The document outlines the development of an augmented reality system designed to recognize iconic landmarks in Oslo, utilizing computer vision techniques such as Harris corner detection and SIFT. It details the data collection process, feature extraction, and the implementation results, showing successful real-time landmark identification with some performance challenges. Future improvements include adopting advanced algorithms and expanding the landmark database for enhanced detection accuracy.

1. DEVELOPING AN AUGMENTED
REALITY LANDMARK
DETECTION SYSTEM FOR
TOURISTS IN OSLO
The goal is to develop a prototype AR system that identifies and recognizes iconic
landmarks in Oslo using local image descriptors. The feature will assist tourists by
providing real-time recognition of landmarks like Vigeland Park, Oslo Opera
House, and Akershus Fortress.Problem: To build a system that can accurately
detect and match landmark features using computer vision techniques such as
Harris Corner Detection and SIFT (Scale-Invariant Feature Transform).

2. DATA COLLECTION AND PREPROCESSING
FOR LANDMARK DETECTION
• Landmarks: Vigeland Park (Monolith and Angry Boy sculptures)Oslo Opera
HouseAkershus FortressImage Collection: Collected 5 images for each landmark
from different angles and under different lighting conditions. Preprocessing:
Converted all images to grayscale and resized them for uniformity before feature
extraction.

3. FEATURE EXTRACTION USING HARRIS
CORNER DETECTION AND SIFT
• Harris Corner Detection: Used for identifying interest points in landmark images. It
works by detecting corners (points of interest) in the image.SIFT (Scale-Invariant
Feature Transform): Extracts feature descriptors that are scale and rotation-invariant,
making them robust for landmark detection across different views.Implementation:
Detected Harris corners, followed by SIFT to extract features.Visualized keypoints on
the images.

4. RESULTS OF LANDMARK DETECTION AND
MATCHING
• Matching: Used SIFT descriptors for matching between reference images and live
input images. Matching Method: Applied ratio test for nearest neighbor matching to
find the best match. Evaluation: Successfully detected keypoints and matched them
with corresponding landmark images. Performance: The system performed well in
detecting and matching landmarks under varied conditions (e.g., different angles
and lighting).

5. CONCLUSIONS AND FUTURE
IMPROVEMENTS
• The prototype successfully implemented landmark detection using Harris Corner Detection
and SIFT. The system identified key landmarks in real-time, proving the feasibility of using
computer vision in AR applications for tourists.
• Challenges: Variations in lighting and weather conditions affected detection
performance.SIFT performed well but struggled in extreme rotations or significant
occlusions.
• Improvements: Incorporate more advanced algorithms like ORB or SURF for better
performance in real-world environments. Extend the database of landmarks to improve
detection accuracy across more landmarks. Optimize the real-time performance for mobile
deployment.