The repository was created as part of a master's thesis at the Brno University of Technology. The main motivation for the project was the deployment of a computer vision application in the robotic cell EDUset ONE, located at Intemac Solution s.r.o. Within the thesis, three leading detection frameworks, R-CNN, SSD, and YOLO, were deployed. Subsequently, the frameworks were compared based on their own criteria. The repository can serve as a starting point for implementing these frameworks on custom datasets.

The custom dataset used to train the detection neural networks is available at the following two links:

1. Roboflow
2. Kaggle

In the samples folder, there are examples of Jupyter notebooks that were used to work with detection frameworks. These examples demonstrate how it is possible to deploy each of the detection frameworks on a custom dataset. The background for the examples is in the eduset subfolder. This folder can serve as the default package for subsequent work.

In the main folder, there is a custom program for semi-finished goods detection. The main principle is that an API is created, which communicates with an industrial camera from BASLER. The API is invoked by an OPC UA client, which is connected to an OPC UA server running on a PLC from Siemens. The client and API can run independently of each other and can also be started independently using the following commands:

Running the API on localhost:

uvicorn.run(app, host="localhost", port=8000)

Running the OPC UA client:

url = "opc.tcp://192.168.0.10:4840"
    
api_endpoint = "http://localhost:8000/placement"

control_signal = {"ns": "http://EdusetONE", "i": 75}
placement = {"ns": "http://EdusetONE", "i": 19}

client = OPCUAClient(url=url)

try:
    await client.set_encryption() if client.encryption is True else None
    await client.run(control_signal, placement, time=2, api_endpoint=api_endpoint)
except KeyboardInterrupt:
    await client.disconnect()

License:

The project is licensed under the GNU General Public License version 3 (GPLv3)

Contact:

For inquiries and support, please contact [email protected] for more information.