基于特征融合粒子滤波的神经丝自动跟踪

Automatic Tracking of Neurofilament Protein Based on Feature Fusion in Particle Filtering

神经丝蛋白质是研究轴突信息传递和神经退化疾病的标志物质.传统的神经丝蛋白质的跟踪,通常需要人工手动完成,进而分析其活动特性.这一过程不仅劳动强度高,而且存在很大的人为错误因素.对此,本文提出以特征融合的改进粒子滤波方法实现对神经丝蛋白质的自动跟踪.特征融合的过程中针对提取视频的特点和神经丝蛋白质运动的不规律性,利用空间—颜色核函数直方图法实现目标的高效率、高鲁棒性跟踪;而对于神经丝蛋白质容易变形的特点,引入重心修正粒子分布以提高粒子的有效性.在此基础上对特征融合函数引入分配权值系数.实验表明,本研究提出的方法在对神经丝蛋白质视觉跟踪具有很好的效果和鲁棒性.

Neurofilament protein is a significant substances about studying axon information transmission and neurodegenerative diseases. Traditional neurofilament protein tracking is manually completed to analyze the characteristics of its activities. This manual tracking process is labor-intensive. Thus, a particle filter feature fusion method for automatic tracking of neurofilament protein is presented in this paper. For extracting the screen features and neurofilament protein movement irregularities the feature fusion process utilizes the histogram method of space and color kernel function to achieve highly efficient and robust tracking; For the easy deformation characteristics of the neurofilament protein, the gravity correction particle distribution is introduced to improve the effectiveness of the particles. On this basis, the introduction of feature fusion function assigned weight coefficient γ. Experiments show that this tracking method of neurofilament protein has a good efficiency and robustness.