DiffTAD: Denoising diffusion probabilistic models for vehicle trajectory anomaly detection

Chaoneng Li, Guanwen Feng, Yunan Li, Ruyi Liu, Qiguang Miao, Liang Chang

Abstract: Highlights•A new framework formalizes anomaly detection as a noisy-to-normal paradigm.•Reconstructing clean trajectories from noisy ones and detecting anomalies.•Transformer-based temporal and spatial encoders are integrated in diffusion models.•The interval sampling accelerates the inference of diffusion models.

External IDs:dblp:journals/kbs/LiFLLMC24