Statistical Characteristic-Based Road Structure Recognition in Automotive FMCW Radar Systems

Abstract

This paper proposes an efficient road structure recognition method using statistical characteristics of received signals in automotive frequency-modulated continuous wave radar systems. Generally, roads consist of various structures, some of which, such as tunnels and soundproof walls made of iron, generate undesired echoes, called clutter. When the clutter flows into the radar system, the target detection performance cannot be guaranteed completely. This causes great danger to the driver using the radar function such as adaptive cruise control. Thus, an efficient method to recognize the structures that deteriorate the radar detection performance is desired. Depending on the types of road structures, frequency components of the received signals have distinctive distributions. Focusing on this point, parameters that reflect statistical properties of each distribution are extracted. These parameters can be used as standards for the recognition because they show different values according to the road structures. For more enhanced recognition, we use a support vector machine method with a linear classifier or a Gaussian kernel, and the resulting confusion matrices are derived. According to the results, the proposed method successfully classifies the structures with high accuracy. If the recognition of the road structures that degrade radar's function is performed effectively, the safety of the driver in the radar-equipped vehicle can be ensured by applying additional signal processing or giving a warning message to the driver.