MAMOC

Experiment description

MAMOC’s open call project’s goal was to show that an AI application enables automatic annotations of video material in the sense of « recognizing the activities that happen in the video ». To achieve this, we combined object detection and hand pose estimation, enriched with position data. With this data we trained a neural network to recognize the actions happening. The solution is particularly intended for the optimization and digital support of workplaces in manual production.

Technical impact

We gained a much better insight in the MIDIH-Architecture and managed also to integrate our MAMOC solution with the MIDIH Apache components: Nifi, Kafka, and Cassandra. In this area we found that it is very simple and advantageous to use standardized, reusable and easy to integrate components to support standardization in the industry, even with open source products. We were able to implement the components in a very short time. In the next figure MAMOC architecture is illustrated.

Figure: MAMOC archtecure.

Economical / Business impact

As of now we have a prototype setup for showing  our expertise in the topic of modern workplace supervision. It also strengthens our skills and introduced us to some new topics of modern machine learning foundations.

We will still focus on product integration to video analysis toolset from the dmc-Ortim and try to advertise our solution to their customers in the field of industrial engineering to get a better feedback and even some more business opportunities.