This article explains how Baidu Maps leverages AI and spatio‑temporal big‑data to predict real‑time parking space availability for sensor‑less lots, detailing the overall approach, feature engineering, grid computation, real‑time feature calculation, and a multi‑branch deep learning model validated at KDD.
Following the tragic 2014 Shanghai Bund crowd incident, JD Urban developed the JUST spatio‑temporal data engine, which integrates diverse real‑time sources—mobile signaling, video, traffic, trajectories, weather, POI and event data—into six unified models, applies advanced indexing and AI forecasting to monitor and predict crowd flows, enabling rapid safety interventions.
This article presents a novel method for real-time reachable area analysis that leverages recent trajectory data, introduces a Skip Graph Index for efficient query processing, predicts optimal trajectory‑splicing parameters with machine learning, and demonstrates its effectiveness through extensive experiments on multiple real‑world datasets.
The article introduces various trajectory similarity metrics—including Euclidean distance, Dynamic Time Warping, Longest Common Sub‑Sequence, Edit Distance on Real sequences, Hausdorff distance, Fréchet distance, One‑Way Distance, and Locality In‑between Polylines—explaining their definitions, advantages, limitations, and visual illustrations for spatio‑temporal data analysis.
VAUD, a visual analysis system developed by Zhejiang University and Alibaba Cloud, enables users to locate lost smartphones or track vehicles by simply dragging and dropping time‑space conditions, eliminating complex SQL queries and leveraging spatio‑temporal indexing, quad‑tree structures, and cross‑domain data inference.
