多方法融合的粒子滤波算法的神经丝自动跟踪

神经丝蛋白质是医学中研究肌萎缩侧索硬化症病情进展的标志物.为了能精确捕获某种神经丝蛋白质在神经鞘中的活动特性,引入了一种多方法融合的粒子滤波算法跟踪神经丝蛋白质的运动.该算法汲取颜色直方图法、核函数法及图模法等的优点,融合粒子滤波算法,实现自动跟踪神经丝蛋白质.此外,为了解决粒子滤波中样本贫化,即在粒子滤波计算中很大一部分粒子重叠到一个单独的点上的情况,需要重采样计算解决此问题.但在重采样过程中,容易造成一些粒子丢失各向异性而导致跟踪精度降低,甚至跟踪目标失败,故结合粒子滤波算法提出了一种改进重采样约束方法.实验结果表明,基于改进重采样法及多方法融合的粒子滤波算法较传统算法能有效地减少样本贫化问题,并且可以高精度地跟踪移动、变形的神经丝蛋白质,为医学中神经丝蛋白质研究提供了新支撑方法.

Neurofilament protein automatic tracking of the particle filter algorithm based on multiple methods fusion

The neurofilament protein serves as the marker of the state for ALS (Amyotrophic Lateral Sclerosis) in the medical filed. In order to accurately capture the motion characteristics of the neurofilament protein in the axon, a new-type algorithm based on the particle filtering of multiple methods-fusion is introduced in this paper. This fusion algorithm integrates the advantages of the color histogram, kernel function method, and graph model strength into the particle filtering algorithm. In addition, in order to solve the problem of sample impoverishment, which will lead to the majority of particles overlapping on one single point in the computation of the particle filter, the re-sampling method is utilitied. However, the re-sampling method easily causes the loss of the particle anisotropy, which will reduce the tracking precision or even fail to the track. We present a new re-sampling constrained method to improve the particle anisotropy in the particle filtering. Experimental results indicate that the algorithm based on the improved method of re-sampling and the particle filter of multiple methods-fusion can effectively reduce the number of overlapping particles and precisely track the deformed neurofilament protein. Such a tracking method will be helpful in the research on the neurofilament protein in the medical filed.