The KI‑DeltaLearning project, carried out from 1 January 2020 to 30 April 2023, delivered a fully integrated data‑management and processing platform that supports the entire workflow from raw sensor data acquisition to the generation of ground‑truth labels for machine‑learning models. The platform, developed in work package 1.1, consists of a user interface that provides a comprehensive overview of all available datasets, including key metadata such as dataset length, creation date, and the status of both automated and manual processing steps. Users can filter and search datasets, view video data, and examine the temporal evolution of individual attributes. For each dataset, the interface allows the manual release of “key‑frames”; these frames generate timestamps that are forwarded to a labeling team. The team annotates the corresponding video and LiDAR data, assigning object classes to image regions, thereby creating the ground‑truth required for training algorithms.

Behind the scenes the platform orchestrates a sequence of containerised processes: data validation and import, video creation from raw streams, automatic key‑frame selection, extraction of all sensor data for the selected key‑frames (including anonymisation of faces and licence plates), and the creation of a session in the C.LABEL system for the labeling team. The containerised architecture affords high flexibility, enabling the addition or modification of individual processing steps without disrupting the overall workflow. The platform’s design meets the project’s objectives: datasets can be inspected and processed rapidly, the majority of the workflow is automated, and privacy‑preserving anonymisation is integrated. Users have reported that the system is highly usable and relevant for their needs, and the platform is being positioned as a reusable component for future customer projects.

In addition to the core data‑processing work, CMORE Automotive contributed to the evaluation phase (TP 5). In work package 5.1 the consortium reviewed results from earlier increments and derived requirement changes for subsequent phases. Work package 5.2 saw CMORE develop baseline and training reference models, while in 5.3 the platform was provided to other partners as a demonstrator. The project’s financial outlay amounted to €232,285.35 in personnel costs, €377,037.00 for external R&D services, and €27,917.59 in other direct project costs, underscoring the substantial investment required to build a system from scratch rather than adapting existing solutions.

The project’s timeline was extended by four months to accommodate delays caused by the COVID‑19 pandemic and the protracted negotiation of data‑protection agreements. Despite these setbacks, the consortium achieved all milestones, and the platform’s containerised cloud implementation ensures long‑term scalability and adaptability. The anonymisation model developed within the project is being further refined as a reusable component, enabling its deployment in future customer applications. By involving OEMs, Tier‑1 suppliers, and other stakeholders across the automotive value chain, the consortium facilitated the rapid transfer of cutting‑edge research into industry practice, positioning the platform as a reference for data‑driven automotive research and development. The project was funded by German research agencies, reflecting national support for advancing data‑centric automotive technologies.