DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Kim, Dae-Shik | - |
dc.contributor.advisor | 김대식 | - |
dc.contributor.author | Jo, KangUn | - |
dc.date.accessioned | 2019-09-04T02:40:13Z | - |
dc.date.available | 2019-09-04T02:40:13Z | - |
dc.date.issued | 2018 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=734054&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/266710 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학부, 2018.2,[iv, 32 p. :] | - |
dc.description.abstract | Visual object tracking algorithm is an important problem for many applications like surveillance system, robot vision system, and augmented reality system. Particularly, in the object tracking algorithm, it is important that the algorithm is robust to occlusions, motion blurring, and object deformation. The trackers using CNN is robust to object deformations, but it is weak to occlusion and motion blurring. However, trackers using LSTM is robust to occlusion and motion blurring, but relatively weak to object deformations. In this work, we developed an multimodal object tracking algorithm that can track an object stably by using CNN and Deep LSTM, which can learn more complex features than single LSTM. We used OTB-30 dataset to evaluate our tracker, and verified that our tracker has the highest performance than the other trackers. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | Object tracking▼aLong short-term memory▼aConvolutional neural network | - |
dc.subject | 물체 추적▼a롱 쇼트-텀 메모리▼a콘볼루셔널 뉴럴 네트워크 | - |
dc.title | (A) real-time object tracker using CNN and deep LSTM | - |
dc.title.alternative | CNN과 Deep LSTM을 이용한 실시간 물체 추적 알고리즘 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전기및전자공학부, | - |
dc.contributor.alternativeauthor | 조강운 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.