138 lines
3.8 KiB
Python
138 lines
3.8 KiB
Python
import cv2
|
|
import numpy as np
|
|
|
|
|
|
def gstreamer_pipeline(cam_id=0,
|
|
capture_width=1920,
|
|
capture_height=1080,
|
|
framerate=30,
|
|
format="YUYV",
|
|
):
|
|
"""
|
|
Use libgstreamer to open csi-cameras.
|
|
"""
|
|
return (
|
|
f"v4l2src device=/dev/video{cam_id} ! "
|
|
f"video/x-raw,format={format},width={capture_width},height={capture_height},framerate={framerate}/1 ! "
|
|
"videoconvert ! "
|
|
"video/x-raw,format=YUYV ! " # 转为 OpenCV 能直接用的 BGR 格式
|
|
"appsink"
|
|
|
|
)
|
|
|
|
|
|
def convert_binary_to_bool(mask):
|
|
"""
|
|
Convert a binary image (only one channel and pixels are 0 or 255) to
|
|
a bool one (all pixels are 0 or 1).
|
|
"""
|
|
return (mask.astype(np.float) / 255.0).astype(int)
|
|
|
|
|
|
def adjust_luminance(gray, factor):
|
|
"""
|
|
Adjust the luminance of a grayscale image by a factor.
|
|
"""
|
|
return np.minimum((gray * factor), 255).astype(np.uint8)
|
|
|
|
|
|
def get_mean_statistisc(gray, mask):
|
|
"""
|
|
Get the total values of a gray image in a region defined by a mask matrix.
|
|
The mask matrix must have values either 0 or 1.
|
|
"""
|
|
return np.sum(gray * mask)
|
|
|
|
|
|
def mean_luminance_ratio(grayA, grayB, mask):
|
|
return get_mean_statistisc(grayA, mask) / get_mean_statistisc(grayB, mask)
|
|
|
|
|
|
def get_mask(img):
|
|
"""
|
|
Convert an image to a mask array.
|
|
"""
|
|
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
|
ret, mask = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY)
|
|
return mask
|
|
|
|
|
|
def get_overlap_region_mask(imA, imB):
|
|
"""
|
|
Given two images of the save size, get their overlapping region and
|
|
convert this region to a mask array.
|
|
"""
|
|
overlap = cv2.bitwise_and(imA, imB)
|
|
mask = get_mask(overlap)
|
|
mask = cv2.dilate(mask, np.ones((2, 2), np.uint8), iterations=2)
|
|
return mask
|
|
|
|
|
|
def get_outmost_polygon_boundary(img):
|
|
"""
|
|
Given a mask image with the mask describes the overlapping region of
|
|
two images, get the outmost contour of this region.
|
|
"""
|
|
mask = get_mask(img)
|
|
mask = cv2.dilate(mask, np.ones((2, 2), np.uint8), iterations=2)
|
|
cnts, hierarchy = cv2.findContours(
|
|
mask,
|
|
cv2.RETR_EXTERNAL,
|
|
cv2.CHAIN_APPROX_SIMPLE)[-2:]
|
|
|
|
# get the contour with largest aera
|
|
C = sorted(cnts, key=lambda x: cv2.contourArea(x), reverse=True)[0]
|
|
|
|
# polygon approximation
|
|
polygon = cv2.approxPolyDP(C, 0.009 * cv2.arcLength(C, True), True)
|
|
|
|
return polygon
|
|
|
|
|
|
def get_weight_mask_matrix(imA, imB, dist_threshold=5):
|
|
"""
|
|
Get the weight matrix G that combines two images imA, imB smoothly.
|
|
"""
|
|
overlapMask = get_overlap_region_mask(imA, imB)
|
|
overlapMaskInv = cv2.bitwise_not(overlapMask)
|
|
indices = np.where(overlapMask == 255)
|
|
|
|
imA_diff = cv2.bitwise_and(imA, imA, mask=overlapMaskInv)
|
|
imB_diff = cv2.bitwise_and(imB, imB, mask=overlapMaskInv)
|
|
|
|
G = get_mask(imA).astype(np.float32) / 255.0
|
|
|
|
polyA = get_outmost_polygon_boundary(imA_diff)
|
|
polyB = get_outmost_polygon_boundary(imB_diff)
|
|
|
|
for y, x in zip(*indices):
|
|
# opencv requires a tuple of ints
|
|
xy_tuple = tuple([int(x), int(y)])
|
|
distToB = cv2.pointPolygonTest(polyB, xy_tuple, True)
|
|
|
|
if distToB < dist_threshold:
|
|
distToA = cv2.pointPolygonTest(polyA, xy_tuple, True)
|
|
distToB *= distToB
|
|
distToA *= distToA
|
|
G[y, x] = distToB / (distToA + distToB)
|
|
|
|
return G, overlapMask
|
|
|
|
|
|
def make_white_balance(image):
|
|
"""
|
|
Adjust white balance of an image base on the means of its channels.
|
|
"""
|
|
B, G, R = cv2.split(image)
|
|
m1 = np.mean(B)
|
|
m2 = np.mean(G)
|
|
m3 = np.mean(R)
|
|
K = (m1 + m2 + m3) / 3
|
|
c1 = K / m1
|
|
c2 = K / m2
|
|
c3 = K / m3
|
|
B = adjust_luminance(B, c1)
|
|
G = adjust_luminance(G, c2)
|
|
R = adjust_luminance(R, c3)
|
|
return cv2.merge((B, G, R))
|