人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

Understanding Pedestrian Behaviors from Stationary Crowd Groups
CVPR2015

本文主要探讨 静态人群对行人行为的影响
人群行为的建模以前主要考虑的因素有:scene layout (e.g. entrances, exits, walls, and obstacles), pedestrian beliefs (the choice of source and destination), and interactions with other moving pedestrians

但是 静态人群 stationary crowd groups 没有被考虑,这里我们将其纳入到我们的人群行为分析的模型中。

人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
从上图我们可以看出 stationary crowd groups 对每个行人的影响不一样。对于离开或加入某一个静态人群的,它可以看作始发地或目的地,对于路过静态人群的人来说,它是一个障碍物。但是它和真正的障碍物 scene obstacles,因为 静态人群是可以被穿过的,而 scene obstacles 是不能被穿过的。再就是静态人群随着时间的变化可能是变化的,而 scene obstacles 是不会变化的。

3 Pedestrian Behavior Modeling
行人路径的选择有点类似于水流。一个人一般情况下会选择最便捷和有效的路径达到目的地。基于这个假设,一个广义的场景能量图 M 用于建模 场景中每个位置路过的难度。能量值高的区域对应容易穿过的地方。更多的行人倾向于选择这样的地方穿过。能量值低的地方对应人出现的概率低。这些地方不利于人的经过。

在我们的模型中,我们考虑三个因素: scene layout, moving pedestrians, and stationary groups,不同的因素对行人路径的选择影响也不一样。对于每个因素的权值我们通过从训练数据中学习得到。

考虑到每个行人喜好不同,我们设置了一个参数 a personality parameter P,Personalized energy maps 个人能量图 M_P 基于 广义能量图 W和个人喜好参数 P 来生成。

3.1. General energy map modeling
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
Θ = [θ1 ,θ2 ,θ3 ,θ4 ]T 表示每个因素的权重。
Scene Layout, Moving Pedestrians, and Stationary Groups. These channels are represented by f_SL , f_MP , and f_SG

3.2. Scene layout factor
行人的行为是都到场景布局的约束的。人不能出现在某些地方,如 walls and other static obstacles,同时人也倾向于远离这些 障碍物。这里我们用下面的公式对其建模:
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
where SL is a set of locations occupied by scene obstacles which are unreachable。
d 1 (x,SL) = min y∈SL ||x − y|| measures the distance from the current location x to its nearest scene obstacle location y, and θ1 is a parameter indicating the influence bandwidth (which also can be viewed as the importance) of the scene layout term

人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

3.3. Influence of moving pedestrians
其他运动行人的影响是另一个需要考虑的因素,一般我们倾向于选择与其他人保持一定的距离,所以一般不会选择已经有人的地方作为我们经过的地方。
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
where MP_i (i ∈ [1,m]) is the i th moving pedestrian
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

3.4. Influence of stationary crowd groups
这里我们从两个角度对 stationary crowd groups 进行建模,一个是将 stationary crowd groups 当做障碍物,有的人选择绕过, The group has a repulsive force around the group region to keep moving pedestrians away,第二个角度就是有人选择穿过这个 stationary crowd groups,there should be a penalty inside the group region,这一点是和障碍物最大的不同之处。
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
d3 (x,SG i ) = min y∈SG i ||x − y|| measures the distance from x to the stationary crowd group region SG
d4 (SG i ) ∈ (0,+∞) is used to measure the sparsity of stationary crowd group region SG

3.5. Personalized energy map modeling
a personality parameter P
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

3.6. Path generation
行人路径的选择,这里我们使用 Fast Marching 算法来计算得到
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

3.7. Model learning
model parameters can be optimized by maximizing likelihood on the training data
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

4 Pedestrian walking route dataset
这里我们搞了一个人群移动轨迹数据库,人工标记了一个小时的视频。
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

我们学习到的不同因素影响权重
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

5 Applications
有了这个能量图,我们可以干什么了?
5.1. Prediction on pedestrian walking paths
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

5.2. Prediction of pedestrian destinations
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

5.3. Personality attribute estimation
根据人的行为可以分为三类 aggressive, conservative, and abnormal
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

5.5. Abnormal behavior detection
预测到的和观察到的很不一样,我们认为是异常情况
人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups

人群行为分析--Understanding Pedestrian Behaviors from Stationary Crowd Groups