Volkswagen Group, CARIAD: Active Learning for Self-Driving Cars- Developing active learning algorithms for self-driving cars- Deep learning model training for semantic segmentation and object detection- Export, model 'surgery', and deployment of models with ONNX and ONNX Runtime- Large investigation and usage of multiple experiment tracking tools- Developing a python based Machine learning system (MLOps, CI/CD in azure, hydra configs, experiment tracking, testing, code review, MLOps, pre-commits, linting, type-hints and environments)

2022.01-2023.12: Volkswagen Group, CARIAD - Active Learning for Self-Driving Cars- Developing active learning algorithms for self-driving cars- Deep learning model training for semantic segmentation and object detection- Export, model 'surgery', and deployment of models with ONNX and ONNX Runtime- Large investigation and usage of multiple experiment tracking tools- Developing a python based Machine learning system (MLOps, CI/CD in azure, hydra configs, experiment tracking, testing, code review, MLOps, pre-commits, linting, type-hints and environments)01.2021-12.2021: AGCO A/S - Datascience for Combine Harvester- Data science (Python) and software development (C++) on embedded platform to automate and optimize harvesting on a small fleet of Combine Harvesters. - Using Datascience to monitor and visualize operations of the small fleet.04.2020-12.2021: Large danish company: Deep Learning on a Microcontroller- Deep learning algorithms on highly constraint hardware (OpenMV). - Model training in Tensorflow/keras and transfer learning in Tensorflow JS. - Deployment of quantized models on a microcontroller (TFlite, C++).- Technologies: Tensorflow / Keras, Tensorflow JS, Tensorflow tflite for microcontrollers, image classification, object detection, Python, Typescript, C++

The objective of the industrial postdoc is to research and develop a vision-based system for underwater navigation. The vision system will use new and state-of-the-art computer vision and deep learning. A visual global positioning will be based on a map of selected visual landmarks and a recognizer to detect landmarks. Visual SLAM will measure movement between landmarks and create a point cloud to navigate and avoid collision. A pipeline detector will estimate the relative pose and guide the AUV along the pipeline. Furthermore, visual recognition algorithms must perform automated inspection by detecting events such as anode, joint, debris and damage on the pipeline.

The thesis proposes a detection system (TractorEYE), a dataset (FieldSAFE), and procedures to fuse information from multiple sensor technologies to improve detection of obstacles and to generate a map. Computer vision, Machine learning, Deep Learning, Object detection and Imaging sensors (RGB, Thermal and stereo camera) for autonomous systems in agriculture.