• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.9
• /
• pp.1330-1339
• /
• 2022

3D object detection generally aims to detect relatively large data such as automobiles, buses, persons, furniture, etc, so it is vulnerable to small object detection. In addition, in an environment with limited resources such as embedded devices, it is difficult to apply the model because of the huge amount of computation. In this paper, the accuracy of small object detection was improved by focusing on local features using only one layer, and the inference speed was improved through the proposed knowledge distillation method from large pre-trained network to small network and adaptive quantization method according to the parameter size. The proposed model was evaluated using SUN RGB-D Val and self-made apple tree data set. Finally, it achieved the accuracy performance of 62.04% at mAP@0.25 and 47.1% at mAP@0.5, and the inference speed was 120.5 scenes per sec, showing a fast real-time processing speed.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10a
• /
• pp.137-140
• /
• 1996

Finding line segments in an intensity image has been one of the most fundamental issues in computer vision. In complex scenes, it is hard to detect the locations of point features. Line features are more robust in providing greater positional accuracy. In this paper we present a robust "line features extraction" algorithm which extracts line feature in a single pass without using any assumptions and constraints. Our algorithm consists of five steps: (1) edge scanning, (2) edge normalization, (3) line-blob extraction, (4) line-feature computation, and (5) line linking. By using edge scanning, the computational complexity due to too many edge pixels is drastically reduced. Edge normalization improves the local quantization error induced from the gradient space partitioning and minimizes perturbations on edge orientation. We also analyze the effects of edge processing, and the least squares-based method and the principal axis-based method on the computation of line orientation. We show its efficiency with some real images.al images.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.744-747
• /
• 1997

High performance digital computer technology enables the digital computer-based controllers to replace traditional analog controllers used for factory automations. This replacement, however, brings up the side effects caused by discrete quantization and non-real-time execution of control softwares. This paper describes the structure of real-time simulator and controller that can be used for design and verification of real-time digital controllers. The virtual machine concept adopted by real-time simulator make the proposed simulator be independent from the specific hardware platforms. The proposed system can also be used in the loosely coupled distributed environments connected through local area network using real-time message passing algorithm and virtual data table based on the shared memory mechanism.

• Journal of Institute of Control, Robotics and Systems
• /
• v.5 no.1
• /
• pp.115-121
• /
• 1999

Advanced digital computer technology enables the computer-based controllers to replace the traditional analog controllers used in factory automations. This replacement, however, brings up the side effects caused by the quantization error and non-real-time execution of control software. This paper describes the structure of real-time simulator and controller that can be used for design and verification of real-time digital controllers. The virtual machine concept adopted by the proposed real-time simulator makes the proposed simulator be independent from the specific hardware platforms. The proposed system can also be used in the loosely coupled distributed environments connected through local area network using real-time message passing algorithm and virtual data table based on the shared memory mechanism.

• Speech Sciences
• /
• v.12 no.2
• /
• pp.171-182
• /
• 2005

An efficient GMM (Gaussian mixture modeling) method based on LPCA (local principal component analysis) with VQ (vector quantization) for speaker identification is proposed. To reduce the dimension and correlation of the feature vector, this paper proposes a speaker identification method based on principal component analysis. The proposed method firstly partitions the data space into several disjoint regions by VQ, and then performs PCA in each region. Finally, the GMM for the speaker is obtained from the transformed feature vectors in each region. Compared to the conventional GMM method with diagonal covariance matrix, the proposed method requires less storage and complexity while maintaining the same performance requires less storage and shows faster results.

• Journal of Computing Science and Engineering
• /
• v.9 no.4
• /
• pp.204-213
• /
• 2015

Under the conditional independence assumption among local features, the Naive Bayes Nearest Neighbor (NBNN) classifier has been recently proposed and performs classification without any training or quantization phases. While the original NBNN shows high classification accuracy without adopting an explicit training phase, the conditional independence among local features is against the compositionality of objects indicating that different, but related parts of an object appear together. As a result, the assumption of the conditional independence weakens the accuracy of classification techniques based on NBNN. In this work, we look into this issue, and propose a novel Bayesian network for an NBNN based classification to consider the conditional dependence among features. To achieve our goal, we extract a high-level feature and its corresponding, multiple low-level features for each image patch. We then represent them based on a simple, two-level layered Bayesian network, and design its classification function considering our Bayesian network. To achieve low memory requirement and fast query-time performance, we further optimize our representation and classification function, named relation-based Bayesian network, by considering and representing the relationship between a high-level feature and its low-level features into a compact relation vector, whose dimensionality is the same as the number of low-level features, e.g., four elements in our tests. We have demonstrated the benefits of our method over the original NBNN and its recent improvement, and local NBNN in two different benchmarks. Our method shows improved accuracy, up to 27% against the tested methods. This high accuracy is mainly due to consideration of the conditional dependences between high-level and its corresponding low-level features.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.38 no.6
• /
• pp.665-674
• /
• 2001

In this paper, a reduced complexity fillet to simultaneously suppress the blocking and ringing artifacts of compressed video sequence is addressed. A new one dimensional regularized function to incorporate the smoothness to its neighboring pixels into the solution is defined, resulting in very low complexity filter The proposed regularization function consists of two sub-functions that combine local data fidelity and local smoothing constraints. The regularization parameters to control the trade-off between the local fidelity to the data and the smoothness are determined by available overhead information in decoder, such as maroc-block type and quantization step size. In addition, the regularization parameters are designed to have the limited range and stored as look-up-table, and therefore, the computational cost to determine the parameters can be reduced. The experimental results show the capability and efficiency of the proposed algorithm.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.3
• /
• pp.55-63
• /
• 2008

This paper presents a panorama background generation and object tracking technique using a Pan-Tilt-Zoom camera. The proposed method estimates local motion vectors rapidly using phase correlation matching at the prespecified multiple local regions, and it makes minimized estimation error by vector quantization. We obtain the required image patches, by estimating the overlapped region using local motion vectors, we can then project the images to cylinder and realign the images to make the panoramic image. The object tracking is performed by extracting object's motion and by separating foreground from input image using background subtraction. The proposed PTZ-based object tracking method can efficiently generated a stable panorama background, which covers up to 360 degree FOV The proposed algorithm is designed for real-time implementation and it can be applied to many commercial applications such as object shape detection and face recognition in various surveillance video systems.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.33 no.1
• /
• pp.363-375
• /
• 2013

Today, a shortage of the skilled operator has been intensified gradually and the necessity of an earthwork in extreme environment operators are difficult to access is increasing for the purpose of resource development and new living space creation. For this reason, an effort to develop an unmanned excavation robot for fully automated earthwork system is continuing globally. In Korea, a research consortium called 'Intelligent Excavation System' has been formed since 2006 as a part of Construction Technology Innovation Program of Ministry of Land, Transport and Maritime Affairs of Korea. Among detailed technologies of the Task Planning System is one of the core technologies of IES, this paper explains research and development process of the Local Area Design Module, which provides informatization unit to create automated excavators' work command information at operation level such as location, range, target, and sequence for excavation work. Designing of Local Area should be considered various influential factors such as excavator's specification, working mechanism, heuristics, and structural stability to create work plan guaranteed safety and effectiveness. For this research, conceptual and detail design of the Local Area is performed for analyzing design element and variable, and quantization method of design specification corresponding with heuristics and structural safety is generated. Finally, module is developed through constructed algorithm and developed module is verified.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.37 no.5
• /
• pp.54-63
• /
• 2000

We present a multi-resolution block matching algorithm (BMA) for fast motion estimation At the coarsest level, a motion vector (MV) having minimum matching error is chosen via a full search, and a MV with minimum matching error is concurrently found among the MVs of the spatially adjacent blocks Here, to examine the spatial MVs accurately, we propose an efficient method for searching full resolution MV s without MV quantization even at the coarsest level The chosen two MV s are used as the initial search centers at the middle level At the middle level, the local search is performed within much smaller search area around each search center If the method used at the coarsest level is adopted here, the local searches can be done at integer-pel accuracy A MV having minimum matching error is selected within the local search areas, and then the final level search is performed around this initial search center Since the local searches are performed at integer-pel accuracy at the middle level, the local search at the finest level does not take an effect on the overall performance So we can skip the final level search without performance degradation, thereby the search speed increases Simulation results show that in comparison with full search BMA, the proposed BMA without the final level search achieves a speed-up factor over 200 with minor PSNR degradation of 02dB at most, under a normal MPEG2 coding environment Furthermore, our scheme IS also suitable for hardware implementation due to regular data-flow.