An Intelligent Spatio-Temporal Cross-Attention Network with You Only Look Once Version 9 for Adaptive Urban Traffic Management

Timely and adaptive traffic monitoring is a significant aspect of smart city architecture; however, many current models cannot understand temporal object behavior or identify critical objects in real time. Hence, this research proposes a spatiotemporal cross-attention network with you only look once version 9 (STCAN-YOLOv9), which is a new framework for smart video traffic analytics. Initially, input data were collected from the BDD100K dataset, which contains real-world driving videos. Then, the data are preprocessed using contrast-limited adaptive histogram equalization (CLAHE) and de-raining filters to process video frames taken in poor lighting or severe weather. For spatial object detection, Ghost Convolution YOLOv9 (GC-YOLOv9) was employed to achieve real-time accuracy, whereas object tracking was performed across a sequence of frames to collect the trajectory, speed, and lane priority deviation. Subsequently, the TCA module explores the features within the frames and computes a priority score for all the objects based on their behavioral context. These priority-aware insights are fused with spatial features to enable dynamic decision making, such as emergency vehicle prioritization and pedestrian congestion alerts. The proposed STCAN-YOLOv9 model achieved better results in terms of precision (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{8 2. 6 3 \%}$</tex>) and F1-score (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{7 9. 5 1 \%}$</tex>) than the existing GC-YOLOv9 model.