Tensorflowjs Posenet to BVH attempt

Posenet Camera for BVH conversion testing My PoseNet Camera here
Posenet Activated Video Saving

My Github here
A BVH player here

Mixamo OpenSource BVH files at https://www.mixamo.com/#/

The Posenet camera pointed at the Mixamo web page animated characters.

The problem is that the Posenet data is positional. It calculates an x and y value for each pose location, with an accuracy score and a labelled body part.

Example Single Pose, single snapshot PoseNet data

{
   "score": 0.0009414869196274702,
   "keypoints": [
      {
         "score": 0.00006651878356933594,
         "part": "nose",
         "position": {
            "x": 249.84802904564316,
            "y": 304.3941908713693
         }
      },
      {
         "score": 0,
         "part": "leftEye",
         "position": {
            "x": -9.578838174273859,
            "y": -0.28868514571447096
         }
      },
      {
         "score": 0,
         "part": "rightEye",
         "position": {
            "x": 12.497390430497925,
            "y": -2.538142667271784
         }
      },
      {
         "score": 0.00022864341735839844,
         "part": "leftEar",
         "position": {
            "x": 331.26815352697093,
            "y": 357.0778008298755
         }
      },
      {
         "score": 0,
         "part": "rightEar",
         "position": {
            "x": 5.787214730290456,
            "y": -5.047182961618257
         }
      },
      {
         "score": 0.0013141632080078125,
         "part": "leftShoulder",
         "position": {
            "x": 367.18879668049794,
            "y": 299.87085062240664
         }
      },
      {
         "score": 0.00040268898010253906,
         "part": "rightShoulder",
         "position": {
            "x": 140.62266597510373,
            "y": 250.91234439834025
         }
      },
      {
         "score": 0.0010004043579101562,
         "part": "leftElbow",
         "position": {
            "x": 440.36047717842325,
            "y": 487.1903526970954
         }
      },
      {
         "score": 0.0005249977111816406,
         "part": "rightElbow",
         "position": {
            "x": 162.70720954356847,
            "y": 375.969398340249
         }
      },
      {
         "score": 0.0005192756652832031,
         "part": "leftWrist",
         "position": {
            "x": 506.8801867219917,
            "y": 409.49533195020746
         }
      },
      {
         "score": 0.0007147789001464844,
         "part": "rightWrist",
         "position": {
            "x": 387.9429460580913,
            "y": 470.69346473029043
         }
      },
      {
         "score": 0.004131317138671875,
         "part": "leftHip",
         "position": {
            "x": 405.50414937759336,
            "y": 482.1348547717842
         }
      },
      {
         "score": 0.0031223297119140625,
         "part": "rightHip",
         "position": {
            "x": 412.95435684647305,
            "y": 479.4740663900415
         }
      },
      {
         "score": 0.00125885009765625,
         "part": "leftKnee",
         "position": {
            "x": 231.22251037344398,
            "y": 366.92271784232366
         }
      },
      {
         "score": 0.0009179115295410156,
         "part": "rightKnee",
         "position": {
            "x": 166.69839211618256,
            "y": 366.6566390041494
         }
      },
      {
         "score": 0.0008487701416015625,
         "part": "leftAnkle",
         "position": {
            "x": 448.34284232365144,
            "y": 481.07053941908714
         }
      },
      {
         "score": 0.0009546279907226562,
         "part": "rightAnkle",
         "position": {
            "x": 171.4878112033195,
            "y": 453.39834024896265
         }
      }

However, the BVH data is mainly angular, based on the x, y, z angle of change from the main defined location of the hips. The main BVH file once defined is then folowed by 57 values per time frame to describe the new angular location of each joint. Only thje first three values are z,x,y position values for the entire character.

Example single object multiple time frames BVH file

HIERARCHY
ROOT Hips
{
	OFFSET	0.00	0.00	0.00
	CHANNELS 6 Xposition Yposition Zposition Zrotation Xrotation Yrotation
	JOINT Chest
	{
		OFFSET	 0.00	 5.21	 0.00
		CHANNELS 3 Zrotation Xrotation Yrotation
		JOINT Neck
		{
			OFFSET	 0.00	 18.65	 0.00
			CHANNELS 3 Zrotation Xrotation Yrotation
			JOINT Head
			{
				OFFSET	 0.00	 5.45	 0.00
				CHANNELS 3 Zrotation Xrotation Yrotation
				End Site 
				{
					OFFSET	 0.00	 3.87	 0.00
				}
			}
		}
		JOINT LeftCollar
		{
			OFFSET	 1.12	 16.23	 1.87
			CHANNELS 3 Zrotation Xrotation Yrotation
			JOINT LeftUpArm
			{
				OFFSET	 5.54	 0.00	 0.00
				CHANNELS 3 Zrotation Xrotation Yrotation
				JOINT LeftLowArm
				{
					OFFSET	 0.00	-11.96	 0.00
					CHANNELS 3 Zrotation Xrotation Yrotation
					JOINT LeftHand
					{
						OFFSET	 0.00	-9.93	 0.00
						CHANNELS 3 Zrotation Xrotation Yrotation
						End Site 
						{
							OFFSET	 0.00	-7.00	 0.00
						}
					}
				}
			}
		}
		JOINT RightCollar
		{
			OFFSET	-1.12	 16.23	 1.87
			CHANNELS 3 Zrotation Xrotation Yrotation
			JOINT RightUpArm
			{
				OFFSET	-6.07	 0.00	 0.00
				CHANNELS 3 Zrotation Xrotation Yrotation
				JOINT RightLowArm
				{
					OFFSET	 0.00	-11.82	 0.00
					CHANNELS 3 Zrotation Xrotation Yrotation
					JOINT RightHand
					{
						OFFSET	 0.00	-10.65	 0.00
						CHANNELS 3 Zrotation Xrotation Yrotation
						End Site 
						{
							OFFSET	 0.00	-7.00	 0.00
						}
					}
				}
			}
		}
	}
	JOINT LeftUpLeg
	{
		OFFSET	 3.91	 0.00	 0.00
		CHANNELS 3 Zrotation Xrotation Yrotation
		JOINT LeftLowLeg
		{
			OFFSET	 0.00	-18.34	 0.00
			CHANNELS 3 Zrotation Xrotation Yrotation
			JOINT LeftFoot
			{
				OFFSET	 0.00	-17.37	 0.00
				CHANNELS 3 Zrotation Xrotation Yrotation
				End Site 
				{
					OFFSET	 0.00	-3.46	 0.00
				}
			}
		}
	}
	JOINT RightUpLeg
	{
		OFFSET	-3.91	 0.00	 0.00
		CHANNELS 3 Zrotation Xrotation Yrotation
		JOINT RightLowLeg
		{
			OFFSET	 0.00	-17.63	 0.00
			CHANNELS 3 Zrotation Xrotation Yrotation
			JOINT RightFoot
			{
				OFFSET	 0.00	-17.14	 0.00
				CHANNELS 3 Zrotation Xrotation Yrotation
				End Site 
				{
					OFFSET	 0.00	-3.75	 0.00
				}
			}
		}
	}
}
MOTION
Frames:    2
Frame Time: 0.033333
 8.03	 35.01	 88.36	-3.41	 14.78	-164.35	 13.09	 40.30	-24.60	 7.88	 43.80	 0.00	-3.61	-41.45	 5.82	 10.08	 0.00	 10.21	 97.95	-23.53	-2.14	-101.86	-80.77	-98.91	 0.69	 0.03	 0.00	-14.04	 0.00	-10.50	-85.52	-13.72	-102.93	 61.91	-61.18	 65.18	-1.57	 0.69	 0.02	 15.00	 22.78	-5.92	 14.93	 49.99	 6.60	 0.00	-1.14	 0.00	-16.58	-10.51	-3.11	 15.38	 52.66	-21.80	 0.00	-23.95	 0.00	
 7.81	 35.10	 86.47	-3.78	 12.94	-166.97	 12.64	 42.57	-22.34	 7.67	 43.61	 0.00	-4.23	-41.41	 4.89	 19.10	 0.00	 4.16	 93.12	-9.69	-9.43	 132.67	-81.86	 136.80	 0.70	 0.37	 0.00	-8.62	 0.00	-21.82	-87.31	-27.57	-100.09	 56.17	-61.56	 58.72	-1.63	 0.95	 0.03	 13.16	 15.44	-3.56	 7.97	 59.29	 4.97	 0.00	 1.64	 0.00	-17.18	-10.02	-3.08	 13.56	 53.38	-18.07	 0.00	-25.93	 0.00

I think the cosine law can be used to convert two sets of x y locations to angles, but that is where I am having some problems. This is as far as I have got at the moment
BVH formater BVH formater here

This Github, ... this Github Website Version, ... this Hosted Website Version, ... Tensorflowjs

By Jeremy Ellis
Twitter@rocksetta
Website http://rocksetta.com
Use at your own risk!

Tensorflowjs Posenet to BVH attempt

The problem is that the Posenet data is positional. It calculates an x and y value for each pose location, with an accuracy score and a labelled body part.

I think the cosine law can be used to convert two sets of x y locations to angles, but that is where I am having some problems. This is as far as I have got at the moment BVH formater BVH formater here

I think the cosine law can be used to convert two sets of x y locations to angles, but that is where I am having some problems. This is as far as I have got at the moment
BVH formater BVH formater here