Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Xander44-4/traffic_reducer/llms.txt

Use this file to discover all available pages before exploring further.

Traffic Reducer requires Python, a small set of pip packages, and a few supporting assets — YOLO weight files, an optional FFmpeg binary, and a pre-trained sklearn model. This page covers every prerequisite and explains what each component does, so you have a complete picture before running the installation command.

​
Prerequisites

Before installing, confirm that your system meets the following requirements:
  • Python 3.10 or 3.11 — required for compatibility with ultralytics, scikit-learn, and the rest of the dependency stack. Python 3.12+ may work but is not tested. Verify with: 
    python --version
  • pip — the standard Python package manager, included with all Python 3 distributions. Verify with: 
    pip --version
  • Internet connection (for YouTube mode) — yt-dlp must resolve a live HLS URL from YouTube each time the YouTube source is activated. Local video mode works fully offline.

​
Python packages

All required packages are listed in requirements.txt at the project root: 
flask
pandas
numpy
joblib
scikit-learn
opencv-python
ultralytics
yt-dlp
The table below explains what each package contributes to the system:
PackageVersion notePurpose
flaskAny current stable releaseWeb server, Jinja2 template rendering, and REST API routing for the dashboard
opencv-pythonAny current stable releaseFrame capture from local video files and the FFmpeg pipe, zone polygon testing, bounding-box drawing, and JPEG encoding
ultralyticsAny current stable releaseYOLOv8 model loading and inference; also installs imageio-ffmpeg as a dependency
yt-dlpAny current stable releaseExtracts a direct HLS stream URL from a YouTube live link so FFmpeg can pipe raw video frames
pandasAny current stable releaseData handling utilities used across the processing pipeline
numpyAny current stable releaseNumerical operations — polygon centroid tests, argmax phase selection, HSV colour ratio calculations for emergency detection
joblibAny current stable releaseAlternative deserialiser for the pre-trained sklearn .pkl model file; used as a fallback when pickle.load fails
scikit-learnAny current stable releaseProvides the runtime environment for the pre-trained signal-phase classification model
Install everything with a single command run from the project root: 
pip install -r requirements.txt
On Windows, use py instead of python if the python command is not recognised. On macOS and Linux, use python3 and pip3 to avoid accidentally targeting a system Python 2 installation.

​
YOLOv8 model weights

Traffic Reducer ships with three YOLO weight files in the repository root:
FileSizeNotes
yolov8m.ptMediumDefault — best accuracy for multi-vehicle intersection scenes
yolov8s.ptSmallFaster inference; used automatically if yolov8m.pt is absent
The application selects the model at startup using the following priority order, taken directly from traffic_app/app.py: 
_yolov8m_local = os.path.join(PROJECT_ROOT, "yolov8m.pt")
_yolov8s_local = os.path.join(PROJECT_ROOT, "yolov8s.pt")
if os.path.exists(_yolov8m_local):
    YOLO_MODEL_PATH = _yolov8m_local
elif os.path.exists(_yolov8s_local):
    YOLO_MODEL_PATH = _yolov8s_local
else:
    YOLO_MODEL_PATH = "yolov8m.pt"
If neither local file exists, Ultralytics will attempt to download yolov8m.pt from its model registry on first run. Keep at least one .pt file in the project root to avoid this network request at startup.

​
FFmpeg

Traffic Reducer uses FFmpeg to decode the YouTube HLS stream into raw BGR frames that OpenCV can process. You do not need to install FFmpeg manually. The imageio-ffmpeg package — installed automatically as a dependency of ultralytics — bundles a platform-appropriate FFmpeg binary and exposes it at runtime: 
# From traffic_app/video_processor.py
import imageio_ffmpeg
FFMPEG_BIN = imageio_ffmpeg.get_ffmpeg_exe()
The resolved binary path is passed directly to subprocess.Popen when opening a YouTube stream. No PATH configuration or system-level FFmpeg installation is required.

​
Pre-trained sklearn model

The repository includes a pre-trained scikit-learn model for signal-phase classification: 
traffic_reducer_dataset/modelo_entrenado/modelo_semaforo_ia.pkl
This .pkl file is loaded at server startup by traffic_app/app.py using pickle.load with a joblib.load fallback. It is used by the POST /predict endpoint to classify the appropriate signal phase from vehicle count inputs.
The MODEL_PATH constant in traffic_app/app.py is currently hardcoded to the original developer’s local machine path:
MODEL_PATH = r"C:\Users\Xande\important\The Predictors\Haklaton\The Predictors-traffic_reducer\traffic_reducer_dataset\modelo_entrenado\modelo_semaforo_ia.pkl"
This path does not exist on any other machine. Before running the server in production, open traffic_app/app.py and update MODEL_PATH to the correct absolute path of modelo_semaforo_ia.pkl on your system. For example:
MODEL_PATH = r"C:\Projects\traffic_reducer\traffic_reducer_dataset\modelo_entrenado\modelo_semaforo_ia.pkl"
If the file is not found, the application starts normally but logs a warning and disables the /predict endpoint — all other features, including live YOLOv8 detection and the dashboard, continue to work.

​
Verifying the installation

Once pip install -r requirements.txt completes, run the server to confirm everything is in order: 
py traffic_app/app.py
A successful startup prints the following lines before the Flask development server banner: 
[App] Modelo YOLO: yolov8m.pt
[App] Video local: .../traffic_dron_view.mp4 (OK)
[App] Modo por defecto: idle
Iniciando Traffic Reducer Server...
http://127.0.0.1:5000
If you see these lines, all packages loaded correctly, the YOLO weights were found, and the Flask server is accepting connections. Proceed to the Quickstart to connect a video source and activate AI control.

Build docs developers (and LLMs) love

Get started for freeTalk to us