# Revision history [back]

### Camera calibration and pose estimation (OpenGL + OpenCV)

Hello i'm fairly new to OpenCV. I'm trying to estimate the 3D pose of the camera in order to draw a 3D teapot using OpenGL. I have been testing for about a week and I partially understand the theory, I tried to replicate an example but I do not get it to appear correctly. I get the keypoints using SIFT and looks right.

I use this function in order to obtain the intrinsics and extrinsics parameters:

ret, K, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, im_size,None,None)


When I have these parameters I create the loop to draw the teapot:

def setup():
pygame.init()
pygame.display.set_mode((im_size[0], im_size[1]), pygame.OPENGL | pygame.DOUBLEBUF)
pygame.display.set_caption('OpenGL AR demo')

setup()
S = 1 # Selected Image
while True:
event = pygame.event.poll()
if event.type in (pygame.QUIT, pygame.KEYDOWN):
break

draw_background(I[S - 1])
set_projection_from_camera(K)
set_modelview_from_camera(rvecs[S-1], tvecs[S-1])
draw_teapot(100)

pygame.display.flip()
pygame.time.wait(50)


I estimate the projection using this function:

def set_projection_from_camera(K):
glMatrixMode(GL_PROJECTION)
fx = K[0,0]
fy = K[1,1]
fovy = 2*arctan(0.5*im_size[1]/fy)*180/pi
aspect = (im_size[0]*fy)/(im_size[1]*fx)
# define the near and far clipping planes
near = 0.1
far = 500000.0
# set perspective
gluPerspective(fovy, aspect, near, far)
glViewport(0, 0, im_size[0], im_size[1])


I estimate the modelview using this function:

def set_modelview_from_camera(rvec, tvec):
glMatrixMode(GL_MODELVIEW)

Rx = array([[1, 0, 0], [0, 0, -1], [0, 1, 0]]) # rotate the teapot
M = eye(4)
M[:3, :3] = dot(Rx, cv2.Rodrigues(rvec)[0])
M[:3, 3] = tvec.T
cv2GlMat = array([[1,0,0,0],[0,-1,0,0],[0,0,-1,0],[0,0,0,1]]) #OpenCV -> OpenGL matrix
M = dot(cv2GlMat, M)
m = M.T.flatten()


The function to draw the background (the real image)

def draw_background(I):
bg_image = Image.fromarray(I)
bg_data = bg_image.tobytes('raw', 'RGBX', 0, -1)

glMatrixMode(GL_MODELVIEW)
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
# bind the texture
glEnable(GL_TEXTURE_2D)
glBindTexture(GL_TEXTURE_2D, glGenTextures(1))
glTexImage2D(GL_TEXTURE_2D,0,GL_RGBA,bg_image.size[0],bg_image.size[1],0,GL_RGBA,GL_UNSIGNED_BYTE,bg_data)
glTexParameterf(GL_TEXTURE_2D,GL_TEXTURE_MAG_FILTER,GL_NEAREST)
glTexParameterf(GL_TEXTURE_2D,GL_TEXTURE_MIN_FILTER,GL_NEAREST)
# create quad to fill the whole window
glTexCoord2f(0.0,0.0); glVertex3f(-1,-1,-1.0)
glTexCoord2f(1.0,0.0); glVertex3f( 1,-1,-1.0)
glTexCoord2f(1.0,1.0); glVertex3f( 1, 1,-1.0)
glTexCoord2f(0.0,1.0); glVertex3f(-1, 1,-1.0)
glEnd()
# clear the texture
glDeleteTextures(1)


I understand the concepts of Opencv, but I can not put together and the object appears correctly on the screen.

### Camera calibration and pose estimation (OpenGL + OpenCV)

Hello i'm fairly new to OpenCV. I'm trying to estimate the 3D pose of the camera in order to draw a 3D teapot using OpenGL. I have been testing for about a week and I partially understand the theory, I tried to replicate an example but I do not get it to appear correctly. I get the keypoints using SIFT and looks right.

I use this function in order to obtain the intrinsics and extrinsics parameters:

ret, K, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, im_size,None,None)


When I have these parameters I create the loop to draw the teapot:

def setup():
pygame.init()
pygame.display.set_mode((im_size[0], im_size[1]), pygame.OPENGL | pygame.DOUBLEBUF)
pygame.display.set_caption('OpenGL AR demo')

setup()
S = 1 # Selected Image
while True:
event = pygame.event.poll()
if event.type in (pygame.QUIT, pygame.KEYDOWN):
break

draw_background(I[S - 1])
set_projection_from_camera(K)
set_modelview_from_camera(rvecs[S-1], tvecs[S-1])
draw_teapot(100)

pygame.display.flip()
pygame.time.wait(50)


I estimate the projection using this function:

def set_projection_from_camera(K):
glMatrixMode(GL_PROJECTION)
fx = K[0,0]
fy = K[1,1]
fovy = 2*arctan(0.5*im_size[1]/fy)*180/pi
aspect = (im_size[0]*fy)/(im_size[1]*fx)
# define the near and far clipping planes
near = 0.1
far = 500000.0
# set perspective
gluPerspective(fovy, aspect, near, far)
glViewport(0, 0, im_size[0], im_size[1])


I estimate the modelview using this function:

def set_modelview_from_camera(rvec, tvec):
glMatrixMode(GL_MODELVIEW)

Rx = array([[1, 0, 0], [0, 0, -1], [0, 1, 0]]) # rotate the teapot
M = eye(4)
M[:3, :3] = dot(Rx, cv2.Rodrigues(rvec)[0])
M[:3, 3] = tvec.T
cv2GlMat = array([[1,0,0,0],[0,-1,0,0],[0,0,-1,0],[0,0,0,1]]) #OpenCV -> OpenGL matrix
M = dot(cv2GlMat, M)
m = M.T.flatten()


The function to draw the background (the real image)

def draw_background(I):
bg_image = Image.fromarray(I)
bg_data = bg_image.tobytes('raw', 'RGBX', 0, -1)

glMatrixMode(GL_MODELVIEW)
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)
# bind the texture
glEnable(GL_TEXTURE_2D)
glBindTexture(GL_TEXTURE_2D, glGenTextures(1))
glTexImage2D(GL_TEXTURE_2D,0,GL_RGBA,bg_image.size[0],bg_image.size[1],0,GL_RGBA,GL_UNSIGNED_BYTE,bg_data)
glTexParameterf(GL_TEXTURE_2D,GL_TEXTURE_MAG_FILTER,GL_NEAREST)
glTexParameterf(GL_TEXTURE_2D,GL_TEXTURE_MIN_FILTER,GL_NEAREST)
# create quad to fill the whole window
glTexCoord2f(0.0,0.0); glVertex3f(-1,-1,-1.0)
glTexCoord2f(1.0,0.0); glVertex3f( 1,-1,-1.0)
glTexCoord2f(1.0,1.0); glVertex3f( 1, 1,-1.0)
glTexCoord2f(0.0,1.0); glVertex3f(-1, 1,-1.0)
glEnd()
# clear the texture
glDeleteTextures(1)


I understand the concepts of Opencv, Opencv Calibration, but I can not cannot put together into OpenGL and show the object appears correctly on the screen.