Radar Fault Detection via Camera-Radar Branches Learning Network

초록

Radars are widely used in autonomous driving technology. Self-driving usually relies on radar signals to recognize pedestrians and vehicles, identify the surrounding environments reliably, avoid car crashes and navigation, and provide a reliable route to avoid collisions. The radar plays an important role in vehicle systems, and maintaining its proper functioning is necessary for the safety of self-driving systems to be considered. However, sensor faults are unavoidable. When the radar sensor is faulty, the radar signal will not receive the correct feedback information. Currently, it is hard to detect fault errors in radars, and the algorithm is complicated to work with. To analyze the radar cross section (RCS) signal and distance relationship, we used the RCS signal feature and combined the real-time features of the vehicle camera with the convolutional neural network (CNN) model to identify the fault information as expected. The paper uses a new data generator feature and deep learning model, recognizes the input signal as normal and abnormal, and the accuracy improves to 95.54%.

키워드