The problem with a master-plan image
Housing societies usually publish their layout as a large marketing image. It looks like a map, but to a computer it is only pixels: there are no plot boundaries, no coordinates, and nothing you can search for "plot 214". Redrawing tens of thousands of plots by hand is slow and error-prone.
The process, step by step
- Clean the image. Remove watermarks and text overlays that would confuse detection, so plot boundaries stand out.
- Find every plot. Image segmentation (for example the watershed technique) separates the image into individual plot shapes.
- Read the plot numbers. Optical character recognition (OCR) reads the number printed inside each plot and links it to that shape.
- Read the sizes. Many plans colour plots by size in a legend; reading those colours assigns a size category to each plot.
- Turn shapes into vectors. Pixel outlines become clean polygons that can be stored in a spatial database such as PostGIS.
- Georeference. The polygons are fitted to known GPS reference points so each plot sits in the right place on a real map.
- Review in QGIS. A person checks the result block by block, fixes mistakes and confirms numbers. Every plot number keeps a record of where it came from.
What we built
For Plot Atlas, Kryyvo started from a 14,002 × 9,977 px marketing image and produced about 31,000 searchable plots. The pipeline uses Python, OpenCV, OCR, QGIS and PostGIS, with a Next.js front end, and has 400+ automated tests. Serving the map as optimized data cut its payload from 25.7 MB to 1.9 MB. QGIS is the review desk, with bookmarks for 70 blocks. See the Plot Atlas case study.
How accurate is it?
This approach is right for search, listings and navigation. It is not survey-grade: positions are aligned to reference points and reviewed by hand, not measured on the ground, so it should not be used for legal boundaries.
FAQ
Can this work for any housing society map?
It works best on clear, high-resolution plans with printed plot numbers. Low-resolution or heavily stylized images need more manual review.
What can you build on top of the data?
A searchable plot map, property listings anchored to real plot numbers, and mobile apps for agents and buyers. Kryyvo's Zameen Expert app is built on this data.
Why review by hand at all?
Automatic detection and OCR are fast but not perfect. A structured human review step is what makes the data trustworthy enough for a real product.