Enhancing vision-based traffic crash detection performance consistency across day-night scenes: A depth-aware and...

Closed-circuit television (CCTV)-based traffic video crash detection systems require stable and consistent cross-scene performance to support all-day crash response and rescue efficiency. However, due to the substantial domain discrepancies between daytime and nighttime scenes-particularly in illumination and imaging quality-traffic crash detection still suffers from significant performance degradation when transferred across heterogeneous lighting conditions. To address this issue, this research proposed a depth-aware and domain-adaptive network built upon the Visual State Space Model (VSSM) to achieve robust crash detection across heterogeneous lighting environments. The proposed model employed a two-stream architecture that integrated appearance, motion and 3D depth information, in which the depth enhancement module captured fine-grained spatial geometry to provide complementary structural constraints, while the domain adaptation constraint effectively mitigated domain shift, thereby improving the overall robustness and reliability of crash detection. Experimental results demonstrated that the proposed model achieved a recall of 96.043 %, a miss rate of only 2.507 %, and an F1-score of 97.003 %, significantly outperforming several widely used baseline models. Ablation experiments further confirmed the critical roles of optical flow representation, 3D depth features, and the dual-level domain adaptation mechanism in enhancing spatiotemporal consistency. Moreover, the model required only 0.623 GFLOPs and achieved a real-time inference speed of 118 frames per second (FPS), demonstrating high computational efficiency. The proposed framework effectively mitigates the performance discrepancy between daytime and nighttime crash detection, and its high inference speed can contribute to faster emergency response and reduced casualty risk, offering a practical foundation for developing stable and transferable intelligent traffic safety monitoring systems.