Unit
- Definition: Photogrammetry is the science of making measurements from photographs. It extracts reliable 3D spatial information from 2D images by analyzing geometric relationships between overlapping images. Traditionally used in topographic mapping, it now powers many 3D modeling workflows.
- History: The discipline began in the mid-19th century with terrestrial cameras. Aerial photogrammetry matured during World War I for reconnaissance. Digital revolution in the 1990s replaced analog film with digital sensors and automated processing, making the technique accessible to a wider audience.
- Basic Principle: When two photographs capture the same scene from different positions, a point in the real world projects to different pixel locations in each image. By measuring these image coordinates and knowing the camera geometry, we can triangulate the 3D position of the point using the collinearity condition.
- Key Outputs: Photogrammetric processing produces digital elevation models (DEMs), orthophotos (geometrically corrected images), dense point clouds, textured 3D meshes, and vector maps. Each output serves different analysis or visualization needs.
- Comparison with LIDAR: Both produce 3D data but differ in cost, speed, and accuracy. LIDAR is more expensive and emits its own energy, making it usable in darkness. Photogrammetry uses passive sunlight, is cheaper, and captures rich color texture, but may struggle in low-contrast or textureless areas.
- Workflow Overview: A standard photogrammetry workflow includes image acquisition, camera calibration (if not known), feature extraction and matching, orientation (structure-from-motion), dense matching, georeferencing, orthorectification, and finally model export. Each step depends on the quality of previous steps.
- Software and Tools: Commercial software such as Agisoft Metashape, Pix4Dmapper, and RealityCapture dominate. Open-source alternatives include OpenMVG, COLMAP, and Meshroom. Python libraries like OpenCV provide building blocks for custom pipelines.
- Application Domains: Surveying and mapping, archaeology, construction monitoring, industrial inspection, cultural heritage preservation, agriculture, and autonomous vehicle navigation all rely on photogrammetry. Each domain imposes specific requirements on accuracy, resolution, and processing scale.
- Limitations and Challenges: Photogrammetry requires sufficient texture and lighting; flat surfaces, repetitive patterns, or reflective materials can cause matching failures. Atmospheric conditions, motion blur, and occlusions also degrade quality. Additionally, the process is computationally intensive for large datasets and demands careful planning for ground control points to achieve geospatial accuracy.
- Image Acquisition Guidelines: For reliable results, capture images with at least 60 percent frontal overlap and 30 percent sidelap. Maintain consistent lighting, avoid harsh shadows, and use a fixed focal length. Ensure sharp focus and high shutter speed to minimize motion blur. Overlapping coverage from multiple angles helps reconstruct complex geometry.
Code Example
Python
import cv2
import numpy as np
# Read two overlapping images
img1 = cv2.imread('left.jpg')
img2 = cv2.imread('right.jpg')
# Convert to grayscale
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
# Detect and match SIFT features
sift = cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(gray1, None)
kp2, des2 = sift.detectAndCompute(gray2, None)
bf = cv2.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
# Apply ratio test
good = []
for m, n in matches:
if m.distance good.append(m)
print('Number of good matches:', len(good))This snippet demonstrates the first step of any photogrammetry pipeline: detecting and matching features across overlapping images. The ratio test filters out ambiguous matches, preserving only reliable correspondences for subsequent orientation steps.
Quick Revision
- Photogrammetry extracts 3D measurements from overlapping 2D images using geometric principles.
- Digital workflows automate triangulation and produce DEMs, orthophotos, and textured models.
- Key outputs include point clouds, meshes, and maps, each serving specific applications.
- OpenCV provides basic tools for feature detection and matching in a custom pipeline.