深色识别错误 [英] recognition errors with dark colors

问题描述

Kinect 1.0在识别暗色衣服的个体方面存在问题，这些衣服反映了浅色或深色皮肤的人。似乎没有跟踪/识别被黑色/反光衣服覆盖的身体部位的关节，或者黑色皮肤的人，b
关节是裸体的。

Kinect 1.0 had issues with recognition of individuals with dark colored clothes that reflect the light or dark skin people. It seems that joints are not tracked/recognized for parts of body covered with dark/reflective clothes, or people with dark skin who's joints are nude.

显然似乎那些Kinect 2.0也出现了问题。是否有解决方案或任何即将到来的解决方案？

Apparently it seems that those issues arise with Kinect 2.0 as well. Is there a solution or any upcoming solution to this matter?

推荐答案

它不一定是材料物理性质的颜色。如果材料吸收或高度反射，则会导致损耗或噪声过大，从而导致传感器无法获得所需的数据。由于v2使用
航班的时间，这比v1好，但你必须检查各种数据（ir，深度，身体甚至）并确定数据是否良好。

It isn't necessarily the color by the physical property of the material. If the material absorbs or is highly reflective, then this is going to cause loss or too much noise and therefore cause the sensor not to get the data it needs. Since v2 uses time of flight, this is better than v1, but you will have to check the various data(ir, depth, body even) and make determinations on if the data is good.

例如，如果你希望地板覆盖率至少为框架的50％（图像的下半部分），你可以想出一个简单的算法来检查每个像素，即使有人在帧中，这些像素应该达到平均+/-百分之几。
如果有人走进穿着高吸水性的完整黑色衣服的框架，那将会降低平均IR像素值。你将不得不进行实验。

For example, if you expect to have a floor cover at least 50% of the frame(bottom half of the image) you can come up with a simple algorithm that checks each pixel and even if someone is in frame those pixels should come to an average +/- some percent. If someone walks into frame that is wearing complete black clothing that is highly absorbent, that will drop the average IR pixel values. You will have to experiment.

另一个例子是，如果您希望跟踪人员。我知道我有一条能吸收红外线的牛仔裤，所以我无法主动跟踪我的腿。您可以确定是这种情况，因为您可能能够获得所有其他身体关节但下部关节
始终是"推断"的。或未跟踪。

another example is if you expect to track people. I know I have a pair of jeans that will absorb the IR so I cannot actively track my legs. You can determine that is the case since you might be able to get the all the other body joints but the lower joints are always "inferred" or not tracked.

这篇关于深色识别错误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！