世界坐标中的眼动仪 [英] eye tracker in world coordinates

问题描述

我正在尝试使用facetracklib来跟踪3D中脸部的位置。

I'm trying to use the facetracklib to track the position of face in 3D.

我使用以下2种方法来获取脸部的最终位置，

I use the following 2 ways to get the final positions of the face,

1。 IFTModel :: Get3DShape

1. IFTModel::Get3DShape

2。 IFTModel :: GetProjectedShape - > INuiCoordinateMapper :: MapColorFrameToDepthFrame - > INuiCoordinateMapper :: MapDepthPointToSkeletonPoint

2. IFTModel::GetProjectedShape -> INuiCoordinateMapper::MapColorFrameToDepthFrame -> INuiCoordinateMapper::MapDepthPointToSkeletonPoint

但结果变得不同，方法1距离方法2几厘米。有人知道为什么吗？ facetracklib是否在内部做了一些聪明的事情以提高准确性？

But the results turns to be different, method 1 gives a few cm away from method 2. Does anyone know why? Is the facetracklib doing something smart inside to improve the accuracy?

推荐答案

3d网格模型与颜色框对齐。根据面部检测的准确程度，您得到的色点可能与面部不完全相同。这里的因素是噪声和图像分辨率。

The 3d mesh model is aligned to the color frame. Depending on how accurate the face detection was the color point you get may not be exact to the face. The factors here are noise and resolution of image.

映射网格到彩色帧的距离有多近？给定颜色/深度的FOV，面部边缘可能急剧下降，导致最终深度点的一些估计。所有的积分是否相同，某些积分比其他积分更多？

How close is the mapped mesh onto the color frame? Given the FOV of color/depth, the face edges can fall off drastically causing some estimations in the final depth point. Are all the points off by the same amount, of certain ones off more than others?

这篇关于世界坐标中的眼动仪的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！