Über
scanned architectural floor plans (PDFs) —including NYC as-built condominium filings—into
structured, coordinate-based spatial data
suitable for 3D modeling and GIS workflows. This role focuses on
reconstructing accurate geometry from raster scans , not extracting existing vectors and not training end-to-end black-box models. The core challenge is turning noisy, skewed, real-world blueprint scans into
watertight room polygons with real-world coordinates . We are looking for a senior engineer who is strong in
classical computer vision, computational geometry, and raster-to-vector reconstruction , and who enjoys solving hard, practical problems with deterministic and explainable systems. Key Responsibilities Geometry & Vision Pipeline Design and implement a
raster-to-geometry pipeline
for scanned architectural PDFs Build robust preprocessing tools for: deskewing binarization noise reduction normalization of low-quality scans Isolate architectural linework (walls, boundaries) from: text dimensions symbols stamps and annotations Handle
door gaps and broken boundaries
to ensure enclosed, “watertight” regions Extract enclosed regions (rooms, corridors) using connected components and topology analysis Convert raster regions into
clean polygon geometry contour extraction polygon simplification vertex snapping consistent winding and validity checks Spatial Accuracy & Scaling Develop deterministic methods to convert pixel geometry into
real-world X/Y coordinates Calibrate scale using: architectural dimension annotations scale notes when available Validate geometry numerically: closed polygons area consistency tolerance-based error detection Text & Semantic Integration Integrate OCR outputs to: associate room labels with polygons parse dimension strings (feet/inches, metric) extract height or ceiling notes Map semantic text to spatial geometry using proximity and containment logic Output & Integration Produce structured JSON outputs aligned with downstream 3D/GIS systems Ensure outputs are explainable, debuggable, and consistent across floors and documents Build internal visualization/debugging tools (overlays, masks, polygon previews) What This Role Is
Not Not prompt engineering Not LLM application development Not training large end-to-end neural networks Not purely academic research This role is about
deterministic geometry extraction from real-world scanned documents . Required Technical Skills 5+ years of experience in
computer vision, image processing, or computational geometry Strong command of
classical CV techniques , including: thresholding and morphology (dilate/erode/open/close) edge and line detection (e.g., Hough transforms) connected components and region analysis contour tracing and polygon simplification (e.g., Douglas–Peucker) Solid understanding of
planar geometry and numerical robustness Experience converting
raster data into vector or polygon representations Strong Python skills (NumPy, OpenCV, scikit-image, etc.) Comfortable debugging visually and iterating on messy real-world data Strongly Preferred Experience with architectural drawings, floor plans, CAD, BIM, GIS, or maps Familiarity with OCR systems and bounding-box–based text extraction Experience parsing architectural dimensions (feet/inches or metric) Experience validating polygon geometry (self-intersection, closure, area) Prior work on document image analysis or technical drawings Nice to Have Experience using
pretrained segmentation models
to supplement classical CV Exposure to GIS or BIM formats (GeoJSON, IFC, IMDF) Knowledge of NYC as-built or Department of Buildings / Finance document conventions Experience building internal QA or visualization tools Familiarity with downstream 3D geometry pipelines
Sprachkenntnisse
- English
Hinweis für Nutzer
Dieses Stellenangebot stammt von einer Partnerplattform von TieTalent. Klicken Sie auf „Jetzt Bewerben“, um Ihre Bewerbung direkt auf deren Website einzureichen.