Geospatial Building Orientation Detection

Key Metrics

11M+

Building Footprints

95%

Coverage Rate (CA/TX)

95%

Memory Reduction

<1s

Query Time

Project Overview

Built a high-performance Python geospatial pipeline that processes over 11 million building footprints across California and Texas to determine the front-facing orientation of houses. The system combines geometric analysis, spatial indexing and street network data to resolve the critical challenge of identifying which side of a building faces the street.

The solution employs a hybrid approach using both OpenStreetMap and Microsoft Buildings datasets, implementing streaming JSON parsing with ijson to achieve 95% memory reduction and sub-second query times. The system intelligently handles edge cases including corner lots with 180° orientation ambiguities through advanced street name matching and geometric edge analysis.

Core Algorithm

1 Geocode address to latitude/longitude coordinates using Google Maps API

2 Stream through state GeoJSON files, filter buildings using bounding box

3 Query Overpass API for nearby streets with weighted road classification

4 Extract and match street name from address against OSM street data

5 Analyze building edges, calculate perpendicular normal vectors

6 Select outward-facing edge aligned with matched street bearing

7 Return orientation angle (0-360°) as building front direction

Technologies Used

Python 3.8+ Shapely R-tree Indexing ijson (Streaming) Google Maps API Overpass API NumPy

Technical Implementation

Streaming JSON Parser: ijson processes multi-GB GeoJSON files incrementally without loading entire dataset into memory
Spatial Indexing: R-tree index implementation enables efficient nearest-neighbor queries across millions of footprints
Geometric Edge Analysis: Computes perpendicular normal vectors to building edges and selects outward-facing direction
Street Name Matching: Regex-based extraction and normalization handles address variations for accurate matching

Key Features

Hybrid Data Sources: Combines OpenStreetMap and Microsoft Buildings dataset for maximum coverage
Intelligent Fallback: Three-tier strategy ensures results even in data-sparse areas
Performance Caching: 24-hour TTL cache with session pooling reduces API calls by 80%
Scalable Architecture: Connection pooling and streaming I/O support production deployment

Challenges & Solutions

Memory Constraints: Implemented streaming parser reducing peak memory from 20GB to 500MB
Query Performance: Built spatial index achieving sub-second queries on 11M records
Corner Lot Ambiguity: Developed street name matching resolving 180° orientation conflicts

Results & Impact

The system successfully processes 11 million building footprints with 95% coverage for California and Texas addresses. Memory optimization through streaming reduces resource requirements from 20GB to 500MB (95% reduction), enabling deployment on standard hardware and cloud instances.

Sub-second query times ensure real-time responsiveness for address-to-orientation lookups, making the system suitable for production applications. The intelligent street name matching resolves corner lot ambiguities with high accuracy, while the hybrid data source approach maintains 95% coverage even in areas with incomplete OpenStreetMap data.

Geospatial Building Orientation Detection System