Detect raised hand using 3D skeleton

[vvasco]

From the latest tests I've been doing on the robot, it turned out that for high distances robot-user (~3m), humanStructure flickers in detecting that the hand is lifted. This occurs as the threshold applied to the distance elbow-wrist does not scale up with the depth and leads to a continuous flickering of start and stop commands, sent to googleSpeech. Using 3D skeletons, rather than 2D, should provide us with a more robust detection. managerTUG can directly deal with this.

cc @vtikha @pattacini

[vvasco]

The following compares the output of humanStructure and managerTUG when using 2D and 3D skeleton respectively:

• the message Starting speech is printed when a hand up is detected;
• the message Stopping speech is printed when the hand down is detected.

2D skeleton	3D skeleton

In total the hand is raised 4 times, so 4 couples of Starting speech / Stopping speech messages should be printed:

• when using the 2D skeleton, more messages are produced, as the detection starts flickering. Also, when the skeleton is far, the hand up is not detected until the robot approaches the person.
• when using the 3D skeleton, the detection is more robust, as the skeleton is normalized and we can easily apply a threshold on the distance elbow - hand computed along the z axis (wrt the start-line). This gives us more robustness to distance and differences in body structures.