POI Detector and Validator

Drug Store Detection
Restaurant Detection
Library Detection

📖 About the Project

💡 Inspiration

This project was created as part of an initiative with HERE Technologies, who provided access to a rich geospatial database of roads and Points of Interest (POIs). The complexity of validating spatial data—especially when POIs are located near multi-digitized roads—inspired us to explore a hybrid solution that leverages both geospatial analysis and computer vision.

🧠 What I Learned

Deepened my understanding of spatial relationships using libraries like GeoPandas and Shapely.
Gained experience with HERE's APIs for retrieving satellite imagery.
Learned how to apply YOLO (You Only Look Once) object detection to satellite images.
Improved my skills in Python scripting, data cleaning, and coordinate-based logic.

🛠️ How We Built It

Used Python as the primary programming language.
Implemented spatial validation using geopandas, pandas, BallTree (from sklearn.neighbors), and shapely.geometry.
Integrated HERE’s API to fetch satellite imagery for further inspection of flagged POIs.
Applied YOLO (from ultralytics) for image-based object detection on the satellite views.
Managed image preprocessing with cv2, PIL.Image, matplotlib, and utility scripts.

🚧 Challenges Faced

Accurately modeling the spatial logic required to detect misplaced POIs near complex, multi-digitized roads.
Balancing false positives and false negatives in image detection using YOLO.
Handling varied coordinate systems and making sure spatial joins were consistent and precise.
Ensuring file paths and dependencies were flexible enough to support different environments.