解读DP-SLAM （5）

解读DP-SLAM （5）

上一次谈到Naive approach的缺点，在这篇论文里，作者提出一个DP-Mapping的概念，以及两种数据结构

the ancestry tree和the map，并且使粒子滤波的计算复杂度与迭代次数无关

今天来分析一下这些数据结构，先来看the ancestry tree:

1, The tree itself is rooted with an initial particle, of which all other particles are progeny

2, Each particle maintains a pointer to its parent and is assigned a unique numerical ID

3, Each particle maintains a list of grid squares that it has updated

Ancestry tree的特点：

The tree has bounded size regardless of the number of iterations of the particle filter

维护Ancestry tree过程 （maintain the ancestry tree）:

概括描述：

We maintain a bounded size tree by pruning away unnecessary nodes

具体措施：

1, note that certain particles may not have children and can simply be removed from the tree

2, we can recursively prune away dead branches of the tree

剪枝的结果：

the only particle which are stored in our ancestry tree are exactly those particles which are ancestors of the

current generation of particles

PS：从句有一点复杂,慢慢理解

可想而知，这样的剪枝会产生一个结果：

If a particle has only one  child in our ancestry tree, we can essentially remove it, by collapsing that branch of

the tree

剪枝操作可以得到minimal ancestry tree，它有一个良好的性质：

Independent of the number of iterations of particle filtering, a minimal ancestry tree of P particles

1. has exactly P leaves

2. has branching factor of at least 2, and

3. has depth no more than P

ancestry tree的数据结构就介绍到这里，至于它怎样运用在SLAM过程，后面章节会介绍

我这里画一个草图，

哈哈，下次来分析Map representation

这是官方ppt的讲解图：