An internationally active automotive supplier based in Germany – one of the world’s largest providers of mechatronic systems for vehicle doors, seats, and electric motors – continuously invests in cutting-edge manufacturing technologies to maintain the highest quality standards in premium car seat production.
This production process combines intricate manual steps such as stamping, welding, painting, and gluing with robot-assisted assembly lines. Among the most critical stages is the automated assembly of seat frames and recliners, where components from various models and suppliers must match precisely.
Proof of Concept: Tracking a German car brand logo
To evaluate the feasibility and precision of AI-based visual monitoring, DEEPEYES was initially commissioned to carry out a small-scale pre-project:
The objective was to detect and track the manufacturer logo on individual components in motion during robotic assembly.
Using recorded video footage, the DEEPEYES system successfully followed the small logo throughout the assembly process — demonstrating its capability for high-precision part-level traceability, even under dynamic production conditions.
This successful proof of concept validated the potential of DEEPEYES technology for broader applications and laid the foundation for more comprehensive computer vision monitoring of the seat assembly process.
Challenge
Following the PoC, the client aimes to tackle a recurring issue on the robotic seat production line: Human error occasionally leads to incorrect combinations of seat and recliner frames being loaded into the assembly process.
- Mismatched parts from different brands or models entered the line
- These errors often remain undetected until full robotic assembly is complete
- Once assembled, the components can’t be separated — leading to costly scrapping of entire units
- The result: high material waste, delays, and production downtime
Solution
Based on the PoC success, DEEPEYES will develop a dedicated computer vision module for real-time validation of components before robotic assembly begins:
- Incorrect parts are detected immediately, before the robot starts its operation
- Operators receive instant visual feedback at their workstations via an intuitive user interface
- The system runs fully on-premises, meeting data privacy and regulatory requirements
Goals
- Zero faulty assemblies since the introduction of AI-based monitoring
- Significantly reduced material waste and improved production stability
- High acceptance among operators due to real-time and easy-to-understand feedback
- Scalable rollout planned across additional lines and production facilities
Technical Setup for the POC:
- Four low cost 1080P cameras cameras, positioned on all four sides of the robot cage at a height of 2.5 meters.
- The dynamic tracking proved more effective using film sequences than individual image frames.
- The project ran for 6 weeks (approx. 1.5 months)
- A dedicated 2-person team worked on the implementation
- Multi-angle video enabled robust detection – even in difficult lighting or when parts were partially obscured
- The subsequent project will use the same technical setup.
