• Title/Summary/Keyword: search region

Search Result 646, Processing Time 0.025 seconds

Multi-level Cross-attention Siamese Network For Visual Object Tracking

  • Zhang, Jianwei;Wang, Jingchao;Zhang, Huanlong;Miao, Mengen;Cai, Zengyu;Chen, Fuguo
    • KSII Transactions on Internet and Information Systems (TIIS)
    • /
    • v.16 no.12
    • /
    • pp.3976-3990
    • /
    • 2022
  • Currently, cross-attention is widely used in Siamese trackers to replace traditional correlation operations for feature fusion between template and search region. The former can establish a similar relationship between the target and the search region better than the latter for robust visual object tracking. But existing trackers using cross-attention only focus on rich semantic information of high-level features, while ignoring the appearance information contained in low-level features, which makes trackers vulnerable to interference from similar objects. In this paper, we propose a Multi-level Cross-attention Siamese network(MCSiam) to aggregate the semantic information and appearance information at the same time. Specifically, a multi-level cross-attention module is designed to fuse the multi-layer features extracted from the backbone, which integrate different levels of the template and search region features, so that the rich appearance information and semantic information can be used to carry out the tracking task simultaneously. In addition, before cross-attention, a target-aware module is introduced to enhance the target feature and alleviate interference, which makes the multi-level cross-attention module more efficient to fuse the information of the target and the search region. We test the MCSiam on four tracking benchmarks and the result show that the proposed tracker achieves comparable performance to the state-of-the-art trackers.

An Index Structure based on Space Partitions and Adaptive Bit Allocations for Multi-Dimensional Data (다차원 데이타를 위한 공간 분할 및 적응적 비트 할당 기반 색인 구조)

  • Bok, Kyoung-Soo;Kim, Eun-Jae;Yoo, Jae-Soo
    • Journal of KIISE:Databases
    • /
    • v.32 no.5
    • /
    • pp.509-525
    • /
    • 2005
  • In this paper, we propose the index structure based on a vector approximation for efficiently supporting the similarity search of multi-dimensional data. The proposed index structure splits a region with the space partition method and allocates to the split region dynamic bits according to the distribution of data. Therefore, the index structure splits a region to the unoverlapped regions and can reduce the depth of the tree by storing the much region information of child nodes in a internal node. Our index structure represents the child node more exactly and provide the efficient search by representing the region information of the child node relatively using the region information of the parent node. We show that our proposed index structure is better than the existing index structure in various experiments. Experimental results show that our proposed index structure achieves about $40\%$ performance improvements on search performance over the existing method.

Fast Block Matching Algorithm Using The Distribution of Mean Absolute Difference at The Search Region Overlapped with Neighbor Blocks and Subsampling (이웃 블록과 중첩된 탐색영역에서의 MAD 분포 및 부표본화를 이용한 고속 블록 정합)

  • 이법기;정원식;이경환;최정현;김경규;김덕규
    • The Journal of Korean Institute of Communications and Information Sciences
    • /
    • v.24 no.8B
    • /
    • pp.1506-1517
    • /
    • 1999
  • In this paper, we propose two fast block matching algorithm using the distribution of mean absolute difference (MAD) at the search region overlapped with neighbor blocks and pixel subsapmling. The proposed methods use the lower and upper bound of MAD at the overlapped search region which is calculated from the MAD of neighbor block at that search position and MAD between the current block and neighbor block. In the first algorithm, we can reduce the computational complexity by executing the block matching operation at the only necessary search points. That points are selected using the lower bound of MAD. In the second algorithm, we use the statictical distribution of actual MAD which exists between the lower bound and upper bound of MAD. By using the statistical distribution of actual MAD, we can significantly reduce the computational complexity for motion estimation. after striking space key 2 times.

  • PDF

Stochastic Optimization Approach for Parallel Expansion of the Existing Water Distribution Systems (추계학적 최적화방법에 의한 기존관수로시스템의 병열관로 확장)

  • Ahn, Tae-Jin;Choi, Gye-Woon;Park, Jung-Eung
    • Water for future
    • /
    • v.28 no.2
    • /
    • pp.169-180
    • /
    • 1995
  • The cost of a looped pipe network is affected by a set of loop flows. The mathematical model for optimizing the looped pipe network is expressed in the optimal set of loop flows to apply to a stochastic optimization method. Because the feasible region of the looped pipe network problem is nonconvex with multiple local optima, the Modified Stochastic Probing Method is suggested to efficiently search the feasible region. The method consists of two phase: i) a global search phase(the stochastic probing method) and ii) a local search phase(the nearest neighbor method). While the global search sequentially improves a local minimum, the local search escapes out of a local minimum trapped in the global search phase and also refines a final solution. In order to test the method, a standard test problem from the literature is considered for the optimal design of the paralled expansion of an existing network. The optimal solutions thus found have significantly smaller costs than the ones reported previously by other researchers.

  • PDF

Adaptive Learning Control of Electro-Hydraulic Servo System Using Real-Time Evolving Neural Network Algorithm (실시간 진화 신경망 알고리즘을 이용한 전기.유압 서보 시스템의 적응 학습제어)

  • Jang, Seong-Uk;Lee, Jin-Geol
    • Journal of Institute of Control, Robotics and Systems
    • /
    • v.8 no.7
    • /
    • pp.584-588
    • /
    • 2002
  • The real-time characteristic of the adaptive leaning control algorithms is validated based on the applied results of the hydraulic servo system that has very strong a non-linearity. The evolutionary strategy automatically adjusts the search regions with natural competition among many individuals. The error that is generated from the dynamic system is applied to the mutation equation. Competitive individuals are reduced with automatic adjustments of the search region in accordance with the error. In this paper, the individual parents and offspring can be reduced in order to apply evolutionary algorithms in real-time. The feasibility of the newly proposed algorithm was demonstrated through the real-time test.

Evolutionary Computation for the Real-Time Adaptive Learning Control(II) (실시간 적응 학습 제어를 위한 진화연산(II))

  • Chang, Sung-Ouk;Lee, Jin-Kul
    • Proceedings of the KSME Conference
    • /
    • 2001.06b
    • /
    • pp.730-734
    • /
    • 2001
  • In this study in order to confirm the algorithms that are suggested from paper (I) as the experimental result, as the applied results of the hydraulic servo system are very strong a non-linearity of the fluid in the computer simulation, the real-time adaptive learning control algorithms is validated. The evolutionary strategy has characteristics that are automatically. adjusted in search regions with natural competition among many individuals. The error that is generated from the dynamic system is applied to the mutation equation. Competitive individuals are reduced with automatic adjustments of the search region in accord with the error. In this paper, the individual parents and offspring can be reduced in order to apply evolutionary algorithms in real-time as the description of the paper (I). The possibility of a new approaching algorithm that is suggested from the computer simulation of the paper (I) would be proved as the verification of a real-time test and the consideration its influence from the actual experiment.

  • PDF

Customized Web Search Rank Provision (개인화된 웹 검색 순위 생성)

  • Kang, Youngki;Bae, Joonsoo
    • Journal of Korean Institute of Industrial Engineers
    • /
    • v.39 no.2
    • /
    • pp.119-128
    • /
    • 2013
  • Most internet users utilize internet portal search engines, such as Naver, Daum and Google nowadays. But since the results of internet portal search engines are based on universal criteria (e.g. search frequency by region or country), they do not consider personal interests. Namely, current search engines do not provide exact search results for homonym or polysemy because they try to serve universal users. In order to solve this problem, this research determines keyword importance and weight value for each individual search characteristics by collecting and analyzing customized keyword at external database. The customized keyword weight values are integrated with search engine results (e.g. PageRank), and the search ranks are rearranged. Using 50 web pages of Goolge search results for experiment and 6 web pages for customized keyword collection, the new customized search results are proved to be 90% match. Our personalization approach is not the way that users enter preference directly, but the way that system automatically collects and analyzes personal information and then reflects them for customized search results.

Intelligent information filtering using rough sets

  • Ratanapakdee, Tithiwat;Pinngern, Ouen
    • 제어로봇시스템학회:학술대회논문집
    • /
    • 2004.08a
    • /
    • pp.1302-1306
    • /
    • 2004
  • This paper proposes a model for information filtering (IF) on the Web. The user information need is described into two levels in this model: profiles on category level, and Boolean queries on document level. To efficiently estimate the relevance between the user information need and documents by fuzzy, the user information need is treated as a rough set on the space of documents. The rough set decision theory is used to classify the new documents according to the user information need. In return for this, the new documents are divided into three parts: positive region, boundary region, and negative region. We modified user profile by the user's relevance feedback and discerning words in the documents. In experimental we compared the results of three methods, firstly is to search documents that are not passed the filtering system. Second, search documents that passed the filtering system. Lastly, search documents after modified user profile. The result from using these techniques can obtain higher precision.

  • PDF

Optimization of Triple Response Systems by Using the Dual Response Approach and the Hooke-Jeeves Search Method

  • Fan, Shu-Kai S.;Huang, Chia-Fen;Chang, Ko-Wei;Chuang, Yu-Chiang
    • Industrial Engineering and Management Systems
    • /
    • v.9 no.1
    • /
    • pp.10-19
    • /
    • 2010
  • This paper presents an extended computing procedure for the global optimization of the triple response system (TRS) where the response functions are nonconvex (nonconcave) quadratics and the input factors satisfy a radial region of interest. The TRS arising from response surface modeling can be approximated using a nonlinear mathematical program involving one primary (objective) function and two secondary (constraints) functions. An optimization algorithm named triple response surface algorithm (TRSALG) is proposed to determine the global optimum for the nondegenerate TRS. In TRSALG, the Lagrange multipliers of target (secondary) functions are computed by using the Hooke-Jeeves search method, and the Lagrange multiplier of the radial constraint is located by using the trust region (TR) method at the same time. To ensure global optimality that can be attained by TRSALG, included is the means for detecting the degenerate case. In the field of numerical optimization, as the family of TR approach always exhibits excellent mathematical properties during optimization steps, thus the proposed algorithm can guarantee the global optimal solution where the optimality conditions are satisfied for the nondegenerate TRS. The computing procedure is illustrated in terms of examples found in the quality literature where the comparison results with a gradient-based method are used to calibrate TRSALG.

Data Acquisition System Using the Second Binary Code (2차원 부호를 이용한 정보 획득 시스템)

  • Kim, In-Kyeom
    • The Journal of Information Technology
    • /
    • v.6 no.1
    • /
    • pp.71-84
    • /
    • 2003
  • In this paper, it is presented the efficient system for data recognition using the proposed binary code images. The proposed algorithm finds the position of binary image. Through the process of the block region classification, it is classified each block with the edge region using the value of gray level only. Each block region is divided horizontal and vertical edge region. If horizontal edge region blocks are classified over six blocks in any region, the proposed algorithm should search the vertical edge region in the start point of the horizontal edge region. If vertical edge region blocks were found over ten blocks in vertical region, the code image would found. Practical code region is acquired from the rate of the total edge region that is computed from the binary image that is processed with the average value. In case of the wrong rate, it is restarted the code search in the point after start point and the total process is followed. It has a short time than the before process time because it had classified block information. The block processing is faster thant the total process. The proposed system acquires the image from the digital camera and makes binary image from the acquired image. Finally, the proposed system extracts various characters from the binary image.

  • PDF