📦 google-mediapipe-gst-python-implementation

🚀 Real-Time GStreamer Video Processing with Shared Memory and Batch Processing

google-mediapipe-gst-python-implementation is a high-performance, real-time video processing pipeline that:

• Streams frames from a GStreamer pipeline
• Batches frames efficiently into shared memory
• Distributes batches across multiple processes using multiprocessing queues
• Supports scalable, parallel feature extraction (e.g. using Mediapipe)

This project is designed for low-latency, high-throughput computer vision pipelines.

---

📂 Project Structure

.
├── gaze_tracker.py       # gaze tracking function
├── utils.py              # Shared utilities (e.g., batch handling, frame reshaping)
├── requirements.txt      # Python package dependencies
└── README.md             # Project documentation

---

⚙️ Features

• ✅ Real-time GStreamer frame ingestion
• ✅ Batch processing for improved memory and CPU efficiency
• ✅ Shared memory transport (zero-copy frame passing)
• ✅ Multi-process consumer architecture (supports parallel Mediapipe tasks)
• ✅ Easily configurable batch size and frame resolution
• ✅ Scalable for multi-camera or high-frame-rate pipelines

---

🛠️ Requirements

• Python 3.8+
• GStreamer 1.18+

Install Python dependencies:

pip install -r requirements.txt

Example requirements.txt:

numpy
opencv-python
matplotlib

---

🚀 How It Works

1. Producer:

   • Reads frames from the GStreamer pipeline using appsink
   • Writes frames into a shared memory batch buffer
   • Signals when batches are ready via a lightweight queue

2. Consumers:

   • Listen for batch-ready signals
   • Read batches directly from shared memory (zero-copy)
   • Process each frame independently (e.g. using Mediapipe feature extraction)

---

📦 Usage

Run the pipeline:

python manage.py runserver

Customize:

• Adjust BATCH_SIZE and FRAME_SHAPE in utils.py
• Integrate your own Mediapipe feature extraction functions in consumer.py

---

📸 Example Pipeline

pipeline = Gst.parse_launch('videotestsrc ! video/x-raw,format=BGR,width=1280,height=720 ! appsink name=sink')

You can easily replace this with your own GStreamer input (webcam, video file, live stream, etc.)

---

🔗 Future Work

• Support for double-buffering to enable full parallel read/write cycles
• Multi-camera ingestion
• GPU-accelerated processing support
• Optional Kafka or distributed queue integration

---

📄 License

This project is licensed under the MIT License.

---

🤝 Contributions

Pull requests and improvements are welcome! Please open an issue first to discuss major changes.

---

✨ Credits

Built by http://github.com/KarthikMgk for scalable, real-time vision pipelines.