부산대학교 도서관

Predicting lane change behavior is pivotal within Intelligent Transportation Systems (ITS) for enhancing vehicle adaptability and collision avoidance. Our innovation lies in deploying lane change prediction models in roadside Mobile Edge Computing (MEC) units for real-time predictions. We introduce a novel spatiotemporal attention model leveraging LSTM networks to extract interactive features, coupled with a Mixture Density Network for trajectory distribution. Experiments conducted on the NGSIM dataset affirm our model's superior performance. Furthermore, we delve into the lane change event detection and publication process within our MEC platform, which harnesses roadside camera-collected data. Notably, low latency in our approach establishes a robust foundation for real-time applications in ITS, rendering it a compelling candidate for future research and development.