Introduction

The Raspberry Turk uses a Raspberry Pi camera module and computer vision algorithms to determine which pieces are in which squares. In order to see the board properly, it needs to be calibrated first.

Warp Perspective

From chess_camera.py

def current_chessboard_frame(self):
    _, frame = self._capture.read()
    h, w = frame.shape[:2]
    M = get_chessboard_perspective_transform()
    bgr_img = cv2.warpPerspective(frame, M, (BOARD_SIZE,BOARD_SIZE))
    img = cv2.cvtColor(bgr_img, cv2.COLOR_BGR2RGB)
    flipped_img = cv2.flip(img, -1)
    return ChessboardFrame(flipped_img)

In order to go from A to B, cv2.warpPerspective from OpenCV is used.

Finding the Transformation Matrix

There is a small script used to find the correct transformation matrix. The full source can be found here.

Essentially it does the following:

1. Find the inner chessboard corners using cv2.findChessboardCorners
2. Ensure it found the chessboard and the camera is centered
3. Extrapolate and find the outer four corners of the board
4. Map the actual four corners of the board to the corners of the new image and create the perspective matrix using cv2.getPerspectiveTransform
5. Save the transformation matrix for later use

If the camera moves or the robot is deconstructed and reconstructed the perspective transformation matrix can be easily recalculated.

$ python -m raspberryturk.embedded.vision.chessboard_perspective_transform
2017-02-11 16:05:41,371 - __main__ - INFO - Begin camera position recalibration...
2017-02-11 16:05:41,371 - __main__ - INFO - Camera position recalibration successful.