Fusion of doppler radar and geometric attributes for motion estimation of extended objects

A prime requirement for autonomous driving is a fast and reliable estimation of the motion state of dynamic objects in the ego-vehicle´s surroundings. An instantaneous approach for extended objects based on two Doppler radar sensors has recently been proposed. In this paper, that approach is augmented by prior knowledge of the object´s heading angle and rotation center. These properties can be determined reliably by state-of-the-art methods based on sensors such as LIDAR or cameras. The information fusion is performed utilizing an appropriate measurement model, which directly maps the motion state in the Doppler velocity space. This model integrates the geometric properties. It is used to estimate the object´s motion state using a linear regression. Additionally, the model allows a straightforward calculation of the corresponding variances. The resulting method shows a promising accuracy increase of up to eight times greater than the original approach.