Non-Line-of-Sight Pedestrian Recognition Approach Using Multiple Reflections for Millimeter Wave Radar Sensing Applications

We introduce a target recognition scheme based on multiple signal reflections assuming a non-line-of-sight (NLOS) scenario for a millimeter wave (MMW) radar used in automotive sensor applications. For a pedestrian or a cyclist jumping out from the rear of an obstacle, such as a parked vehicle, a target recognition scheme in a NLOS region is highly required for collision prevention, especially in driving support and autonomous driving systems. Since the diffraction signal obtained from a NLOS region is too weak for identifying a specific target type, particularly at high frequencies and under full NLOS conditions, we introduce a target recognition approach based on multiple reflections. Two machine learning approaches, based on the support vector machine (SVM) and long-short-term memory (LSTM) algorithms are introduced with only single transmitter and receiver data. The experiments conducted using a 24-GHz band MMW radar in an actual road scenario demonstrated that the proposed approach can accurately recognize a real human body from an artificial object by exploiting the unique characteristics of reflected signals.