본문 바로가기
HOME> 논문 > 논문 검색상세

학위논문 상세정보

각 가중치 파티클 필터를 이용한 객체추적에 관한 연구 : A STUDY ON OBJECT TRACKING WITH ANGULAR WEIGHTED PARTICLE FILTER 원문보기

  • 저자

    무스타파

  • 학위수여기관

    경성대학교 일반대학원

  • 학위구분

    국내박사

  • 학과

    전기전자공학과 전기전자공학

  • 지도교수

    윤병우

  • 발행년도

    2014

  • 총페이지

    vi, 68 p.

  • 키워드

    pattern recognition object tracking particle filter;

  • 언어

    eng

  • 원문 URL

    http://www.riss.kr/link?id=T13535209&outLink=K  

  • 초록

    Intelligent image processing technology is being applied to various areas as security surveillance, medical engineering, and industrial automation etc. Especially, there are various studies about it to prevent traffic accidents and control traffic flow effectively, which is named as Intelligent Transportation Systems(ITS). ITS based on image processing involves the functions such as detection of reverse driving, fallen objects, and pedestrians on the road for the purpose of the analysis of traffic load and early detection of traffic accidents. These perception of situation depends on the performance of object detection and tracking. Usually, Adaboost algorithm and SVM algorithm are used for object detection, and optical flow, Kaman filter, and particle filter are used for object tracking. Particle filter can track the objects well in non-Gaussian medium, so it can adapt well for the randomness in the real world. However, it requires too large amounts of computations. In this dissertation, a modified particle filter algorithm is proposed to improve the performance of object tracking and decrease the computational complexity. It uses the information of color histogram for defining the target object, and uses the direction of car driving to change the statistical characteristic of each particle for reducing the number of particles. In the first stage of the proposed algorithm, the motion angle of the target is calculated. In the second stage, the angle between the motion angle of the car and each particle is calculated for all particles. In the last stage, the probability of each particle is weighted according to the calculated angle difference. In this way, the particles moving similar direction with the target get higher weights, and the state parameters of them contribute more to tracking the target. Because of this, the proposed algorithm becomes more stable and robust against noises. The experimental results for many videos from various roads, showed that proposed algorithm provides better tracking performance than the conventional algorithm. It can track the targets more accurately than the conventional algorithm, and the computation is reduced because of the decrement of particles number. We expect that the proposed particle filter can contribute to the improvement of the ITS based on image.


 활용도 분석

  • 상세보기

    amChart 영역
  • 원문보기

    amChart 영역