MAMOC
Automatic recognition of manual activities – now even more precise and suitable for industrial use thanks to an intelligent combination of object and gesture recognition!
MAMOC in Action
Mit dem Laden des Videos akzeptieren Sie die Datenschutzerklärung von YouTube.
Mehr erfahren
Foreword
MAMOC is a funded experiment in the context of the MIDIH (Manufacturing Industry Digital Innovation Hubs) EU research project. The aim of MIDIH is to provide digital innovations and services for the manufacturing industry. The focus here is on the needs of small and medium-sized enterprises (SMEs).
MAMOC (Machine Learning Application for Motion Capture)
The aim is to enable precise and detailed action recognition for industry. It is particularly useful for use at manual workstations and can support time management and industrial engineering issues.
Even the most complex processes in production plants are divided and structured into individual work steps. Each work step has – in its most rudimentary form – a clearly defined start, a clearly defined end and an operation time.
The sum of the process times of all the individual work steps of a product determines to a large extent the total working time required to manufacture a product. These process times are therefore also a significant factor in production costs.
The process times, which can be determined to the second in fully automated production, are much more difficult to determine for manual activities. This is where MAMOC’s automatic action recognition comes in.
We use it to create objective data collection with sufficient accuracy for several use cases, which helps to answer the following questions:
- How long does a single work step take?
- How long do the work steps take on average?
- Has a work step been carried out correctly?
- Have all work steps been carried out in the correct order?
Once the work steps have been defined and broken down into activities, MAMOC can help to answer these questions in a statistically significant way. Each of the automatically recognized activities is given a precise time stamp and thus provides the statistical basis for calculating the mean values for these work steps.
What is behind MAMOC?
The basis for MAMOC is an intelligent combination of the AI technologies object recognition and hand gesture recognition.
After initial training, the system is able to recognize the trained manual activities and automatically record their process times.
In the current showcase, this concerns the following activities: Hammering, screwing and cutting.
However, with sufficient training, other activities and entire work steps can also be recognized.
All the hardware required for this can be provided in the form of embedded solutions (e.g. from the manufacturer Nvidia).
Unlike other AI applications, no powerful gaming PC or even processor power from the cloud is required. Only specialized, lightweight embedded hardware without an active internet connection is used to operate MAMOC.
