Discrimination of Automotive Radar Distributed Targets

Automotive radars are the main sensor enabling autonomous driving and active safety, and therefore, are required to provide high resolution in dense urban environments characterized by multiple distributed and close objects. Real targets are usually distributed over multiple ranges, Doppler frequencies, and angular bins. Super-resolution techniques allow distinguishing adjacent point targets, but they are not able to handle distributed targets. This work proposes a computationally attractive approach for discrimination between closely distributed objects in practical urban scenarios. Each distributed radar target is represented as a set of multiple scattering points with the associated joint probability density function, defined considering their position, shape, and velocity. The distributed targets are detected and enumerated according to the maximum likelihood and the Akaike information criterion. The ability of the proposed approach to accurately discriminate close distributed targets is evaluated via simulations.