qshub / pdf_engine /element_detector.py
basel00omar's picture
THE QS HUB v2.0 - Complete React + FastAPI Platform (clean deploy)
1b4b6d2
Raw History Blame Contribute Delete
7.97 kB
"""
كشف العناصر الإنشائية والمعمارية من المخططات
يستخدم OpenCV بدون AI — معالجة بصرية بحتة مع تحسين الفلترة وإزالة الشبكات
"""
import cv2
import numpy as np
from typing import List, Dict
def preprocess_drawing(page_array: np.ndarray) -> np.ndarray:
"""
Preprocesses the drawing image:
1. Converts to grayscale.
2. Identifies if it is dark mode and normalizes it to a white background.
3. Removes grid lines using morphological opening.
4. Enhances high-contrast line features.
"""
gray = cv2.cvtColor(page_array, cv2.COLOR_BGR2GRAY)
# Check if background is dark (mean pixel value < 127)
is_dark = np.mean(gray) < 127
if is_dark:
gray = cv2.bitwise_not(gray) # Invert to standard black-on-white layout
h, w = gray.shape
_, thresh = cv2.threshold(gray, 200, 255, cv2.THRESH_BINARY_INV)
# Morphological grid line detection (long thin grids)
grid_h = max(int(h * 0.04), 10)
grid_w = max(int(w * 0.04), 10)
ver_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, grid_h))
hor_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (grid_w, 1))
ver_lines = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, ver_kernel, iterations=1)
hor_lines = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, hor_kernel, iterations=1)
# Combine grids
grids = cv2.add(ver_lines, hor_lines)
# Remove grids by setting grid pixels to background (white)
cleaned = gray.copy()
cleaned[grids > 200] = 255
# Morphological closing to heal small text elements and drawing gaps
kernel = np.ones((2, 2), np.uint8)
cleaned = cv2.morphologyEx(cleaned, cv2.MORPH_CLOSE, kernel)
return cleaned
def detect_columns(
page_array: np.ndarray,
pixels_per_meter: float,
min_size_m: float = 0.20,
max_size_m: float = 1.50,
) -> List[Dict]:
"""يكشف الأعمدة كمستطيلات صغيرة مملوءة في المخطط"""
if pixels_per_meter <= 0:
return []
min_px = int(min_size_m * pixels_per_meter * 0.5)
max_px = int(max_size_m * pixels_per_meter * 2.0)
# Use preprocessed clean image
gray = preprocess_drawing(page_array)
_, binary = cv2.threshold(gray, 80, 255, cv2.THRESH_BINARY_INV)
contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
columns = []
for cnt in contours:
x, y, w, h = cv2.boundingRect(cnt)
if not (min_px <= w <= max_px and min_px <= h <= max_px):
continue
aspect = max(w, h) / max(min(w, h), 1)
if aspect > 3.0:
continue
roi = binary[y:y+h, x:x+w]
fill_ratio = np.sum(roi > 0) / (w * h)
if fill_ratio < 0.60:
continue
columns.append({
"x": x, "y": y, "w": w, "h": h,
"width_m": round(w / pixels_per_meter, 3),
"height_m": round(h / pixels_per_meter, 3),
"area_m2": round((w / pixels_per_meter) * (h / pixels_per_meter), 4),
"center_x": x + w // 2,
"center_y": y + h // 2,
})
return columns
def detect_walls(page_array: np.ndarray, pixels_per_meter: float) -> List[Dict]:
"""يكشف الجدران كخطوط سميكة في المخطط بعد إزالة شبكة الإحداثيات"""
if pixels_per_meter <= 0:
return []
# Use preprocessed clean image
gray = preprocess_drawing(page_array)
edges = cv2.Canny(gray, 50, 150)
min_wall_length_px = int(0.5 * pixels_per_meter)
lines = cv2.HoughLinesP(
edges, 1, np.pi / 180,
threshold=100,
minLineLength=min_wall_length_px,
maxLineGap=20,
)
if lines is None:
return []
walls = []
for line in lines:
x1, y1, x2, y2 = line[0]
length_px = ((x2 - x1) ** 2 + (y2 - y1) ** 2) ** 0.5
angle = np.degrees(np.arctan2(abs(y2 - y1), abs(x2 - x1)))
direction = "horizontal" if angle < 15 else "vertical" if angle > 75 else "diagonal"
walls.append({
"x1": x1, "y1": y1, "x2": x2, "y2": y2,
"length_m": round(length_px / pixels_per_meter, 3),
"direction": direction,
"angle_deg": round(angle, 1),
})
return walls
def detect_rooms(page_array: np.ndarray, pixels_per_meter: float) -> List[Dict]:
"""يكشف الغرف كمناطق مغلقة في مخطط الطابق"""
if pixels_per_meter <= 0:
return []
# Use preprocessed clean image
gray = preprocess_drawing(page_array)
kernel = np.ones((3, 3), np.uint8)
_, binary = cv2.threshold(gray, 200, 255, cv2.THRESH_BINARY_INV)
dilated = cv2.dilate(binary, kernel, iterations=2)
contours, _ = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
min_area_px = int((1.0 * pixels_per_meter) ** 2)
max_area_px = int((30.0 * pixels_per_meter) ** 2)
rooms = []
for cnt in contours:
area_px = cv2.contourArea(cnt)
if not (min_area_px <= area_px <= max_area_px):
continue
x, y, w, h = cv2.boundingRect(cnt)
area_m2 = round(area_px / (pixels_per_meter ** 2), 2)
perim_m = round(cv2.arcLength(cnt, True) / pixels_per_meter, 2)
rooms.append({
"x": x, "y": y, "w": w, "h": h,
"area_m2": area_m2,
"perim_m": perim_m,
"width_m": round(w / pixels_per_meter, 2),
"height_m": round(h / pixels_per_meter, 2),
"center_x": x + w // 2,
"center_y": y + h // 2,
})
rooms.sort(key=lambda r: r["area_m2"], reverse=True)
return rooms
def draw_detections(
page_array: np.ndarray,
columns: List[Dict],
walls: List[Dict],
rooms: List[Dict],
) -> np.ndarray:
"""يرسم كل العناصر المكتشفة على الصورة"""
img = page_array.copy()
for room in rooms:
cv2.rectangle(img, (room["x"], room["y"]),
(room["x"] + room["w"], room["y"] + room["h"]),
(255, 150, 0), 2)
cv2.putText(img, f"{room['area_m2']}m2",
(room["x"] + 5, room["y"] + 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 100, 0), 1)
for wall in walls:
cv2.line(img, (wall["x1"], wall["y1"]), (wall["x2"], wall["y2"]),
(0, 200, 0), 2)
for col in columns:
cv2.rectangle(img, (col["x"], col["y"]),
(col["x"] + col["w"], col["y"] + col["h"]),
(0, 0, 255), 3)
cv2.putText(img, f"{col['width_m']}x{col['height_m']}",
(col["x"], col["y"] - 5),
cv2.FONT_HERSHEY_SIMPLEX, 0.4, (0, 0, 200), 1)
return img
def calculate_detection_summary(
columns: List[Dict],
walls: List[Dict],
rooms: List[Dict],
) -> Dict:
"""يلخص نتائج الكشف في أرقام جاهزة للـ QTO engine"""
total_wall_h = sum(w["length_m"] for w in walls if w["direction"] == "horizontal")
total_wall_v = sum(w["length_m"] for w in walls if w["direction"] == "vertical")
total_area = sum(r["area_m2"] for r in rooms)
total_perim = sum(r["perim_m"] for r in rooms)
col_sizes: Dict[str, int] = {}
for col in columns:
key = f"{col['width_m']}x{col['height_m']}"
col_sizes[key] = col_sizes.get(key, 0) + 1
return {
"column_count": len(columns),
"column_schedule": col_sizes,
"wall_h_total_m": round(total_wall_h, 2),
"wall_v_total_m": round(total_wall_v, 2),
"total_wall_m": round(total_wall_h + total_wall_v, 2),
"room_count": len(rooms),
"total_area_m2": round(total_area, 2),
"total_perim_m": round(total_perim, 2),
"rooms": rooms,
}