A vehicle trajectory prediction method and system based on spatio-temporal feature fusion of video data

Abstract

This invention belongs to the field of intelligent transportation and autonomous driving technology, and discloses a vehicle trajectory prediction method and system based on video data fusion and spatiotemporal features. The method includes: acquiring drone aerial video streams; generating vehicle temporal trajectories using target detection and tracking algorithms; converting pixel coordinates to real-world coordinates using a size prior mapping method; constructing historical trajectory sequences and performing normalization and standardization processing; generating temporally enhanced feature vectors using a long short-term memory network and attention mechanism; dynamically filtering neighboring vehicles and generating spatial interaction feature vectors through a social pooling interaction module; inputting the fused spatiotemporal features into a trajectory prediction decoder to generate a future trajectory prediction sequence; and denormalizing the prediction sequence and mapping it to the video image coordinate system for visualization output. This invention achieves end-to-end integrated perception and prediction, significantly improving the accuracy, stability, and real-time performance of vehicle trajectory prediction in complex driving scenarios.

Images