Increasing product variety, cost pressure and demand for skilled workers are leading to a growing need for full automation in logistics worldwide. Existing automation systems do not provide the necessary capabilities to cope with the complexity and high process dynamics in logistics. Accordingly, a paradigm shift in logistics can be observed - from rigid programmed machines to intelligent machines. Machine learning methods produce excellent results when classical approaches fail. The deciding factor for the success of machine learning is the amount of data that the model can access to.
In machine image processing, previous image databases focus on common scenes of common objects such as people, clothing, and food. A large-scale integration into the logistics processes has failed so far, since modern image data sets do not cover the required task and environment-specific object classes. In order to solve this problem, the Chair of Material Handling, Material Flow, Logistics created and published the LOCO record (Logistics Objects in Context), which deals with industrial scenes and logistic-specific objects such as pallets, load carriers or conveyor systems.
The aim of the project is to create and publish a dataset of annotated, intra logistic images for the training of learning algorithms (e.g. deep learning). This dataset enables the transfer of knowledge from the area of machine learning into the logistics domain. The open-innovation approach allows the further development of the object recognition algorithms in the field of logistics and will enable the cost-effective development of intelligent systems, such as assistance systems for industrial trucks or autonomous transport systems for both research institutes and small and medium-sized enterprises.
In the first step of the research project, a market analysis for suitable industrial partners will be carried out. Then we will contact and inform the possible industrial partners about the project. We are looking for companies with production and logistics centres of any industry (automotive, retail, food, etc.).
In the second step, the pictures will be taken. During the on-site visit, photographs are taken with different cameras in the production and logistics area of the industrial partners. If employees are in the area of the recordings, they are already pixelated during the recording and made illegible - only when that happened, the recordings are saved.
In the third step, the recordings are manually annotated and finally published. If you are interested in a cooperation or the dataset, please contact the contact person listed below.