Direct Multi-Target Localization With Cooperative Automotive Radar

We explore cooperative automotive sensing, where a pair of communicating vehicles cooperate in locating multiple targets in the surroundings. Each vehicle is equipped with a mono-static radar, while a bi-static radar, which offers a different probing angle of the environment and thus geometric diversity, can be formed by leveraging the inter-vehicle communication signal for sensing. Target localization is typically carried out using a two-step approach, which first estimates some intermediate location-dependent parameters (e.g., delays) and then converts them into location estimates via nonlinear regression. For multi-target localization, the two-step approach has to solve a combinatorial data association problem. To address this challenge, we consider a direct approach to locate targets directly from the observed signals. We first introduce the optimum joint maximum likelihood estimator (MLE), which also entails a high complexity, as a benchmark method. We then propose a sparse recovery and refinement (SRR) method, which integrates the alternating direction method of multipliers (ADMM) algorithm with successive cancellation in an iterative way. SRR is able to benefit from the computational efficiency of ADMM without suffering its grid-mismatch problem. A joint Cramér-Rao bound (CRB) is presented as a performance assessment tool for cooperative localization. Numerical results show that geometric diversity enables cooperative sensing to surpass the conventional non-cooperative sensing in terms of accuracy and energy efficiency; among the cooperative sensing schemes, SRR yields improved estimation performance and can better resolve closely spaced targets than the two-step approach.