摘要
在贝叶斯框架下,基于增量正交投影非负矩阵分解目标跟踪算法能够适应各种复杂的场景,准确处理跟踪目标外观变化,取得了较好的跟踪效果,但是该算法计算量大,难以满足实时性要求。针对这一缺点,提出了一种基于L2范数和增量正交投影非负矩阵分解的目标跟踪算法,建立基于L2范数最小化和增量正交投影非负矩阵分解的目标表示模型,在贝叶斯框架下得出跟踪结果。实验结果表明,新算法能够较好地处理视频场景中的光照变化、尺度变化、局部遮挡、角度变化等干扰,有较低的中心位置误差平均值和较高的重叠率平均值,平均处理视频达4.08帧·s-1,能够满足实时性的要求。
Under the framework of the Bayesian estimation, the tracking algorithm based on incremental orthogonal changes obI projective non-negative matrix factorization can deal with complex scenes and the appearance ,ject in the video scene and can get better results. However, the computation cost is too large to meet real-time tracking. To solve the problem, object tracking algorithm combining the L2 norm and in- cremental orthogonal projective non-negative matrix factorization is proposed under the framework of the Bayesian estimation. Experimental results show that the proposed method can deal with illumination chan- ges, scale changes , handling occlusion and angle changes with lower average center location and higher av- erage overlap rates. Additionally, the proposed method can work well at a speed about 4. 08 frame · s-1 and can achieve real-time tracking.
出处
《黑龙江大学自然科学学报》
CAS
北大核心
2015年第2期262-269,共8页
Journal of Natural Science of Heilongjiang University
基金
山东省自然科学基金高校
科研单位联合专项计划资助项目(ZR2014FL020)
滨州市科技发展计划项目(2013ZC0103)