方法： Check height of the foot

优点：easy

缺点： easily fooled，if a character skids to a stop

Baseline mehtod 2

方法： Check speed of the foot

优点：easy

缺点： unreliable  the markers have some speed even during foot plants.         marker data is noisy

Bindiganavale 98  Proceeding of the International Workshop on Modeling and Motion Capture Techniques for Virtual Environments

方法： detect zero-crossing in acceration space of the end effectors

优点： work well for non-noisy data

缺点：  unreliable on motion capture data  not reliabel when working with noisy signals    require manualy tagged-objects to avoid checking for collison with all the objects in the scene

Liu and Popovic 2002  Siggraph

方法：detect frames in which the feet are stationary

优点：work well for non-noisy data

缺点： unreliable on motion capture data, not automatic .This method is dedicated to keyframed animation and is not intended to be applied to motion capture as it does not consider noise in the data.

Kovar 2002  Symposium on Computer Animation

题目：Footskate cleanup for motion capture editing

方法：use specific thresholds on the position and velocity of the feet to detect them.

优点：

缺点：not reliable for motion capture animation as derivatives tend to amplify nosie in signals

Lee 2002 Siggraph

题目：Interactive control of AVstars animated with human motion data

方法： consider body segments and objects in the environment relative velocity and position to decide whether a body segment is in contact with an object in the scene or not

优点：

缺点：not reliable for motion capture animation as derivatives tend to amplify nosie in signals

S.Menareais    2004  Symposium on Computer Animation

题目： Synchronization for Dynamic blending of motions

方法：use specific thresholds on the position and velocity of the feet to detect them

优点：

缺点 not reliable for motion capture animation as derivatives tend to amplify nosie in signals

Ikemoto 06 Symposium on Interactive 3D Graphics

方法：use a classifier to detect when foot plants should occur.By labeling a small set of frames, a user trains a classifier to detect when the foot should be planted.The classifier then automatically labels the remainder of the frames.

优点： semi-automatic(训练部分需要手动参与),

缺点： This method is dedicated to footplants detection and would be difficult to generilized to any kind of effectors and /or constraints .Indeed ,detecting another type of constraints would require to build a new kind of teature vectors and to train the calssifier once more.

想法：这个方法没看懂。。。说实话。。（一下午都在搞这个。。出了配了个Emacs。。。）

1） 首先怎么把三维mark点的轨迹映射到二维上，而且都是对齐的？ 从root点来搞？（貌似root点的确可以搞）

2） 下面就剩一些细节的东西。。21帧的问题。。。

貌似的确是SKELETON相关的。。所以不适合我们的问题。。。summer说的的确是不错的。。

Le 06 Symposium on Computer Animation

题目：Robust kinematic constraint detection for motion data

方法：

优点：

缺点：

这个Roubust Kinematic 看得我真是头大的很啊。。。SVD分解，线性代数。。。映射空间。。。高斯噪声。。。噪声模板。。。我勒个去。。先补基础。。

首先以用Abstract的一句话 Spatial Proximities of end-effectors with tagged objects during zero-crossing in acceleration space are used to isolate significant events and abstract constraints from an agents`s action

  空间的距离约束的目的就是为了抽去出具有明显意义的帧

  The zero-crossing point in trajectory implies changes in motion such as starting from rest ,coming to a stop ,or changing the velocity direction .

  以抓取被子为例，首先在手上有一个mark点，杯子上有一个tag点。

  当二者之间的距离小于一定值的时候，开始进行跟踪。然后在加速度空间中进行Zero Crossing

   第一个图是距离约束，第二个图加速度约束，其中黄色的代表横向加速度，方向如右图所示。进行过零检测就是在Acceleration空间中进行的。。但不清楚这个结果怎么样。。

  感觉这个实验就是检测加速度变化，但套上了一个ZeroCrossing的帽子。。

  与我们实验的区别:

   1 首先这个加速度是一个二维的。。人体的Gait Analysis是一个3维的过程，但可以再前进方向上进行投影。。从而达到二维要求。

   2 我们的Gait Analysis中没有tag点。。如果说有，也只能是地面算作是一个tag点。。黄武师兄说，这其中有一个地面标定的过程。。这个的确需要。。

还需要将3D数据，从标定坐标系中转换到地面坐标系中，因为我们使用的约束条件都是地面坐标系中的。。

   3 这个实验实现起来比较方便。。

    Motion analysis data provides large amounts of dta to describe motion such as walking speed and gait events, as well as joint angles,forces, and moments as function of the percent of the gait cycle. Using those data joint kinetics ,joint moments and joint powers have been used to for gait recognition lately .

    On the other hand Biometrix gait analysis concentrates on individual`s gait recognition in a variety of eifferent areas and scenarios .As a resultof this ,biometric gait analysis is based on visual data capture and analysis systems.

   Due to lavk of information like motion analysis system biometric gait recognition users computer vision method to describe motion that is exclusively being applied to identification tasks. However, gait can disclose more that identity.