CAJ | 학술논문

针对传统粒子滤波跟踪算法重采样时存在粒子退化、目标与背景颜色相似和尺度变化导致的目标定位不准确问题,本研究提出了一种基于特征融合的粒子群优化粒子滤波跟踪算法,算法利用粒子群优化进行粒子权值更新,用当前状态估计值与各粒子状态的差值大小作为评价标准,促使粒子采样向真实状态区域移动,减缓粒子退化,提高了粒子滤波跟踪算法的跟踪性能。针对跟踪目标尺度变化导致的定位不准确情况,引入了归一化转动惯量( Normalized moment of inertia, NMI)特征,并将它与颜色特征采用乘性融合策略进行融合来描述目标特征,提高目标复杂场景下的定位精度。通过在多个标准测试视频上进行试验,实验结果表明,本研究提出的方法对动态背景场景中尺度变化目标和背景颜色相似目标的跟踪具有较好的准确性和鲁棒性。
침대전통입자려파근종산법중채양시존재입자퇴화、목표여배경안색상사화척도변화도치적목표정위불준학문제,본연구제출료일충기우특정융합적입자군우화입자려파근종산법,산법이용입자군우화진행입자권치경신,용당전상태고계치여각입자상태적차치대소작위평개표준,촉사입자채양향진실상태구역이동,감완입자퇴화,제고료입자려파근종산법적근종성능。침대근종목표척도변화도치적정위불준학정황,인입료귀일화전동관량( Normalized moment of inertia, NMI)특정,병장타여안색특정채용승성융합책략진행융합래묘술목표특정,제고목표복잡장경하적정위정도。통과재다개표준측시시빈상진행시험,실험결과표명,본연구제출적방법대동태배경장경중척도변화목표화배경안색상사목표적근종구유교호적준학성화로봉성。
For the presence of particles of degraded traditional particle filter tracking algorithm when resampling, Not a good solution to partial occlusion morph targets and target tracking problems, This paper introduces the multi-feature fusion based particle swarm optimized particle filter tracking method. Particle swarm optimization algorithm uses particle weights be updated with the current estimate of the difference between the size of the state and the state of each particle as evaluation criteria, prompting the true state of the particle sampling area to move, reduce particle degradation, improve the tracking performance of the particle filter tracking algorithm. For target deformation and occlusion, This article introduced the Normalized moment of inertia ( NMI) feature, Will it with color features multiplicative fusion strategy fusion is used to describe the target characteristics. Through the experiments on several standard test video, experimental results show that the proposed method for dynamic background scene morph targets and partial occlusion target tracking with better accuracy and robustness.