Millimeter-wave (mmWave) vehicular-to-everything (V2X) links are highly vulnerable to sudden blockages in dense urban traffic. Since terrestrial roadside links can degrade rapidly, and alternative ground paths are often limited, maintaining reliable service with only ground networking resources remains challenging. To enhance link reliability by exploiting aerial relay resources in air–ground integrated networks, this paper proposes a vision-assisted unmanned aerial vehicle (UAV) relay triggering framework. The framework uses roadside multi-camera images to predict the future link state of a target vehicle and triggers a UAV decode-and-forward (DF) relay before the direct roadside-unit (RSU)–vehicle link becomes unreliable. To enable target-specific prediction, a template-guided image-matching module is developed to localize the target vehicle in multi-view images. The matched features are fused and temporally modeled to predict future LoS, NLoS, and Absent states, with the predicted NLoS probability further used to determine the UAV activation decision through a probability-based triggering policy. Simulation results on a 3D ray-tracing urban V2X dataset show that the proposed dual-view predictor achieves about 99% validation accuracy, compared with about 87% for the single-view baseline. The proposed relay triggering scheme reduces the outage probability from 15.08% for RSU-only transmission and 4.49% for reactive relaying to 0.76%, and improves the 5th-percentile rate from 11.72 Mbps to 22.49 Mbps over reactive relaying.
Accurate wireless positioning has remained challenging under dynamic bandwidth conditions and outdoor multipath environments that are typical in Internet of Things (IoT) and autonomous aerial vehicle (AAV) applications. Conventional learning-based localization methods rely on bandwidth-specific channel state information (CSI) representations, which causes the trained models to be inapplicable or less adaptive when the signal bandwidth differs from that used during training. To overcome this limitation, a unified and neural network-oriented framework is proposed, which constructs bandwidth-adaptive power delay profile (PDP) representations for learning-based models. A PDP preprocessing scheme through adaptive zero-padding and oversampled IFFT of heterogeneous CSI is introduced to generate dimension-consistent and delay-aligned neural network inputs. To enhance robustness, a sub-band-sliced PDP representation is developed to enhance model robustness, where each bandwidth is divided into equal-width sub-bands whose PDPs are independently processed and organized as Transformer tokens. A dedicated Transformer is designed to get the location estimation from PDPs of multi-access points. Simulation results have demonstrated that the proposed preprocessing-PDP-plus-Transformer framework achieves superior cross-bandwidth generalization and localization accuracy, compared to analytical and learning-based baselines.