Dubey, AnandAnandDubeyFuchs, Jonas BenjaminJonas BenjaminFuchsMadhavan, VenkatVenkatMadhavanLübke, MaximilianMaximilianLübkeWeigel, RobertRobertWeigelLurz, FabianFabianLurz2021-01-112021-01-112020-09-21IEEE National Radar Conference (2020)http://hdl.handle.net/11420/8372The inherent smaller radar cross sections of vulnerable road users resulting in smaller signal-to-noise-ratios make an accurate detection of them somewhat challenging. Mutual radar interference in typical automotive scenarios further imposes the difficulty of a target detection by additionally raising the noise floor. The traditional signal processing pipeline consists of multiple but separate stages for interference detection, mitigation and target detection. In this paper, a convolutional neural network based autoencoder architecture is used to perform a combined single-stage target detection while generalizing over different interference noise. The proposed approach achieves significant improvement over state-of-the-art methods while preserving the instance of each target and is able to identify them uniquely in case of a partial occlusion or overlapping of multiple targets.enautomotive radarconvolutional - autoencoderinterference mitigationVRU detectionConference Paper10.1109/RadarConf2043947.2020.9266434Other