Visual Gesture Builder中的粒度更高可能吗？ [英] Is more granularity in Visual Gesture Builder possible?

问题描述

我正在寻找一些真正不使用头部，对肩或躯干的手势（连续和离散） - 它实际上只是手臂。有没有办法从构建解决方案时分析的骨架中排除额外的骨骼，
类似于排除腿部或另一只手臂的方式？我一直看到前10个弱分类器输出中包含不需要的骨骼，我觉得它可能会降低我项目中的手势检测效果。

I'm looking to create some gestures (both continuous and discrete) that really don't use the head, opposite shoulder, or torso - it's really just arm only. Is there a way to exclude the extra bones from the skeleton that's analyzed when building a solution, similar to the way you can exclude legs or the other arm? I keep seeing the unneeded bones being included in the Top 10 Weak Classifiers output, and I feel like it might be degrading the gesture detection in my project.

此外，是否可以跟踪个人手指/指尖超出简单的开/关/套索状态？我找不到任何关于此的文档，但我认为我在Kinect工作的一段时间后看到了一个视频*真的很好地做了手指跟踪。虽然I
可能很疯狂。

Additionally, is it possible to track individual fingers/fingertips beyond the simple open/closed/lasso states? I can't find any documentation on this, but I thought I saw a video a while back of the Kinect working *really* well doing finger tracking. I might be crazy though.

推荐答案

您是否已深入了解Visual Gesture Builder，包括白皮书？

https://social.msdn.microsoft.com/Forums/en-US/02e0302a-e3bd-46d3-9146-0dacd11d2a8d/deep-dive-videos-and- whitepaper-for-visual-gesture-builder？forum = kinectv2sdk

当你完成向导时，会询问你想要使用什么关节。改进跟踪的唯一方法是使用良好的标记策略并使用大量记录数据。

When you go through the wizard, it asks what joints you want to use. The only way to improve tracking is with a good tagging strategy and using a lot of recording data.

除运行时提供的基本Body跟踪数据外，您无法检测到任何其他内容。由于未在核心运行时中跟踪头部方向，因此无法使用。根据录制情况以及如何检测手部状态，这可能会增加更多噪音。在进行录音时（地板6英尺处，传感器可以完全无障碍地看到地板，用户的头部和肩部朝向相机），请务必设置最佳传感器设置。

You cannot detect anything other than the base Body tracking data the runtime provides. Since the head orientation is not tracked in the core runtime, then that cannot be used. Depending on recording and how the hand states are detected, this might add more noise. Be sure to setup the optimal sensor setup when doing your recordings(6 feet from floor and the sensor has full unobstructed view of the floor with the users head and shoulders facing the camera)

这篇关于Visual Gesture Builder中的粒度更高可能吗？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！