class MaskedAnnotation(Annotation):
def __init__(self, label, score, box, mask=None):
super(MaskedAnnotation, self).__init__(label, score, box)
self.mask = Mask(mask if mask is not None else [])
def crop(self, image_np):
transparent = self.mask.as_transparency(image_np)
return self.rect.crop(transparent)
def draw(self, image_np, color, draw_label=True, draw_rect=True):
self.mask.draw(image_np, color)
if draw_rect:
self.rect.draw(image_np, color)
@classmethod
def from_results(cls, num_detections, labels, scores, boxes, masks=None, min_confidence=0.5):
masks = masks if masks is not None else []
annotations = []
zipped = zip(labels[:num_detections], scores[:num_detections], boxes[:num_detections], masks[:num_detections])
for i, (label, score, box, mask) in enumerate(zipped):
if score >= min_confidence:
annotations.append(cls(label, score, box, mask))
return annotations
def __str__(self):
return "{} :: {}".format(self.label, self.score)
def apply(self, image_np):
gray = cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY)
edges = cv2.Sobel(gray, cv2.CV_64F, 0, 1, ksize=3)
ret, thresh = cv2.threshold(edges, 8, 255, cv2.THRESH_BINARY_INV)
opened = cv2.morphologyEx(thresh.astype(np.uint8), cv2.MORPH_OPEN,
(3, 3))
mask = Mask(opened)
return mask.apply(image_np)
def apply(self, image_np):
self.subtractor.apply(image_np)
motion_mask = self.subtractor.apply(image_np)
motion_mask = cv2.dilate(motion_mask, self.kernel, iterations=20)
motion_mask = cv2.erode(motion_mask, self.kernel, iterations=20)
self.motion_mask = Mask(np.logical_not(motion_mask))
return self.motion_mask.apply(image_np)
def apply(self, image_np):
gray = cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY)
edges = cv2.Laplacian(gray, cv2.CV_64F)
thresh = cv2.threshold(edges, 8, 255, cv2.THRESH_BINARY_INV)[1].astype(np.uint8)
thresh = cv2.erode(thresh, (1, 1), iterations=1)
mask = Mask(thresh)
return mask.apply(
self.quant.apply(
image_np
# self.bilateral_filter.apply(image_np)
)
)
class BackgroundSubtractor(ImageHandler):
def __init__(self):
""" Removes background (stationary) elements from an image, accumulating information across multiple images.
"""
self.annotations = None
self.subtractor = cv2.createBackgroundSubtractorKNN(history=1000)
self.kernel = np.ones((3, 3), np.uint8)
def apply(self, image_np):
self.subtractor.apply(image_np)
motion_mask = self.subtractor.apply(image_np)
motion_mask = cv2.dilate(motion_mask, self.kernel, iterations=20)
motion_mask = cv2.erode(motion_mask, self.kernel, iterations=20)
self.motion_mask = Mask(motion_mask)
return self.motion_mask.apply(image_np)
def __init__(self, label, score, box, mask=None):
super(MaskedAnnotation, self).__init__(label, score, box)
self.mask = Mask(mask if mask is not None else [])
def apply(self, image_np):
edges = cv2.Canny(cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY), 100, 200)
ret, thresh = cv2.threshold(edges, 127, 255, cv2.THRESH_BINARY_INV)
mask = Mask(np.squeeze(thresh))
return mask.apply(image_np)
def apply(self, image_np):
gray = cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 127, 255, self.mode)
mask = Mask(thresh)
return mask.apply(image_np)