Work Zone Digital Twin — Pipeline Testing, Validation & Bug Fixes Project Overview We have a completed 11-stage Python pipeline that converts dashcam video and GPS/IMU telemetry into a top-down map of highway work zone assets (cones, drums, signs). The pipeline is built and running end-to-end. We need an engineer to test it thoroughly, validate outputs, and fix remaining issues. Pipeline Summary 11-stage offline pipeline: Stage 1-3: Telemetry preparation, keyframe extraction, video/GPS sync Stage 4: COLMAP sparse SfM reconstruction Stage 5: Metric scale alignment (GPS + physical reference measurement) Stage 6-7: Manual and auto annotation tools (browser-based) Stage 8: Multiview triangulation of asset 3D positions Stage 9: Taper ordering and work zone measurements Stage 10: Ground truth validation Stage 11: Report generation (topdown.png, summary.md, evidence.csv) Tech Stack Python 3.10+ COLMAP (custom GPU build) RT-DETR object detection model (HuggingFace) NumPy, pandas, OpenCV, matplotlib ENU coordinate system, GPS/IMU sensor fusion What We Need Testing — run the full pipeline on multiple video segments and verify outputs are correct at each stage Validation — compare pipeline measurements (cone spacing, taper length) against known ground truth values, document accuracy Bug Fixes — identify and fix remaining issues including: Stage 9 taper/lane closure measurement accuracy Asset deduplication and clustering logic Scale factor computation for new recordings Category mapping between detector output and pipeline schema Documentation — document any changes made and update the existing codebase comments What You'll Receive Full codebase on private GitHub repo Detailed handoff documentation (HANDOFF.md) Sample video recordings with telemetry Fine-tuned RT-DETR model for work zone detection Existing test results for reference Ideal Candidate Strong Python debugging skills Familiar with computer vision pipelines Experience with SfM / COLMAP a plus GPS/IMU sensor data experience a plus Can read and understand an existing codebase quickly Detail-oriented — comfortable with numerical validation