Ask Your Question
0

python openCV - Assertion failed (npoints >= 0 && (depth == CV_32F || depth == CV_64F)) in projectPoints

asked 2017-08-07 04:23:36 -0500

linengmiao gravatar image

I am trying to run the code you see below based on this C++ code. The goal of this piece of code is to know the orientation of ones head and draw a line from his nose to some arbitrary plane. The function fails when trying to execute projectPoints(). When trying to run this code I get the following error and output:

nose end point: [    0.     0.  1000.]
rotVector: [[-3.04843683]
 [ 0.03318151]
 [ 0.31324333]]
translVector: [[  91.013357  ]
 [  -4.95273482]
 [ 401.22675029]]
camera matrix: [[ 320.    0.  320.]
 [   0.  320.  240.]
 [   0.    0.    1.]]
dist coeffs: [ 0.  0.  0.  0.]
noseEndPoint2d: [ 0.  0.]
OpenCV Error: Assertion failed (npoints >= 0 && (depth == CV_32F || depth == CV_64F)) in projectPoints, file /home/John/opencv2/opencv-2.4.11/modules/calib3d/src/calibration.cpp, line 3349
Traceback (most recent call last):
  File "./myHeadOrientation.py", line 62, in <module>
    cv2.projectPoints(noseEndPoint3d, rotVector, translVector, cameraMatrix, distCoeffs, noseEndPoint2d)
cv2.error: /home/John/opencv2/opencv-2.4.11/modules/calib3d/src/calibration.cpp:3349: error: (-215) npoints >= 0 && (depth == CV_32F || depth == CV_64F) in function projectPoints

This is my code:

#! /usr/bin/env python2

import cv2
import alignDlib
import openface

import numpy as np

def getModelPoints():
    modelPoints = np.array([[0,0,0],[0, -330, -65],[-225, 170, -135],[225, 170, -135],[-150, -150, -125],[150, -150, -125]], dtype=np.float)

    return modelPoints


if __name__=='__main__':


    #camera setup
    video_capture = cv2.VideoCapture(0)
    video_capture.set(3, 640)
    video_capture.set(4, 480)

    #camera intrinsics
    imgCols = 640
    focalLength = imgCols/2
    cameraMatrix = np.array([[focalLength, 0, 640/2],[0, focalLength, 480/2],[0, 0, 1]], dtype=np.float)
#    distCoeffs = np.array([[0],[0],[0],[0]], dtype=np.float)
    distCoeffs = np.array([0, 0, 0, 0], dtype=np.float)

    rotVector = np.array([0], dtype=np.float)
    translVector = np.array([0], dtype=np.float)
    noseEndPoint3d = np.array([0, 0, 1000], dtype=np.float)
    noseEndPoint2d = np.array([0, 0], dtype=np.float)

    pathLandMartExtr = '../models/dlib/shape_predictor_68_face_landmarks.dat'
    align = openface.AlignDlib(pathLandMartExtr)

    while(True):

        ret, frameRGB = video_capture.read()
        bb = align.getLargestFaceBoundingBox(frameRGB)

        landmarkLoc = align.findLandmarks(frameRGB, bb)
        cv2.circle(frameRGB, landmarkLoc[30], 2, (0,0,255), 1,8,0) #nose
        cv2.circle(frameRGB, landmarkLoc[8], 2, (0,0,255), 1,8,0) #chin
        cv2.circle(frameRGB, landmarkLoc[36], 2, (0,0,255), 1,8,0) #left eye
        cv2.circle(frameRGB, landmarkLoc[45], 2, (0,0,255), 1,8,0) #right eye
        cv2.circle(frameRGB, landmarkLoc[48], 2, (0,0,255), 1,8,0) #left mouth corner
        cv2.circle(frameRGB, landmarkLoc[54], 2, (0,0,255), 1,8,0) #right mouth corner

        twoDImagePoints = np.array([ [landmarkLoc[30]], [landmarkLoc[8]], [landmarkLoc[36]],[landmarkLoc[45]], [landmarkLoc[48]], [landmarkLoc[54]]], dtype=np.float)
        modelPoints = getModelPoints()
        retval, rotVector, translVector = cv2.solvePnP(modelPoints, twoDImagePoints, cameraMatrix, distCoeffs, rotVector, translVector)
        print('nose end point: '+ str(noseEndPoint3d))
        print('rotVector: '+ str(rotVector))
        print('translVector: '+ str(translVector))
        print('camera matrix: ' + str(cameraMatrix))
        print('dist coeffs: ' + str(distCoeffs))
        print('noseEndPoint2d: ' +str(noseEndPoint2d))


        cv2.projectPoints(noseEndPoint3d, rotVector, translVector, cameraMatrix, distCoeffs, noseEndPoint2d)
        cv2.line(frameRGB, landmarkLoc[30], noseEndPoint2d, (255,0,0), 2)
        cv2.imshow('videostream', frameRGB ...
(more)
edit retag flag offensive close merge delete

Comments

it expects the equiv. of a vector<Point3f> as objPoints in projectPoints(), your noseEndPoint3d somewhat looks like a single point ?

berak gravatar imageberak ( 2017-08-07 05:46:54 -0500 )edit

@berak OK they may expect a vector, but vectors don't exist in Python, they use lists. I used a list (which is afaik the closest equivalent). What else do you suggest me?

linengmiao gravatar imagelinengmiao ( 2017-08-07 06:34:00 -0500 )edit

what about: cv2.projectPoints( [noseEndPoint3d],

berak gravatar imageberak ( 2017-08-07 06:50:47 -0500 )edit

@berak , I added that (without modifying anything else) and got: TypeError: objectPoints is not a numpy array, neither a scalar. objectPoints as you can see in the cpp function definition is the name of the first parameter

linengmiao gravatar imagelinengmiao ( 2017-08-07 06:54:14 -0500 )edit

a numpy array then. just needs one dim more, than you have now.

berak gravatar imageberak ( 2017-08-07 07:03:57 -0500 )edit

@berak as you can see in the main post I already used a numpy array! noseEndPoint3d = np.array([0, 0, 1000], dtype=np.float)

And that doesn't work

linengmiao gravatar imagelinengmiao ( 2017-08-07 07:05:53 -0500 )edit

noseEndPoint3d = np.array([[0, 0, 1000]], dtype=np.float)

berak gravatar imageberak ( 2017-08-07 07:09:24 -0500 )edit

1 answer

Sort by ยป oldest newest most voted
0

answered 2019-07-11 12:16:23 -0500

use dtype = np.float64

That solved my issue !

edit flag offensive delete link more
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2017-08-07 04:23:36 -0500

Seen: 12,057 times

Last updated: Aug 07 '17