A lightweight desktop tool for civil engineers: box-select elevation numbers on scanned grading plans and automatically generate flow direction arrows.
Open a scanned PDF grading plan, click the physical measurement points on the map, box-select the corresponding elevation numbers, and the tool will:
- OCR the elevation values using PP-OCRv6 tiny_rec (local ONNX Runtime) — no API key, no internet required
- Compute elevation differences (Δ) between consecutive points
- Draw red flow direction arrows with delta labels; equal elevations render as a plain FLAT line with no arrowhead
- Automatically identify HP (High Point) and LP (Low Point) extrema
- Export the annotated plan as WYSIWYG image (PNG/JPG) or native PDF vector annotations
Built by a practicing civil engineer for civil engineers.
This tool relies on vision-based OCR to read elevation numbers (e.g. "31.95 FS") from scanned PDF pages — often on grayscale, noisy or hand-annotated grading plans.
- PP-OCRv6 tiny_rec (Baidu PaddleOCR, released 2026.6.11) is a dedicated OCR recognition model with only 1.1M parameters.
- It runs locally via ONNX Runtime — no GPU needed, no cloud dependency, ~2ms per inference.
- Achieves 94–99.99% confidence on elevation numbers with suffixes (FS, EL) and HP/LP labels.
- 4-angle rotation handling: automatically tries 0°/90°/180°/270° and picks the highest-confidence result, so rotated numbers on PDFs are recognized correctly.
- The full pipeline replaces the previous Google Gemini API OCR — removing API costs, network dependency, and configuration overhead.
- Download
FlowlineChecker.exefrom the Releases page. - Double-click
FlowlineChecker.exeto launch. - Click Open PDF, select your grading plan, and start drawing.
No API key, no internet, no configuration. The 4.3MB ONNX model is embedded in the EXE. OCR runs entirely offline.
# Clone the repo
git clone https://github.com/wyuebei-cloud/PDF-flowline.git
cd PDF-flowline
# Create virtual environment and install dependencies
python -m venv venv
# Windows:
venv\Scripts\activate
pip install paddleocr onnxruntime opencv-python pillow numpy PyQt6 pymupdf
# Launch (model auto-downloads on first run)
python flowline_checker\main.py| Step | Action |
|---|---|
| 1 | Click Draw Flowline on the toolbar (cursor becomes a red crosshair) |
| 2 | Click the physical location on the plan where elevation is measured |
| 3 | Immediately box-select the elevation text (e.g. "31.95 FS") |
| 4 | A blue label appears — the app returns to drawing mode without waiting |
| 5 | Repeat step 2–4 for subsequent points |
| 6 | Click Done to finalize → arrows, deltas, HP/LP labels are generated |
- Edit a blue number: single-click it before pressing Done
- Select/Deselect an arrow: click it (blue dashed box appears)
- Delete selected arrow: press
Delete - Undo:
Ctrl+Z - Cancel current segment:
Esc - Export Image:
Ctrl+S(WYSIWYG high-res PNG/JPG) - Export PDF:
Ctrl+E(native vector annotations)
- Arrow Size slider — adjusts arrowhead and line thickness globally in real time
- Text Size slider — adjusts delta (Δ) and HP/LP label font size
PDF-flowline/
├── flowline_checker/ # Source code
│ ├── main.py # Entry point
│ ├── core/
│ │ ├── ocr_engine.py # PP-OCRv6 tiny_rec (local ONNX)
│ │ └── pdf_handler.py # PDF rendering, annotation, export
│ ├── ui/
│ │ ├── main_window.py # Main window with toolbar
│ │ ├── pdf_viewer.py # PDF viewer (pan/zoom)
│ │ ├── selection_overlay.py # Selection box overlay
│ │ └── value_dialog.py # Manual value edit dialog
│ └── models/
│ └── data_types.py # ElevationPoint / FlowArrow data models
├── USER_GUIDE.md # Full user manual (English & Chinese)
├── design_history_log.md # Development design log (bilingual)
├── LICENSE # MIT License
└── README.md # This file
The pre-compiled FlowlineChecker.exe is distributed via GitHub Releases, not stored in the source tree.
This project is licensed under the MIT License — see LICENSE for details.
Made by a PE at KPFF. First-gen Chinese immigrant in the PNW. Licensed PE (WA/OR) since 2013.