Measuring Crowd Collectiveness
CVPR2013
http://mmlab.ie.cuhk.edu.hk/projects/collectiveness/
https://github.com/metalbubble/collectiveness
本文针对人群运动的描述 Collective motions ,定义了一个描述子 collectiveness descriptor 可以用它来定量分析人群运动的一些信息。
Collectiveness 用于描述群体运动的一致性的一个定义
Collectiveness, which indicates the degree of individuals acting as a union in collective motion
Collectiveness 的定义是 表示 个体的运动和 群体运动的相似度,个体在群体运动中的参与度,
人群,鱼群,细菌群,羊群 等等都有有序的运动,怎么来定量描述这个运动了?
We refer to the spatially coherent structure of collective motion as Collective Manifold
我们将这种 空间一致性的有序运动称之为 Collective Manifold
从上图可以看出这种一致性局限于 局部领域的个体,也就是局部个体的运动保持高度一致性
behavior consistency remains high among individuals in local neighborhood
2 Measuring Collectiveness
一个 crowd 不仅仅是一群个体的简单组合。在一定的环境下,群体中的个体按照不同的水平的群体运动组成一个整体,所以群体运动的描述由个体的参与度决定,这反应了个体的运动和 群体运动的相似度。我们由从下而上的方式来介绍 collectiveness,首先从相邻的个体行为一致性,再从个体的 collectiveness 到 整体的collectiveness。
这里我们从图论的角度来定义 collectiveness
文中用到的一些符号定义
2.1. Behavior Consistency in Neighborhood
我们首先度量一个邻域 N (is defined as K-nearest-neighbor) 中个体的运动相似性.
C_t (i,j) is the velocity correlation at t between i and j
w_t (i,j) ∈ [0,1] measures an individual’s behavior consistency in a neighborhood
这种行为相似性只局限于同一个邻域中的个体,当两个个体不在一个邻域中,这种相似性基本就不可靠了
下面我们提出了一个更好的基于collective manifold 结构属性的行为一致性描述
A better behavior consistency based on the structural property of collective manifold is proposed below.
2.2. Behavior Consistency on Collective Manifold
因为当两个个体比较远是,行为相似性的估计就不那么准确了。我们提出了一个基于collective manifold 结构属性的行为一致性描述 : paths, which represent the connectivity of the network associated with a graph, In crowd systems, paths have important roles in information transmission and communication among constituent individuals. Thus, path-based similarity can better characterize the behavior consistency among individuals in a crowd.
定义矩阵 W,其中的元素 w_t (i,j) is the similarity between individual i and j in its neighborhood defined in Eq.1
接着定义了一个 path 和 path similarity ν_l (i,j)
Theorem 1. ν_l (i,j) is the (i,j) entry of matrix W^l, (l个 W 相乘)
2.3. Individual Collectiveness from Path Similarity
所谓条条大路通罗马,节点 i,j 直接可以有很多不同的 l path
下图显示了 l = 1 ∼ 30
we define a generating function to integrate all path similarities.
定义一个生成函数,综合所有的path similarities,简单的说就是给每个节点赋予一个权重,近的节点权重大些,远的节点权重小些。
2.4. Crowd Collectiveness with Regularization
z^l can be interpreted as the weight for l-path similarity
一个群体系统 C 的 crowd collectiveness 可以被定义为 所有个体的 individual collectiveness 的均值,可以用如下公式来计算
3 Properties of the Collectiveness
介绍了 我们定义的这个 Collectiveness 三个属性
4 Collective Motion Detection
Based on the collectiveness descriptor, we propose an algorithm called Collective Merging to detect collective motions from time-series data with noises (see Algorithm 1).
有序运动的出现
the level of random perturbation η
Collective Motion Database