The evolution of connected and automated vehicles (CAVs) technology is boosting the development of innovative solutions for the sixth generation (6G) of Vehicular-to-Everything (V2X) networks. Lower frequency networks provide control of millimeter waves (mmWs) or sub-THz beam-based 6G communications. In CAVs, the mmW/Sub-THz guarantees a huge amount of bandwidth (>1GHz) and a high data rate (> 10 Gbit/s), enhancing the safety of CAVs applications. However, high-frequency is impaired by severe path-loss, and line of sight (LoS) propagation can be easily blocked. Static and dynamic blocking (e.g., by non-connected vehicles) heavily affects V2X links, and thus, in a multi-vehicular case, the knowledge of LoS (or visibility) mapping is mandatory for stable connections and proactive beam pointing that might involve relays whenever necessary. In this paper, we design a criterion for dynamic LoS-map estimation, and we propose a novel framework for relay of opportunity selection to enable high-quality and stable V2X links. Relay selection is based on cooperative sensing to cope with LoS blockage conditions. LoS-map is dynamically estimated on top of the static map of the environment by merging the perceptive sensors' data to achieve cooperative awareness of the surrounding scenario. Multiple relay selection architectures are based on centralized and decentralized strategies. 3GPP standard-compliant simulation is the framework methodology adopted herein to reproduce real-world urban vehicular environments and vehicles' mobility patterns.