Agile evaluation and development of AI methods using novel sensing principles requires rapid data acquisition and manual or automated annotation of recorded data. For transfer learning in the area of object detection from a vehicle perspective, datasets need to be created and evaluated.
During test drives, sensor data such as RGB, LiDAR or vehicle bus data (CAN, ...) are initially recorded. Based on the test data, approaches for manual or automated annotation are developed to train AI-based methods with a dataset and improve the detection quality.
The acquired data can be efficiently annotated and evaluated for their suitability to re-train the AI-based method.
Real-world setup on the FZI's CoCar experimental vehicle for data acquisition.