利用巴氏系数判定模型更新的视觉跟踪算法  

Visual object tracking method based on model update strategy by Bhattacharyya coefficient judging

作　　者：范舜奕 管桦[1] 侯志强[1] 余旺盛[1] 戴铂 

机构地区：[1]空军工程大学信息与导航学院,西安710077

出　　处：《计算机应用研究》2017年第3期915-919,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61175029;61473309);陕西省自然科学基金资助项目(2011JM8015)

摘　　要：在视频跟踪时,传统的粒子滤波算法在目标区域出现遮挡、光照变化等情况下通常存在鲁棒性较差的问题,因此提出一种采用巴氏(Bhattacharyya)系数判断模型更新时机的鲁棒视觉跟踪算法。算法以粒子滤波算法为框架,每隔一定帧数抽样检测目标变化,利用当前模型与候选模型之间的巴氏系数统计特征的相似性,从而判断更新时机。仅当目标姿态逐渐改变且无背景干扰时更新目标模型;在发生遮挡或光照改变较大时则不更新,保持当前模型继续跟踪。算法判断是否出现影响目标的匹配因素,从而适时采取模型更新策略。实验结果表明,本算法通过选择性更新模型,在未考虑尺度变化的情况下,能够更加有效地抑制背景干扰和避免模型漂移,在诸多复杂场景中具有一定的鲁棒性。In visual tracking,the particle filter had poor robustness in solving the problem of occlusion and illumination change. Thus,the algorithm proposed a new visual object tracking method that detected target changes within the range of sampling frames based on particle filter,and used the Bhattacharyya coefficient which was an efficient method in image statistical feature matching to determine the update time reasonably. It was only updated when the target itself changes and no background interference,except occlusion or large illumination change. A reasonable judgment was to determine whether the impact of target matching factors in the study in order to get an effective model update strategy timely. The experiment result shows the proposed method can well restrain background distraction with updating the model selectively. Meanwhile,it can effectively update the model and overcome the problem of model drifting,and the tracking algorithm is effective and robust.

关 键 词：视觉跟踪 粒子滤波 巴氏系数 不定时模型更新 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]