data with a set frame rate of up to 28000 frames per second (fps). For our data transfer error detection method, we verified that there are no data transfer errors in the test by confirming the agreement between the input frame rate and the output frame rate within the measurement accuracy of 1 fps.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.3
• /
• pp.465-472
• /
• 2020

Recently, various deep neural networks based methods have been proposed for audio event detection. In this study, we improved the performance of audio event detection by adopting an attention approach to a baseline CRNN. We applied context gating at the input of the baseline CRNN and added an attention layer at the output. We improved the performance of the attention based CRNN by using the audio data of strong labels in frame units as well as the data of weak labels in clip levels. In the audio event detection experiments using the audio data from the Task 4 of the DCASE 2018/2019 Challenge, we could obtain maximally a 66% relative increase in the F-score in the proposed attention based CRNN compared with the baseline CRNN.

• The KIPS Transactions:PartB
• /
• v.12B no.5 s.101
• /
• pp.617-630
• /
• 2005

General scene change detection determines the changes of a scene by using feature comparison of two continuous images that are above the fixed threshold. But existing algerian detects scene change that was used in comparing the features of two images continuously, it usually takes a lot of time in decrypting the image data and false-detection problem occurs when there is an object motion or a change of illumination. In this paper, macroblock were used to extract the information directly from the MPEG compression area and suggests algorithm that will detect scene changes more effectively. Existing algorithm have shown numerous arithmetic problems that were improved in the proposed algorithm. The existing algorithm cannot detect the changes of a scene after analyzing the relationship of the previousand futureimages while the algorithm being proposed can detect the changes of a scene continuously and resolves the problem of false-detection. To this end, the data used in general were tested to prove that this algerian would be able to detect the scene changes faster and more correctly than the existing ones. The performance of the suggested algorithm was analyzed basedontheresultsoftheexperiment. .