Machine Learning-Based Estimation for Tilted Mounting Angle of Automotive Radar Sensor

Abstract

Generally, automotive radars are installed behind bumpers, perpendicular to the ground. After mounting the radar, the mounting angle is often distorted due to external shock during driving. If the radar is tilted toward the ground or sky, its detection performance can be severely degraded, and this can pose a serious threat to the safety of drivers. Therefore, it is important that the radars should be installed at an appropriate angle. This paper proposes an effective method for estimating the tilted mounting angle of the radar in an automotive frequency modulated continuous wave radar system. First, we identify that the frequency spectrum of the received radar signal varies significantly depending on the radar tilt angle. Then, we extract features representing the statistical properties of the received radar signal and use them as criteria for classifying various signals with different tilt angles. In addition, we use the principal component analysis algorithm to reduce the dimensionality of the feature space and increase the execution speed. The classification results using the $k$ -nearest neighbor algorithm as a classifier demonstrate that our proposed method can estimate the tilt angle of the radar with an accuracy greater than 90%.