Driver’s Chest Position Detection using FMCW Radar Data Collected in a Vehicle Mock-up and CNN

In this paper, a 3D chest detection framework for a driver seated in a vehicle mock-up is proposed. The aim of the project is to predict the 3D chest position of the driver while performing specific tasks in a fully autonomous vehicle. The proposed framework is implemented using convolutional neural network (CNN) forked-architecture with two frequency-modulated continuous wave (FMCW) radars and a Microsoft Azure kinect sensor. A dataset of 8,659 frames is built using the aforementioned sensors. Two CNN models are trained, tested and analysed using the dataset, demonstrating the possibility of using only radar data as input to predict human body key point in a vehicle compartment.