Investigation of intelligent radar-vision fusion for highway and tunnel infrastructure and cross-domain application

Millimeter-wave (mmWave) radar has become an important sensing technology for intelligent transportation, urban traffic management, perimeter security, and industrial automation because of its robustness under low illumination, rain, fog, dust, and other adverse conditions. However, most existing studies focus on single-domain optimization, while limited attention has been paid to how radar-vision fusion frameworks should be adapted under heterogeneous deployment constraints such as tunnel multipath, dense urban occlusions, security false alarms, and industrial latency requirements. This study proposes a domain-adaptive radar-vision fusion framework integrating spatiotemporal calibration, adaptive constant false alarm rate (CFAR) processing, and joint probabilistic data association (JPDA)-based multi-target tracking. The framework is validated through field deployments on the Nanning and Qinzhou sections of the G75 Lanzhou–Haikou (Lanhai) Expressway and the Fenghuang No. 1 and No. 2 tunnels on the Laibin–Du’an section of the Hezhou–Bama (Heba) Expressway in Guangxi, China, and further analyzed through structured cross-domain extension application and analysis involving urban traffic, perimeter security, and industrial sensing. The results show that the proposed framework achieves robust vehicle detection and trajectory continuity under low-light, exhaust-smoke, and multipath conditions, while domain-specific hardware configurations and algorithmic thresholds are necessary to satisfy different sensing ranges, clutter characteristics, and latency constraints. The study identifies key limitations of current 3D multiple-input multiple-output (MIMO) radar deployments. These findings provide practical guidance for the transition toward 4D imaging radar and edge-collaborative perception infrastructure.