Child presence detection inside vehicles with NIR camera: Lead role for all AI related activities, effort and feasibility estimation, planning of data recording and labelling campaigns, training and testing of ML models using TensorFlow and Keras. Application development in C++ for deployment on Qualcomm Snapdragon and QNX based automotive HPC system. Tests with synthetic data in addition to real images. Passenger abnormal body pose detection: Combined pose keypoint estimation with pose classification by feeding keypoints into tree based ML model

Detection of objects inside vehicles with NIR/RGB cameras: Collected and (auto)labeled images, trained model using TensorFlow Object Detection API, hyper parameter optimization. Application development in Python and C++ for deployment of AI models. Integrated object detection and pose estimation algorithms for a cabin sensing demonstrator running inside an autonomous shuttle bus. Model quantization with TensorRT for Nvidia Jetson Nano target device. Implemented wrapper for TensorFlow Object Detection API to simplify training on custom datasets, created Docker image for dependencies. Used by colleagues across departments for training models. Developed visual crop row detection node for agriculture robot using a stereo camera, Python, OpenCV and ROS.