Application of Improved LSTM with Fusion Attention in Security Situation Awareness of Intelligent Transportation

Traditional intelligent traffic safety situation awareness methods are unable to meet the requirements of complex scenarios due to weak suppression of environmental interference, high target miss detection rates, and insufficient real-time performance. The mainstream methods suffer from the bottleneck of low frame rate caused by redundant computations, and the lightweight architecture has a low recall rate in occlusion scenarios. Therefore, this study proposes a model based on the fusion of attention mechanism and bidirectional long short-term memory network to solve these problems. By compressing redundant computations and enhancing the ability to capture spatio-temporal dependencies, the model addresses these issues. Experimental results show that the model achieves an average accuracy of 98.2% in congested, high-speed, forested and elevated bridge scenarios, which is 7%–16% higher than that of traditional models, and the error is stable at below 0.02. This proves that the model has excellent accuracy and applicability in complex traffic environments, effectively solving the problem of coordinated optimization of environmental interference, target missed detection and real-time performance, providing a new idea for intelligent traffic active safety prevention and control, and helping traffic management develop towards intelligence and efficiency.