realTimeHandRotationCompute.py

'''
Goal: 從MediaPipe抓取的手land mark轉換成rotation資訊, 
總共有6個rotation資訊, 
MCP兩個旋轉軸, PIP一個旋轉軸, 
index finger以及middle finger各有一個MCP and PIP
總共只會用到7個點, 但是MediaPipe會給21個點
Note, 包含對hand landmark data的預處理, kalman filter以及height width correction
'''

import matplotlib
import numpy as np
import pandas as pd
import json
import enum
import time
import timeit
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

class jointsNames(enum.IntEnum):
    wrist = 0
    thunmbCMC = 1
    thunmbMCP = 2
    thunmbIP = 3
    thunmbTIP = 4
    indexMCP = 5
    indexPIP = 6
    indexDIP = 7
    indexTIP = 8
    middleMCP = 9
    middlePIP = 10
    middleDIP = 11
    middleTIP = 12
    ringMCP = 13
    ringPIP = 14
    ringDIP = 15
    ringTIP = 16
    pinkyMCP = 17
    pinkyPIP = 18
    pinkyDIP = 19
    pinkyTIP = 20

usedJoints = [
    jointsNames.wrist, 
    jointsNames.indexMCP, jointsNames.indexPIP, jointsNames.indexDIP, 
    jointsNames.middleMCP, jointsNames.middlePIP, jointsNames.middleDIP
]

outputJointCount = 4
negateXYZMask = [-1, 1, -1]    # 1 -> not negate, -1 -> negate
kalmanParamR = 0.1
kalmanParamQ = 0.1
kalmanX = {i:[0, 0, 0] for i in jointsNames}    # 上一個時間點的predict結果
kalmanK = 0 # kalman參數(每個joint都一樣, 所有axis都一樣)
kalmanP = 0 # kalman參數

def visualize3DVecs(*args):
    '''
    Goal: 顯示3D空間向量, 方便debug, 顯示部分joints以及vertex即可
    Input: 
    :*args: 多個3d position array組成的list
    '''
    origin = [0, 0, 0]
    # x = [1, 2, 3]
    # y = [4, 5, 6]
    # xCy = np.cross(x, y)
    xyzZip = list(zip(*[origin+_v for _v in args]))
    fig = plt.figure()
    ax = fig.add_subplot(111, projection='3d')
    colormap = matplotlib.cm.inferno
    colors = list(range(len(args)))
    norm = matplotlib.colors.Normalize()
    norm.autoscale(colors)
    for i in range(len(args)):
        ax.quiver(xyzZip[0][i], xyzZip[1][i], xyzZip[2][i], xyzZip[3][i], xyzZip[4][i], xyzZip[5][i], color=colormap(norm(colors[i])), label=i)
    ax.set_xlim([-0.2, 0.2])
    ax.set_ylim([-0.2, 0.2])
    ax.set_zlim([-0.2, 0.2])
    ax.set_xlabel('x')
    ax.set_ylabel('y')
    ax.set_zlabel('z')
    plt.legend()

def vectorProjOnPlane(v, n):
    '''
    Goal: project vector on plane
    Input:
    :v: Vector to be prjected, array with X,Y,Z values
    :n: Plane normal, array with X,Y,Z values
    '''
    n_norm = np.sqrt(sum(n**2))
    proj_of_v_on_n = (np.dot(v, n)/n_norm**2)*n
    return v - proj_of_v_on_n

def computeUsedVectors(positionData):
    '''
    Goal: 計算所需的向量數值, 
            indexWristToMCP, middleWristToMCP, 
            indexMCPToPIP, indexPIPToDIP, 
            middleMCPToPIP, middlePIPToDIP, 
            indexMCPNormal([9]-[5]), palmNormal(indexWristToMCP x indexMCPNormal), middlePalmNormal, 
            indexProjectToMCPNormal, indexProjectToPalmNormal, 
            middleProjectToMCPNormal, middleProjectToPalmNormal
    Input:
    :positionData: list of dicts, 單一時間點所有joints的XYZ 3d position資訊
    Output:
    :: list of arrays, 所有需要的vectors
    '''
    # 1. 得到所需的joint's XYZ 3D position資訊
    wristPos = np.array([
        positionData[jointsNames.wrist]['x'], positionData[jointsNames.wrist]['y'], positionData[jointsNames.wrist]['z']
    ])
    indexMCPPos = np.array([
        positionData[jointsNames.indexMCP]['x'], positionData[jointsNames.indexMCP]['y'], positionData[jointsNames.indexMCP]['z']
    ])
    indexPIPPos = np.array([
        positionData[jointsNames.indexPIP]['x'], positionData[jointsNames.indexPIP]['y'], positionData[jointsNames.indexPIP]['z']
    ])
    indexDIPPos = np.array([
        positionData[jointsNames.indexDIP]['x'], positionData[jointsNames.indexDIP]['y'], positionData[jointsNames.indexDIP]['z']
    ])
    middleMCPPos = np.array([
        positionData[jointsNames.middleMCP]['x'], positionData[jointsNames.middleMCP]['y'], positionData[jointsNames.middleMCP]['z']
    ])
    middlePIPPos = np.array([
        positionData[jointsNames.middlePIP]['x'], positionData[jointsNames.middlePIP]['y'], positionData[jointsNames.middlePIP]['z']
    ])
    middleDIPPos = np.array([
        positionData[jointsNames.middleDIP]['x'], positionData[jointsNames.middleDIP]['y'], positionData[jointsNames.middleDIP]['z']
    ])
    # 2. 向量數值計算
    indexWristToMCP = indexMCPPos - wristPos
    middleWristToMCP = middleMCPPos - wristPos
    indexMCPToPIP = indexPIPPos - indexMCPPos
    indexPIPToDIP = indexDIPPos - indexPIPPos
    middleMCPToPIP = middlePIPPos - middleMCPPos
    middlePIPToDIP = middleDIPPos - middlePIPPos
    indexMCPNormal = middleMCPPos - indexMCPPos
    # indexMCPNormal = indexMCPPos - middleMCPPos # Python改為右手定則, 所以轉軸要相反過來
    indexMCPNormalNormalized = indexMCPNormal/np.linalg.norm(indexMCPNormal)
    # Note that numpy cross product is right-hand-rule
    palmNormal = np.cross(
        indexMCPNormalNormalized, # normalize the vector
        indexWristToMCP/np.linalg.norm(indexWristToMCP)
    )
    middlePalmNormal = np.cross(
        indexMCPNormalNormalized, 
        middleWristToMCP/np.linalg.norm(middleWristToMCP)
    )
    # Vector project to a plane: https://www.geeksforgeeks.org/vector-projection-using-python/
    indexProjectToMCPNormal = vectorProjOnPlane(indexMCPToPIP, indexMCPNormalNormalized)
    indexProjectToPalmNormal = vectorProjOnPlane(indexMCPToPIP, palmNormal)
    middleProjectToMCPNormal = vectorProjOnPlane(middleMCPToPIP, indexMCPNormalNormalized)
    middleProjectToPalmNormal = vectorProjOnPlane(middleMCPToPIP, middlePalmNormal)
    # wrist to MCP needs to be projected to plane too
    # but this is not considered in the previous unity code
    indexWristProjToMCPNormal = vectorProjOnPlane(indexWristToMCP, indexMCPNormalNormalized)
    middleWristProjToMCPNormal = vectorProjOnPlane(middleWristToMCP, indexMCPNormalNormalized)

    # print(indexProjectToMCPNormal)
    return [indexWristToMCP, middleWristToMCP, 
            indexMCPToPIP, indexPIPToDIP, 
            middleMCPToPIP, middlePIPToDIP, 
            indexMCPNormal, palmNormal, middlePalmNormal, 
            indexProjectToMCPNormal, indexProjectToPalmNormal, 
            middleProjectToMCPNormal, middleProjectToPalmNormal, 
            indexWristProjToMCPNormal,middleWristProjToMCPNormal]

def angleBetweenTwoVecs(v1, v2, isSigned: bool=False, rotationDirection=None):
    '''
    Goal: compute the angle from vector1 to vector2
    Input: 
    :rotationDirection: The direction that decide the rotation's sign(right-hand-rule)
    '''
    unit_vector_1 = v1 / np.linalg.norm(v1)
    unit_vector_2 = v2 / np.linalg.norm(v2)
    dot_product = np.dot(unit_vector_1, unit_vector_2)
    angle = np.arccos(dot_product)  # return angle in radian
    angleDegree = angle/ np.pi * 180
    if (isSigned) and (rotationDirection is not None):
        c = np.cross(v1, v2)
        # print('c: ', c)
        _dot_product = np.dot(c, rotationDirection)
        _angle = np.arccos(_dot_product)  # return angle in radian
        _angleDegree = _angle/ np.pi * 180
        # print('angle c and targetDirection: ', _angleDegree)
        if _angleDegree>90:
            angleDegree=-angleDegree
        # if c>0:
        #     angleDegree=-angleDegree
    return angleDegree

def computeUsedRotations(indexWristToMCP, middleWristToMCP, 
            indexMCPToPIP, indexPIPToDIP, 
            middleMCPToPIP, middlePIPToDIP, 
            indexMCPNormal, palmNormal, middlePalmNormal, 
            indexProjectToMCPNormal, indexProjectToPalmNormal, 
            middleProjectToMCPNormal, middleProjectToPalmNormal, 
            indexWristProjToMCPNormal, middleWristProjToMCPNormal):
    '''
    Goal: 利用求得的vectors進一步計算所需的rotations
    Input:
    :usedVecs: list of arrays, [indexWristToMCP, middleWristToMCP, 
            indexMCPToPIP, indexPIPToDIP, 
            middleMCPToPIP, middlePIPToDIP, 
            indexMCPNormal, palmNormal, 
            indexProjectToMCPNormal, indexProjectToPalmNormal, 
            middleProjectToMCPNormal, middleProjectToPalmNormal]
    Output:
    :: list, rotation的計算結果
    '''
    # angle between teo vectors
    # ref: https://www.geeksforgeeks.org/how-to-find-the-angle-between-two-vectors/
    # ref: https://chadrick-kwag.net/get-rotation-angle-between-two-vectors/
    indexPIPAngle = angleBetweenTwoVecs(indexMCPToPIP, indexPIPToDIP)
    # indexMCPnormal需要乘上-1. 因為, python為右手定則, 轉軸需要相反後才會輸出相同方向的旋轉數值
    indexMCPAngle1 = angleBetweenTwoVecs(indexWristProjToMCPNormal, indexProjectToMCPNormal, True, indexMCPNormal*-1)
    # indexMCPAngle1 = angleBetweenTwoVecs(indexProjectToMCPNormal, indexWristToMCP, True, indexMCPNormal)
    indexMCPAngle2 = angleBetweenTwoVecs(indexProjectToPalmNormal, indexWristToMCP, True, palmNormal)

    middlePIPAngle = angleBetweenTwoVecs(middleMCPToPIP, middlePIPToDIP)
    middleMCPAngle1 = angleBetweenTwoVecs(middleWristProjToMCPNormal, middleProjectToMCPNormal, True, indexMCPNormal*-1)
    # middleMCPAngle1 = angleBetweenTwoVecs(middleProjectToMCPNormal, middleWristToMCP, True, indexMCPNormal)
    middleMCPAngle2 = angleBetweenTwoVecs(middleProjectToPalmNormal, middleWristToMCP, True, middlePalmNormal)

    return [indexMCPAngle1, indexMCPAngle2, indexPIPAngle, middleMCPAngle1, middleMCPAngle2, middlePIPAngle]

def kalmanFilter(LMJson, usedJoints):
    '''
    Goal: 對landmark time series做kalman filter平滑化
    Input:
    :LMJson: 單一時間json格式的landmark資料, list of dicts
    :usedJoints: 只會針對這些joints做kalman filter
    
    Output:
    :: kalman filter平滑化結果
    '''
    global kalmanX
    global kalmanK
    global kalmanP
    # measurement update
    kalmanK = (kalmanP+kalmanParamQ) / (kalmanP+kalmanParamQ+kalmanParamR)
    kalmanP = kalmanParamR * (kalmanP+kalmanParamQ) / (kalmanParamR+kalmanP+kalmanParamQ)
    # print('kalman gain: ', kalmanK)
    # estimate = previous estimate + (measurement - previous estimate) * kalman gain
    for aJoint in usedJoints:
        for i, aAxis in enumerate(['x', 'y', 'z']):
            LMJson[aJoint][aAxis] = \
                kalmanX[aJoint][i] + (LMJson[aJoint][aAxis] - kalmanX[aJoint][i])*kalmanK
            kalmanX[aJoint][i] = LMJson[aJoint][aAxis]
    return LMJson

def heightWidthCorrection(LMPos, usedJoints, width, height):
    '''
    Goal: 校正長寬都已經被normalize到[0,1]的landmarks, 
            利用image原始的長寬校正y axis的3D position
    Input: 
    :LMPos: 單一時間json格式的landmark資料, list of dicts
    :usedJoints: 只會針對這些joints做y軸position的校正
    :width: 原始image寬 -> y
    :height: 原始image高 -> x
    '''
    # -1的理由單純只是Unity端也-1, 確認沒問題後可以不要-1
    heightWidthRatio = (height-1) / (width-1)
    for aJoint in usedJoints:
        LMPos[aJoint]['y'] *= heightWidthRatio
    return LMPos

def negateAxes(LMPos, negateMask, usedJoints):
    '''
    Goal: X Y Z的座標軸反轉
    Input:
    :LMPos: 單一時間json格式的landmark資料, list of dicts
    :negateMask: XYZ那些軸需要反轉
    :usedJoints:只會針對這些joints做y軸position的校正
    Output:
    :: 反轉後的landmark結果
    '''
    for aJoint in usedJoints:
        for i, aAxis in enumerate(['x', 'y', 'z']):
            LMPos[aJoint][aAxis] *= negateMask[i]
    return LMPos

# For testing(plot the 3d vectors)
if __name__ == '__main01__':
    # origin = [0, 0, 0]
    x = [1, 2, 3]
    y = [4, 5, 6]
    xCy = np.cross(x, y)
    # xyzZip = list(zip(*[origin+x, origin+y, origin+list(xCy)]))
    # fig = plt.figure()
    # ax = fig.add_subplot(111, projection='3d')
    # ax.quiver(xyzZip[0], xyzZip[1], xyzZip[2], xyzZip[3], xyzZip[4], xyzZip[5])
    # ax.set_xlim([-1, 3])
    # ax.set_ylim([-1, 3])
    # ax.set_zlim([-1, 3])
    # ax.set_xlabel('x')
    # ax.set_ylabel('y')
    # ax.set_zlabel('z')
    visualize3DVecs(x, y, list(xCy))
    plt.show()

# For testing angle coputation
if __name__ == '__main01__':
    tmp1 = np.array([1,0])
    tmp2 = np.array([1,1])
    tmp3 = np.array([0,1])
    tmp4 = np.array([-1,1])
    tmp5 = np.array([-1,0])
    tmp6 = np.array([-1,-1])
    tmp7 = np.array([0,-1])
    tmp8 = np.array([1,-1])
    # print(angleBetweenTwoVecs(tmp1, tmp2, True))
    # print(angleBetweenTwoVecs(tmp1, tmp3, True))
    # print(angleBetweenTwoVecs(tmp1, tmp4, True))
    # print(angleBetweenTwoVecs(tmp1, tmp5, True))
    # print(angleBetweenTwoVecs(tmp1, tmp6, True))
    # print(angleBetweenTwoVecs(tmp1, tmp7, True))
    # print(angleBetweenTwoVecs(tmp1, tmp8, True))
    # right-hand-rule, index finger is X, middle finger is Y, thumb is Z
    print(angleBetweenTwoVecs(np.array([1, 0, 0]), np.array([0, 1, 0]), True, np.array([0, 0, 1])))
    print(angleBetweenTwoVecs(np.array([1, 0, 0]), np.array([0, -1, 0]), True, np.array([0, 0, 1])))
    print(angleBetweenTwoVecs(np.array([1, 0, 0]), np.array([0, 0, 1]), True, np.array([0, -1, 0])))
    print(angleBetweenTwoVecs(np.array([1, 0, 0]), np.array([0, 0, 1]), True, np.array([0, 1, 0])))
    print(angleBetweenTwoVecs(np.array([0, 0, 1]), np.array([1, 0, 0]), True, np.array([0, -1, 0])))
    print(angleBetweenTwoVecs(np.array([0, 0, 1]), np.array([1, 0, 0]), True, np.array([0, 1, 0])))

# For testing (body DB rotation verification)
# Between generic(Hips adjust to 0) and humanoid(without hips modification)
# These rotations are from Unity 
if __name__ == '__main01__':
    # 1. read both rotations files
    humanoidSavePath = 'bodyDBRotation/genericAvatar/' 
    genericSavePath = 'bodyDBRotation/genericAvatar/'

    humanoidRotJson = None
    with open(humanoidSavePath+'leftSideKick_withHip.json', 'r') as fileOpen: 
        humanoidRotJson=json.load(fileOpen)
        humanoidRotJson=humanoidRotJson['results']

    genericRotJson=None
    with open(genericSavePath+'leftSideKick.json', 'r') as fileOpen: 
        genericRotJson=json.load(fileOpen)
        genericRotJson=genericRotJson['results']
        
    humanoidTimeCount = len(humanoidRotJson)
    genericTimeCount = len(genericRotJson)

    # 2. make data to time series
    humanoidRotSeries = [humanoidRotJson[t]['data'][0]['x'] for t in range(humanoidTimeCount)]
    genericRotSeries = [genericRotJson[t]['data'][0]['x'] for t in range(genericTimeCount)]

    # 3. plot data in maplotlib
    plt.figure()
    plt.plot(range(humanoidTimeCount), humanoidRotSeries, label='humanoid')
    plt.plot(range(genericTimeCount), genericRotSeries, label='generic')
    plt.legend()
    plt.show()

# For testing(plot rotation time series curves)
# Unity version and python real time version
# Finished!! 與Unity的結果幾乎相同了, 些微的相位位移不影響
if __name__ == '__main01__':
    # 1. read Unity version
    saveDirPath='HandRotationOuputFromHomePC/'
    unityRotJson = None
    with open(saveDirPath+'leftFrontKick.json', 'r') as fileOpen: 
    # with open(saveDirPath+'leftSideKick.json', 'r') as fileOpen: 
        unityRotJson=json.load(fileOpen)
        unityRotJson=unityRotJson['results']
    unityTimeCount = len(unityRotJson)
    # 2. read python version
    realtimeRotJson=None
    with open(saveDirPath+'leftFrontKickStream.json', 'r') as fileOpen: 
    # with open(saveDirPath+'leftSideKickStream.json', 'r') as fileOpen: 
        realtimeRotJson=json.load(fileOpen)
    pythonTimeCount = len(realtimeRotJson)
    # 3. make data to time series
    # 弄清楚unity收集資料的frequency, 調整到兩者frequency相應的情況
    # Unity -> 1s 33.33次讀取, 1s 20次儲存rotation => 每5筆資料, 有兩筆資料的耗損
    jointInd = 0
    jointAxis = 'x'
    unityRotSeries = [unityRotJson[t]['data'][jointInd][jointAxis] for t in range(unityTimeCount)]
    unityRotSeries = [r-360 if r>180 else r for r in unityRotSeries]
    # realtimeRotSeries = [realtimeRotJson[t]['data'][0]['x'] for t in range(unityTimeCount)]
    realtimeRotSeries = [realtimeRotJson[t]['data'][jointInd][jointAxis] for t in range(unityTimeCount+200) if (t%5==1) or (t%5==3) or (t%5==4)] # 模仿Unity的採樣頻率
    # print(unityRotSeries)
    # print(realtimeRotSeries)
    # 4. plot both version
    plt.figure()
    plt.plot(range(unityTimeCount), unityRotSeries, label='unity')
    plt.plot(range(len(realtimeRotSeries)), realtimeRotSeries, label='real time')
    plt.legend()
    plt.show()

if __name__ == '__main01__':
    # 1. Read hand landmark data(only keep joints in used)
    # 1.1 make it a streaming data (already a streaming data)
    # 1.2 kalman filter
    # 1.3 height width correction
    # 2. 計算向量
    #   2.1 手掌法向量
    # 3. 計算角度
    # 4.0 使用index區間指定部分想要輸出的資料點
    # 4. Store computed rotations
    # 5. Store computation time cost


    # 1. 
    saveDirPath = 'complexModel/'
    handLMJson = None
    # with open(saveDirPath+'frontKick.json', 'r') as fileOpen: 
    # with open(saveDirPath+'walk.json', 'r') as fileOpen: 
    # with open(saveDirPath+'leftSideKick.json', 'r') as fileOpen: 
    # with open(saveDirPath+'runSprint.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/jumpHurdle_rgb.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/walkInjured_rgb_2022_9_12.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/jumpJoy_rgb.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/jumpJoy_rgb_2022_12_14.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/twoLegJump_rgb.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/runSprint_leftToRight_rgb2.json', 'r') as fileOpen: 
    # with open(saveDirPath+'newRecord/runSprint_2_20_1.json', 'r') as fileOpen: 
    with open(saveDirPath+'newRecord/runSprintAndFrontKick_3_2_5.json', 'r') as fileOpen: 
        handLMJson=json.load(fileOpen)
    timeCount = len(handLMJson)
    print('time count: ', timeCount)

    # 1.1 1.2 1.3
    LMPreprocTimeLaps = np.zeros(timeCount)
    for t in range(timeCount):
        handLMJson[t]['data'] = negateAxes(handLMJson[t]['data'], negateXYZMask, usedJoints)
        handLMJson[t]['data'] = heightWidthCorrection(handLMJson[t]['data'], usedJoints, 848, 480)
        handLMJson[t]['data'] = kalmanFilter(handLMJson[t]['data'], usedJoints)
        LMPreprocTimeLaps[t] = timeit.default_timer()
        # break
    LMPreprocCost = LMPreprocTimeLaps[1:] - LMPreprocTimeLaps[:-1]
    print('landmark preproc avg time: ', np.mean(LMPreprocCost))
    print('landmark preproc time std: ', np.std(LMPreprocCost))
    print('landmark preproc max time cost: ', np.max(LMPreprocCost))
    print('landmark preproc min time cost: ', np.min(LMPreprocCost))

    # 2. 3. 
    computedRotations = [None for t in range(timeCount)]
    rotComputeTimeLaps = np.zeros(timeCount)
    for t in range(timeCount):
        usedVecs = computeUsedVectors(handLMJson[t]['data'])
        # MCP的flexion向量visualize
        # visualize3DVecs(*[usedVecs[i].tolist() for i in [0, 2, 6, 9, 13]]+[[1,0,0],[0,1,0], [0,0,1]])
        # MCP的abduction向量visualize
        # visualize3DVecs(*[usedVecs[i].tolist() for i in [0, 2, 7, 10]])
        targetRotations = computeUsedRotations(*usedVecs)
        computedRotations[t] = targetRotations
        rotComputeTimeLaps[t] = timeit.default_timer()
        # break
    rotComputeCost = rotComputeTimeLaps[1:] - rotComputeTimeLaps[:-1]
    print('rotation compute avg time: ', np.mean(rotComputeCost))
    print('rotation compute time std: ', np.std(rotComputeCost))
    print('rotation compute max time cost: ', np.max(rotComputeCost))
    print('rotation compute min time cost: ', np.min(rotComputeCost))
    # plt.show()
    
    # 4.0 (只有部分資料需要使用這個功能, 挑選部分資料點的功能)
    ## 目前使用到這個功能的有side kick
    # indexInterval = [2600, 3500]
    # computedRotations = computedRotations[indexInterval[0] : indexInterval[1]+1]
    # timeCount = len(computedRotations)

    # 4. 
    # - left upper leg(X, Z), left knee(X), right upper leg, right knee
    # - X axis is the flexion, Z axis is the abduction
    rotComputeJsonData = [{'time': t, 'data': [{a: 0 for a in ['x', 'y', 'z']} for i in range(outputJointCount)]} for t in range(timeCount)]
    rotComputeRetSaveDirPath = 'HandRotationOuputFromHomePC/'
    for t in range(timeCount):
        rotComputeJsonData[t]['data'][0]['x'] = computedRotations[t][0]
        rotComputeJsonData[t]['data'][0]['z'] = computedRotations[t][1]
        rotComputeJsonData[t]['data'][1]['x'] = computedRotations[t][2]
        rotComputeJsonData[t]['data'][2]['x'] = computedRotations[t][3]
        rotComputeJsonData[t]['data'][2]['z'] = computedRotations[t][4]
        rotComputeJsonData[t]['data'][3]['x'] = computedRotations[t][5]
    # with open(rotComputeRetSaveDirPath+'leftFrontKickStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'leftSideKickStream.json', 'w') as WFile: 
    # with open(rotComputeRetSaveDirPath+'runSprintStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'runSprintStream2.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'walkStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'hurdleJumpStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'walkInjuredStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'jumpJoyStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'jumpJoyStream_2022_12_14.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'twoLegJumpStream.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'runSprint_leftToRight_rgb2.json', 'w') as WFile:
    # with open(rotComputeRetSaveDirPath+'runSprint_2_20_1.json', 'w') as WFile:
    with open(rotComputeRetSaveDirPath+'runSprintAndFrontKick_3_2_5.json', 'w') as WFile:
        json.dump(rotComputeJsonData, WFile)
        pass

    # 5. store time cost 
    # saveDirPath='timeConsume/frontKick/rotationCompute.csv'
    # timeCostDf = pd.DataFrame({
    #     'lmPreproc': rotComputeCost,
    #     'rotationComp': rotComputeCost
    # })
    # timeCostDf.to_csv(saveDirPath, index=False)