Aruco marker tracking scene.

[Pasquale94]

Hello everyone.
I will leave here my solution to track an object with aruco marker with a camera and follow the movements in simulation.

I use as camera a Intel RealSense D435.

# Required import for python
import Sofa
import math 
import numpy as np
import os
path = os.path.dirname(os.path.abspath(__file__))

import cv2
import cv2.aruco as aruco
from numpy import add 
import glob
import pyrealsense2 as rs
import cv2

# Choose in your script to activate or not the GUI
USE_GUI = True

# Start streaming

def main():
    import SofaRuntime
    import Sofa.Gui

    root = Sofa.Core.Node("root")
    addScene(root)
    Sofa.Simulation.init(root)

    if not USE_GUI:
        for iteration in range(10):
            Sofa.Simulation.animate(root, root.dt.value)
    else:
        Sofa.Gui.GUIManager.Init("myscene", "qglviewer")
        Sofa.Gui.GUIManager.addGUI(root, __file__)
        Sofa.Gui.GUIManager.SetDimension(1080, 1080)
        Sofa.Gui.GUIManager.MainLoop(root)
        Sofa.Gui.GUIManager.closeGUI()

def createScene(root):
    root.gravity=[0, 0, 0]
    root.dt=0.02

    root.addObject('DefaultVisualManagerLoop')
    root.addObject('DefaultAnimationLoop')

    root.addObject('VisualStyle', displayFlags="hideCollisionModels showVisualModels showForceFields")
    root.addObject('RequiredPlugin', pluginName="SofaImplicitOdeSolver SofaConstraint SofaLoader SofaOpenglVisual SofaBoundaryCondition SofaGeneralLoader SofaGeneralSimpleFem SoftRobots SofaPython3 SofaOpenglVisual SofaSparseSolver SofaPreconditioner SofaMiscCollision SofaExporter SofaMiscFem SofaMiscForceField") 
    root.addObject('DefaultPipeline', name="CollisionPipeline")
    root.addObject('BruteForceDetection', name="N2")
    root.addObject('DefaultContactManager', name="CollisionResponse", response="default")
    root.addObject('DiscreteIntersection')
    root.addObject('LocalMinDistance', name='Proximity', alarmDistance=5, contactDistance=1, angleCone=0.0)
    root.addObject('BackgroundSetting', color=[0, 0.168627, 0.211765])

    newSphere = root.addChild('Sphere')
    newSphere.addObject('EulerImplicitSolver')
    newSphere.addObject('CGLinearSolver', threshold='1e-09', tolerance='1e-09', iterations='200')
    newSphere.addObject('MeshVTKLoader' , name = 'sphereloader', filename = path + r'/mesh/cube.vtu')
    newSphere.addObject('MechanicalObject' , src = '@sphereloader', name = 'tetrasphere', template = 'Vec3d', listening =True)
    newSphere.addObject('TetrahedronSetTopologyContainer' , src = '@sphereloader', name = 'topo')
    newSphere.addObject('TetrahedronSetTopologyModifier' , name = 'Modifier')
    newSphere.addObject('TetrahedronSetGeometryAlgorithms' , drawTriangles = '1', name = 'GeomAlgo', template = 'Vec3d')
    newSphere.addObject('UniformMass', totalMass="1")
    newSphere.addObject(Controller(node=root, translation = root.Sphere.tetrasphere.position))

    collisionsphere = newSphere.addChild('collisionSphere')
    collisionsphere.addObject('MeshObjLoader', name = 'loadersphere', filename = path + r'/mesh/cube.obj')
    collisionsphere.addObject('MeshTopology' , src = '@loadersphere', name = 'topo')
    collisionsphere.addObject('MechanicalObject' , name = 'collisMech')
    #collisionsphere.addObject('SphereCollisionModel' , selfCollision = False)
    #collisionFinger.addObject('LineCollisionModel' , selfCollision = False)
    collisionsphere.addObject('PointCollisionModel' , selfCollision = False)
    collisionsphere.addObject('BarycentricMapping')

    modelVisu = newSphere.addChild('visu')
    modelVisu.addObject('MeshObjLoader', name='loader', filename = path + r'/mesh/cube.obj')
    modelVisu.addObject('OglModel', src='@loader', color='yellow')
    modelVisu.addObject('BarycentricMapping')

    return root

class Controller(Sofa.Core.Controller):
    """ This controller monitors new sphere objects.
    Press ctrl and the L key to make spheres falling!
    """
    def __init__(self, *args, **kwargs):
        Sofa.Core.Controller.__init__(self, *args, **kwargs)
        self.translation = kwargs['translation']

        self.traslazione = np.array((0,0,0)).reshape(1,3)

    def onAnimateEndEvent(self, event):
        # Configure color streams
        pipeline = rs.pipeline()
        config = rs.config()
        config.enable_stream(rs.stream.color, 640, 480, rs.format.bgr8, 60)
       #Start stream
        pipeline.start(config)
        frames = pipeline.wait_for_frames()
        color_frame = frames.get_color_frame()
        color_image = np.asanyarray(color_frame.get_data())
        gray = cv2.cvtColor(color_image, cv2.COLOR_BGR2GRAY)
       
# aruco marker detection codes

        markerSize = 5
        totalMarkers = 250
        key = getattr(aruco, f'DICT_{markerSize}X{markerSize}_{totalMarkers}')
        arucoDict = aruco.Dictionary_get(key)
        arucoParam = aruco.DetectorParameters_create()
        bboxs, ids, rejected = aruco.detectMarkers(gray, arucoDict, parameters = arucoParam)

       #Intel Realsense camera Matrix
        mtx = [[613.8421020507812, 0, 321.9654541015625],
                [0, 614.451416015625,249.003173828125], 
                [ 0,0,1]]

        mtx= np.array(mtx).reshape(3,3)
        #distortion coefficient of the camera
          dist = [-1.71057285e-02, -7.15934343e-05, -1.00114895e-03,  1.15475573e-02, 5.46266277e-07]
        dist = np.array(dist).reshape(5,1)
        rvecs, tvecs, _objPoints = aruco.estimatePoseSingleMarkers(bboxs, 0.05, mtx, dist)
        print(ids)
# If the marker is detected track the traslations
        if ids == 0: 
            self.tvecs = np.array(tvecs).reshape(1,3)

            self.traslazione = np.hstack((self.traslazione,self.tvecs))

            size = np.size(self.traslazione)

            print(size)

            if size ==3:

                for i in range(0, len(self.translation)):
                    self.translation[i][0] = self.translation[i][0] + (self.tvecs[0][0])*1000
                    self.translation[i][1] = self.translation[i][1] + (self.tvecs[0][1])*1000
                    self.translation[i][2] = self.translation[i][2] + (self.tvecs[0][2])*1000

            if size>=6:
                self.trasl = [self.traslazione[0][size-6],self.traslazione[0][size-5],self.traslazione[0][size-4]]
                self.trasl = np.array(self.trasl).reshape(1,3)

                if np.linalg.norm((self.trasl-self.tvecs)*1000) >= 2 :
                    for i in range(0, len(self.translation)):
                        self.translation[i][0] = self.translation[i][0] + ((self.trasl[0][0] - self.tvecs[0][0])*1000)
                        self.translation[i][1] = self.translation[i][1] + ((self.trasl[0][1] - self.tvecs[0][1])*1000)
                        self.translation[i][2] = self.translation[i][2] + ((self.trasl[0][2] - self.tvecs[0][2])*1000)

        k = cv2.waitKey(60) & 0xff     
        pipeline.stop()
            

# Function used only if this script is called from a python environment
if __name__ == '__main__':
    main()

The code for now works only in traslation.
I tried to add a rotation matrix to the mesh positions using the variable rvecs from marker extimated pose of the marker, but the cube in the scene rotates but became “flat”, i also tried to use rvecs in the rotation field of the mechanicalobject of the cube, but nothing happens.
So I want to ask the devs if they can help me to improve the scene.
I tag @hugo as reference.
Thanks in advance every one.

[Author]

Posts