Event-based dataset for the detection and classification of manufacturing assembly tasks

Area of Science:

• Robotics and Automation
• Computer Vision
• Human-Computer Interaction

Background:

• Manufacturing assembly involves fundamental human actions.
• Event cameras offer efficient visual data capture for dynamic scenes.
• Real-time human motion analysis is crucial for industrial automation.

Purpose of the Study:

• Introduce the Event-based Dataset of Assembly Tasks (EDAT24).
• Provide a resource for studying human motion in manufacturing tasks.
• Facilitate research in real-time event-based vision for robotics.

Main Methods:

• Collected data using a DAVIS240C event camera capturing light intensity changes.
• Recorded 400 samples across four manufacturing primitives: idle, pick, place, and screw.
• Utilized the CT-Benchmark for object interaction scenarios.
• Provided data in raw (.aedat) and pre-processed (.npy) formats.

Main Results:

• The EDAT24 dataset comprises 400 samples of event and greyscale frames.
• Data captures basic human operator actions in a manufacturing context.
• Includes custom Python code for dataset extension and new primitive integration.

Conclusions:

• EDAT24 is a valuable resource for developing and testing event-based vision algorithms for human motion analysis in manufacturing.
• The dataset supports research in real-time robotic control and human-robot collaboration.
• Availability of raw and processed data, along with code, promotes accessibility and further research.