qshub / engine /advanced_cv_extractor.py
basel00omar's picture
Implement systemic CV Auto-Scaling for partial floors/annexes
7f70f7a
Raw History Blame Contribute Delete
4.3 kB
import cv2
import numpy as np
import math
import os
def process_floor_plan_image(img_path):
"""
Advanced Computer Vision Extractor for Architectural Floor Plans with Auto-Coloring.
Optimized for blazing fast execution even on massive A0 blueprints.
"""
if not os.path.exists(img_path):
return {"cv_perimeter_px": 0, "cv_area_px": 0, "cv_door_count": 0, "cv_window_count": 0, "wall_internal_px": 0}
try:
# Load grayscale image for processing
img_gray = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)
if img_gray is None:
return {"cv_perimeter_px": 0, "cv_area_px": 0, "cv_door_count": 0, "cv_window_count": 0, "wall_internal_px": 0}
img_color = cv2.imread(img_path, cv2.IMREAD_COLOR)
# OPTIMIZATION: Resize image if it's too large to prevent OpenCV hangs
h, w = img_gray.shape
max_dim = 1600
scale = 1.0
if max(h, w) > max_dim:
scale = max_dim / max(h, w)
img_gray = cv2.resize(img_gray, (int(w * scale), int(h * scale)))
img_color = cv2.resize(img_color, (int(w * scale), int(h * scale)))
# Invert image (black lines become white)
_, thresh = cv2.threshold(img_gray, 200, 255, cv2.THRESH_BINARY_INV)
# 1. Detect Main Walls and Perimeter
kernel = np.ones((5, 5), np.uint8)
walls = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=1)
# Only take RETR_EXTERNAL for speed instead of full RETR_TREE
contours, _ = cv2.findContours(walls, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
max_area = 0
max_perimeter = 0
largest_cnt = None
if contours:
largest_cnt = max(contours, key=cv2.contourArea)
max_area = cv2.contourArea(largest_cnt)
max_perimeter = cv2.arcLength(largest_cnt, True)
cv2.drawContours(img_color, [largest_cnt], -1, (255, 0, 0), 3) # BGR: Blue
# 2. Detect Doors/Windows (Fast simplified approach)
kernel_small = np.ones((3, 3), np.uint8)
clean = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel_small, iterations=1)
components, _ = cv2.findContours(clean, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
door_count = 0
window_count = 0
# Only process top 200 largest contours to prevent millions of small hatch lines from hanging Python
components = sorted(components, key=cv2.contourArea, reverse=True)[:200]
for comp in components:
comp_area = cv2.contourArea(comp)
if 20 < comp_area < 2000:
rect = cv2.minAreaRect(comp)
width, height = rect[1]
if width > 0 and height > 0:
aspect_ratio = max(width, height) / min(width, height)
rect_area = width * height
solidity = comp_area / rect_area if rect_area > 0 else 0
if 1.0 <= aspect_ratio <= 1.8 and solidity < 0.4:
door_count += 1
box = np.int32(cv2.boxPoints(rect))
cv2.drawContours(img_color, [box], 0, (0, 255, 255), 2)
elif aspect_ratio >= 3.0 and 0.15 <= solidity <= 0.85:
window_count += 1
box = np.int32(cv2.boxPoints(rect))
cv2.drawContours(img_color, [box], 0, (0, 165, 255), 2)
# Save the Auto-Colored Takeoff Markup
markup_path = img_path.replace(".png", "_markup.png")
cv2.imwrite(markup_path, img_color)
markup_filename = os.path.basename(markup_path)
total_internal_px = 0 # To prevent NameError crash
return {
"cv_perimeter_px": max_perimeter,
"cv_area_px": max_area,
"cv_door_count": door_count,
"cv_window_count": window_count,
"wall_internal_px": total_internal_px,
"markup_url": f"/cache/{markup_filename}"
}
except Exception as e:
print(f"CV Extractor error: {e}")
return {"cv_perimeter_px": 0, "cv_area_px": 0, "cv_door_count": 0, "cv_window_count": 0, "wall_internal_px": 0}