Panorama Stitcher

Create panoramic images by stitching multiple photos.

View Source Code

Overview

Stitch multiple overlapping images into a seamless panorama, like smartphone panorama mode.

Key Techniques:

  • Feature detection (ORB/SIFT)
  • Feature matching
  • Homography estimation (RANSAC)
  • Image warping and blending

Pipeline 

Images → Detect Features → Match → Homography → Warp → Blend → Panorama
   ↓           ↓            ↓          ↓          ↓       ↓        ↓
[Multiple]  [Keypoints]  [Pairs]    [Matrix]   [Align]  [Smooth] [Single]

Key OpenCV Functions

Feature Detection 

# ORB (fast, free)
detector = cv2.ORB_create(nfeatures=2000)
kp1, des1 = detector.detectAndCompute(gray1, None)
kp2, des2 = detector.detectAndCompute(gray2, None)

# SIFT (slower, more accurate) - requires opencv-contrib
detector = cv2.SIFT_create()

Feature Matching 

# For ORB (Hamming distance)
matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = matcher.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)

# For SIFT (L2 distance with ratio test)
matcher = cv2.BFMatcher(cv2.NORM_L2)
matches = matcher.knnMatch(des1, des2, k=2)
good_matches = [m for m, n in matches if m.distance < 0.75 * n.distance]

Homography Estimation 

# Extract matched point coordinates
src_pts = np.float32([kp1[m.queryIdx].pt for m in matches]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in matches]).reshape(-1, 1, 2)

# Find homography with RANSAC
H, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)

Warping and Blending 

# Calculate output canvas size
corners = cv2.perspectiveTransform(corners1, H)
# ... calculate bounds ...

# Warp first image
warped1 = cv2.warpPerspective(img1, H_translated, (output_w, output_h))

# Place second image
warped2 = np.zeros((output_h, output_w, 3), dtype=np.uint8)
warped2[offset_y:offset_y+h2, offset_x:offset_x+w2] = img2

# Blend in overlap region
mask1 = (warped1.sum(axis=2) > 0).astype(float)
mask2 = (warped2.sum(axis=2) > 0).astype(float)
overlap = mask1 * mask2

result = np.where(overlap[:,:,np.newaxis] > 0,
                  (warped1.astype(float) + warped2.astype(float)) / 2,
                  np.maximum(warped1, warped2)).astype(np.uint8)

OpenCV Stitcher Class

For easier use, OpenCV provides a built-in Stitcher:

stitcher = cv2.Stitcher_create(cv2.Stitcher_PANORAMA)
status, panorama = stitcher.stitch(images)

if status == cv2.Stitcher_OK:
    cv2.imshow("Panorama", panorama)
else:
    print(f"Stitching failed: {status}")

Status codes:

  • cv2.Stitcher_OK - Success
  • cv2.Stitcher_ERR_NEED_MORE_IMGS - Not enough images
  • cv2.Stitcher_ERR_HOMOGRAPHY_EST_FAIL - Matching failed

Tips for Better Panoramas

  1. Overlap: 30-50% overlap between images
  2. Rotation: Rotate camera, don’t translate
  3. Exposure: Keep exposure consistent
  4. Static scenes: Moving objects cause artifacts
  5. Features: Include textured areas, avoid blank walls

Controls

Key Action
c Capture frame (webcam mode)
s Stitch captured images
o Use OpenCV Stitcher
r Reset/clear images
f Toggle ORB/SIFT
q Quit

Running the Application 

python curriculum/applications/14_panorama_stitcher.py

Official Documentation