Manufacturing-Product-Blockage-Detection

Overview

This repository contains a real-time computer vision system designed to detect and flag plastic products that get stuck at the exit cavity of a blow molding machine. The solution is optimized for edge deployment on a Raspberry Pi 4, utilizing a custom-trained YOLOv8 model converted to the NCNN framework for low-latency inference.

When a blockage occurs, the system visually highlights the stuck product on a live video stream, which is broadcasted over a local network using a Flask web server.

Features

• Edge-Optimized Inference: Uses a custom YOLOv8 model exported to NCNN format for accelerated performance on ARM architecture.
• Robust Object Tracking: Integrates the ByteTrack algorithm via the Supervision library to maintain object identities across frames.
• Custom Blockage Logic: Identifies stuck products by monitoring the bounding box aspect ratio and calculating the dwell time of each tracked item.
• Live Network Streaming: Streams annotated MJPEG video feeds over the local network via Flask, allowing operators to monitor the machine remotely.

Hardware Requirements

• Raspberry Pi 4 Model B
• Raspberry Pi Camera Module V2 (or compatible PiCamera2 supported hardware)

Software Stack

• Python 3
• Flask (Web Server & MJPEG Streaming)
• Picamera2 (Camera Interface)
• Ultralytics (YOLOv8)
• Supervision (ByteTrack & Detections)
• OpenCV (Image Processing & Annotation)

How It Works

1. Frame Capture: The PiCamera2 captures RGB frames at a resolution of 640x480.
2. Detection: The NCNN model processes the frame to detect plastic tubes.
3. Tracking: ByteTrack assigns and maintains a unique ID for every detected tube.
4. Blockage Logic:
   • The system calculates the width-to-height aspect ratio of each bounding box.
   • If a tracked object maintains an aspect ratio greater than 0.4 for more than 2.0 consecutive seconds, it is classified as stuck.
5. Alerting: The bounding box color shifts from green to red, the label updates to BLOCKED, and a global warning is overlaid on the video stream.

Installation & Setup

1. Clone the Repository

git clone https://github.com/Bashithaperera/Manufacturing-Product-Blockage-Detection.git

cd Manufacturing-Product-Blockage-Detection

2. Install Dependencies

It is recommended to run this in a virtual environment. pip install Flask picamera2 ultralytics opencv-python supervision

3. Model Placement

Ensure your custom-trained NCNN model weights are placed in the correct directory. By default, the script looks for the model at: tube_detection/weights/best_ncnn_model

4. Execution

Run the detection script: python detect.py

5. Viewing the Stream

Once the script is running, open a web browser on any device connected to the same local network and navigate to: http://:5000

Configuration

You can fine-tune the tracking and blockage logic by adjusting the parameters directly in detect.py:

• aspect_ratio_threshold = 0.4: Minimum aspect ratio to trigger the blockage timer.
• block_time_threshold = 2.0: Consecutive seconds the aspect ratio must be exceeded to flag a blockage.
• track_activation_threshold = 0.15: Confidence threshold for ByteTrack.

Output (frame extracted from results_sample.mp4)

License

MIT License