Yang Cao
with geoai for building extraction
e383754
raw
history blame
20.1 kB
"""
Geospatial utilities for image processing and GeoJSON generation.
This module adapts techniques from the geoai library for better polygon generation
with simplified dependencies.
"""
import os
import logging
import uuid
import numpy as np
import cv2
from PIL import Image, TiffTags, TiffImagePlugin
import json
import re
from shapely.geometry import Polygon, MultiPolygon, mapping
from shapely import ops
def extract_contours(image_path, min_area=50, epsilon_factor=0.002):
"""
Extract contours from an image and convert them to polygons.
Uses OpenCV's contour detection with douglas-peucker simplification.
Args:
image_path (str): Path to the processed image
min_area (int): Minimum contour area to keep
epsilon_factor (float): Simplification factor for douglas-peucker algorithm
Returns:
list: List of polygon objects
"""
try:
# Read the image
img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
if img is None:
# Try using PIL if OpenCV fails
pil_img = Image.open(image_path).convert('L')
img = np.array(pil_img)
# Apply threshold if needed
_, thresh = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY)
# Find contours
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
polygons = []
for contour in contours:
# Filter small contours
area = cv2.contourArea(contour)
if area < min_area:
continue
# Apply Douglas-Peucker algorithm to simplify contours
epsilon = epsilon_factor * cv2.arcLength(contour, True)
approx = cv2.approxPolyDP(contour, epsilon, True)
# Convert to polygon
if len(approx) >= 3: # At least 3 points needed for a polygon
polygon_points = []
for point in approx:
x, y = point[0]
polygon_points.append((float(x), float(y)))
# Create a valid polygon (close it if needed)
if polygon_points[0] != polygon_points[-1]:
polygon_points.append(polygon_points[0])
# Create shapely polygon
polygon = Polygon(polygon_points)
if polygon.is_valid:
polygons.append(polygon)
return polygons
except Exception as e:
logging.error(f"Error extracting contours: {str(e)}")
return []
def simplify_polygons(polygons, tolerance=1.0):
"""
Apply polygon simplification to reduce the number of vertices.
Args:
polygons (list): List of shapely Polygon objects
tolerance (float): Simplification tolerance
Returns:
list: List of simplified polygons
"""
simplified = []
for polygon in polygons:
# Apply simplification
simp = polygon.simplify(tolerance, preserve_topology=True)
if simp.is_valid and not simp.is_empty:
simplified.append(simp)
return simplified
def regularize_polygons(polygons):
"""
Regularize polygons to make them more rectangular when appropriate.
Args:
polygons (list): List of shapely Polygon objects
Returns:
list: List of regularized polygons
"""
regularized = []
for polygon in polygons:
try:
# Check if the polygon is roughly rectangular using a simple heuristic
bounds = polygon.bounds
width = bounds[2] - bounds[0]
height = bounds[3] - bounds[1]
area_ratio = polygon.area / (width * height)
# If it's at least 80% similar to a rectangle, make it rectangular
if area_ratio > 0.8:
# Replace with the minimum bounding rectangle
minx, miny, maxx, maxy = polygon.bounds
regularized.append(Polygon([
(minx, miny), (maxx, miny),
(maxx, maxy), (minx, maxy), (minx, miny)
]))
else:
regularized.append(polygon)
except Exception as e:
logging.warning(f"Error regularizing polygon: {str(e)}")
regularized.append(polygon)
return regularized
def merge_nearby_polygons(polygons, distance_threshold=5.0):
"""
Merge polygons that are close to each other to reduce the polygon count.
Args:
polygons (list): List of shapely Polygon objects
distance_threshold (float): Distance threshold for merging
Returns:
list: List of merged polygons
"""
if not polygons:
return []
# Buffer polygons slightly to create overlaps for nearby polygons
buffered = [polygon.buffer(distance_threshold) for polygon in polygons]
# Union all buffered polygons
union = ops.unary_union(buffered)
# Convert the result to a list of polygons
if isinstance(union, Polygon):
return [union]
elif isinstance(union, MultiPolygon):
return list(union.geoms)
else:
return []
def extract_geo_coordinates_from_image(image_path):
"""
Extract geographic coordinates from image metadata (EXIF, GeoTIFF).
Uses rasterio for more reliable GeoTIFF handling.
Args:
image_path (str): Path to the image file
Returns:
tuple: (min_lat, min_lon, max_lat, max_lon) or None if not found
"""
try:
# First try using rasterio for GeoTIFF files
if image_path.lower().endswith(('.tif', '.tiff')):
try:
import rasterio
from rasterio.warp import transform_bounds
logging.info(f"Using rasterio to extract coordinates from {image_path}")
with rasterio.open(image_path) as src:
# Check if the file has a valid CRS
if src.crs is not None:
# Get bounds in the source CRS
bounds = src.bounds
# Transform bounds to WGS84 (lat/lon)
if src.crs.to_epsg() != 4326:
west, south, east, north = transform_bounds(
src.crs, 'EPSG:4326',
bounds.left, bounds.bottom, bounds.right, bounds.top
)
else:
west, south, east, north = bounds
logging.info(f"Extracted coordinates from GeoTIFF: {west},{south} to {east},{north}")
return south, west, north, east # min_lat, min_lon, max_lat, max_lon
except Exception as e:
logging.warning(f"Rasterio extraction failed: {str(e)}, falling back to PIL")
# Fallback to PIL for other image types or if rasterio fails
img = Image.open(image_path)
# Check if it's a TIFF image with geospatial data
if hasattr(img, 'tag') and img.tag:
logging.info(f"Detected image with tags, checking for geospatial metadata")
# Try to extract ModelPixelScaleTag (33550) and ModelTiepointTag (33922)
pixel_scale_tag = None
tiepoint_tag = None
# Check for tags
tag_dict = img.tag.items() if hasattr(img.tag, 'items') else {}
# Remove hardcoded Brazil detection
is_brazil_image = False
if not tag_dict and is_brazil_image:
logging.info(f"Special case for Brazil image detected in: {image_path}")
# Hard code Brazil coordinates for the specific sample
# These coordinates are for the Brazil sample from the GeoAI notebook
# Rio de Janeiro area (near Tijuca Forest)
min_lat = -22.96 # Southern Brazil
min_lon = -43.38
max_lat = -22.94
max_lon = -43.36
logging.info(f"Using known Brazil coordinates: {min_lon},{min_lat} to {max_lon},{max_lat}")
return min_lat, min_lon, max_lat, max_lon
for tag_id, value in tag_dict:
tag_name = TiffTags.TAGS.get(tag_id, str(tag_id))
logging.debug(f"TIFF tag: {tag_name} ({tag_id}): {value}")
if tag_id == 33550: # ModelPixelScaleTag
pixel_scale_tag = value
elif tag_id == 33922: # ModelTiepointTag
tiepoint_tag = value
# Supplementary check for the log output we can see (raw detection)
# Look for any GeoTIFF tag indicators in the output
geotiff_indicators = ['ModelPixelScale', 'ModelTiepoint', 'GeoKey', 'GeoAscii']
has_geotiff_indicators = False
for indicator in geotiff_indicators:
if indicator in str(img.tag):
has_geotiff_indicators = True
logging.info(f"Found GeoTIFF indicator: {indicator}")
break
# Look for any TIFF tag containing geographic info
log_pattern = r"ModelPixelScaleTag.*?value: b'(.*?)'"
log_matches = re.findall(log_pattern, str(img.tag))
# If we detect any GeoTIFF indicators or raw tags, consider it a Brazil image
if (log_matches or has_geotiff_indicators) and not pixel_scale_tag:
logging.info(f"GeoTIFF indicators detected in image")
# Remove hardcoded Brazil coordinates
# Try to extract values from raw tag data if possible
try:
# Parse the modelPixelScale if available
if log_matches:
logging.info(f"Found raw pixel scale data: {log_matches[0]}")
# We'll continue with the standard TIFF tag processing below
except Exception as e:
logging.error(f"Error parsing raw tag data: {str(e)}")
if pixel_scale_tag and tiepoint_tag:
# Extract pixel scale (x, y)
x_scale = float(pixel_scale_tag[0])
y_scale = float(pixel_scale_tag[1])
# Extract model tiepoint (raster origin)
i, j, k = float(tiepoint_tag[0]), float(tiepoint_tag[1]), float(tiepoint_tag[2])
x, y, z = float(tiepoint_tag[3]), float(tiepoint_tag[4]), float(tiepoint_tag[5])
# Calculate bounds based on image dimensions
width, height = img.size
# Calculate bounds
min_lon = x
max_lat = y
max_lon = x + width * x_scale
min_lat = y - height * y_scale
logging.info(f"Extracted geo bounds: {min_lon},{min_lat} to {max_lon},{max_lat}")
return min_lat, min_lon, max_lat, max_lon
logging.info("No valid geospatial metadata found in TIFF")
# Check for EXIF GPS data (typically in JPEG)
elif hasattr(img, '_getexif') and img._getexif():
exif = img._getexif()
if exif and 34853 in exif: # 34853 is the GPS Info tag
gps_info = exif[34853]
# Extract GPS data
if 1 in gps_info and 2 in gps_info and 3 in gps_info and 4 in gps_info:
# Latitude
lat_ref = gps_info[1] # 'N' or 'S'
lat = gps_info[2] # ((deg_num, deg_denom), (min_num, min_denom), (sec_num, sec_denom))
lat_val = lat[0][0]/lat[0][1] + lat[1][0]/(lat[1][1]*60) + lat[2][0]/(lat[2][1]*3600)
if lat_ref == 'S':
lat_val = -lat_val
# Longitude
lon_ref = gps_info[3] # 'E' or 'W'
lon = gps_info[4]
lon_val = lon[0][0]/lon[0][1] + lon[1][0]/(lon[1][1]*60) + lon[2][0]/(lon[2][1]*3600)
if lon_ref == 'W':
lon_val = -lon_val
# Create a small region around the point
delta = 0.01 # ~1km at the equator
min_lat = lat_val - delta
min_lon = lon_val - delta
max_lat = lat_val + delta
max_lon = lon_val + delta
logging.info(f"Extracted EXIF GPS bounds: {min_lon},{min_lat} to {max_lon},{max_lat}")
return min_lat, min_lon, max_lat, max_lon
logging.info("No valid GPS metadata found in EXIF")
# If we get here, we couldn't extract coordinates
logging.warning("Could not extract geospatial coordinates from image")
return None
except Exception as e:
logging.error(f"Error extracting geo coordinates: {str(e)}")
return None
def convert_to_geojson_with_transform(polygons, image_height, image_width,
min_lat=None, min_lon=None, max_lat=None, max_lon=None):
"""
Convert polygons to GeoJSON with proper geographic transformation.
Args:
polygons (list): List of shapely Polygon objects
image_height (int): Height of the source image
image_width (int): Width of the source image
min_lat (float, optional): Minimum latitude for geographic bounds
min_lon (float, optional): Minimum longitude for geographic bounds
max_lat (float, optional): Maximum latitude for geographic bounds
max_lon (float, optional): Maximum longitude for geographic bounds
Returns:
dict: GeoJSON object
"""
# Set default geographic bounds if not provided
if None in (min_lon, min_lat, max_lon, max_lat):
logging.warning("No geographic coordinates provided for GeoJSON transformation. Using default values.")
# Default to somewhere neutral (not in New York)
min_lon, min_lat = -98.0, 32.0 # Central US
max_lon, max_lat = -96.0, 34.0
# Create a GeoJSON feature collection
geojson = {
"type": "FeatureCollection",
"features": []
}
# Function to transform pixel coordinates to geographic coordinates
def transform_point(x, y):
# Linear interpolation
lon = min_lon + (x / image_width) * (max_lon - min_lon)
# Invert y-axis for geographic coordinates
lat = max_lat - (y / image_height) * (max_lat - min_lat)
return lon, lat
# Convert each polygon to a GeoJSON feature
for i, polygon in enumerate(polygons):
# Extract coordinates
coords = list(polygon.exterior.coords)
# Transform coordinates to geographic space
geo_coords = [transform_point(x, y) for x, y in coords]
# Create GeoJSON geometry
geometry = {
"type": "Polygon",
"coordinates": [geo_coords]
}
# Create GeoJSON feature
feature = {
"type": "Feature",
"id": i + 1,
"properties": {
"name": f"Feature {i+1}"
},
"geometry": geometry
}
geojson["features"].append(feature)
return geojson
def process_image_to_geojson(image_path, feature_type="buildings", original_file_path=None):
"""
Complete pipeline to convert an image to a simplified GeoJSON.
Args:
image_path (str): Path to the processed image
feature_type (str): Type of features to extract ("buildings", "trees", "water", "roads")
original_file_path (str, optional): Path to the original uploaded file
Returns:
dict: GeoJSON object
"""
try:
# Open image to get dimensions
img = Image.open(image_path)
width, height = img.size
# Import segmentation module here to avoid circular imports
from utils.segmentation import segment_and_extract_features
# Extract features using advanced segmentation
_, polygons = segment_and_extract_features(
image_path,
output_mask_path=None,
feature_type=feature_type,
min_area=50,
simplify_tolerance=2.0,
merge_distance=5.0
)
if not polygons:
logging.warning("No polygons found in the image after segmentation")
return {"type": "FeatureCollection", "features": []}
# Use the provided original file path if available
original_image_path = original_file_path
# If no original file path was provided, try to find it
if not original_image_path and "_processed" in image_path:
original_image_path = image_path.replace("_processed", "")
# Try the original image path but replace the extension with common formats
if not os.path.exists(original_image_path):
base_path = original_image_path.rsplit('.', 1)[0]
for ext in ['.tif', '.tiff', '.jpg', '.jpeg', '.png']:
if os.path.exists(base_path + ext):
original_image_path = base_path + ext
break
logging.info(f"Using original image path: {original_image_path}")
# Extract bounds from image if possible
coords = None
if original_image_path and os.path.exists(original_image_path):
logging.info(f"Checking original image for geospatial data: {original_image_path}")
coords = extract_geo_coordinates_from_image(original_image_path)
if not coords:
logging.info("Checking processed image for geospatial data")
coords = extract_geo_coordinates_from_image(image_path)
# Use extracted coordinates or defaults
if coords:
min_lat, min_lon, max_lat, max_lon = coords
logging.info(f"Using extracted coordinates: {min_lon},{min_lat} to {max_lon},{max_lat}")
else:
# Try one more time with rasterio directly on the original image if it exists
if original_image_path and os.path.exists(original_image_path) and original_image_path.lower().endswith(('.tif', '.tiff')):
try:
import rasterio
from rasterio.warp import transform_bounds
with rasterio.open(original_image_path) as src:
if src.crs is not None:
bounds = src.bounds
if src.crs.to_epsg() != 4326:
west, south, east, north = transform_bounds(
src.crs, 'EPSG:4326',
bounds.left, bounds.bottom, bounds.right, bounds.top
)
else:
west, south, east, north = bounds
min_lat, min_lon, max_lat, max_lon = south, west, north, east
logging.info(f"Using coordinates from rasterio: {min_lon},{min_lat} to {max_lon},{max_lat}")
except Exception as e:
logging.warning(f"Failed to extract coordinates with rasterio: {str(e)}")
logging.warning("No coordinates found in image, using default location in Central US")
min_lat, min_lon = 32.0, -98.0 # Central US
max_lat, max_lon = 34.0, -96.0
else:
logging.warning("No coordinates found in image, using default location in Central US")
min_lat, min_lon = 32.0, -98.0 # Central US
max_lat, max_lon = 34.0, -96.0
# Convert to GeoJSON with proper transformation
geojson = convert_to_geojson_with_transform(
polygons, height, width,
min_lat=min_lat, min_lon=min_lon,
max_lat=max_lat, max_lon=max_lon
)
return geojson
except Exception as e:
logging.error(f"Error in GeoJSON processing: {str(e)}")
return {"type": "FeatureCollection", "features": []}