AI Secrets Behind City Brain: Video Anomaly Detection, Person Re‑ID & Image Generation

City Brain leverages internet infrastructure and abundant urban data to perform real‑time global analysis, efficiently allocate public resources, improve social governance, and promote sustainable city development.

As Alibaba’s technical committee chair Wang Jian noted, each technological revolution—steam, electricity, now data—drives urban civilization forward, making a data‑driven “brain” essential.

Paper 1: Spatio‑Temporal AutoEncoder for Video Anomaly Detection

This work tackles the challenging problem of detecting abnormal events in video streams, which is critical for traffic safety and security monitoring. Inspired by recent advances in action recognition, the authors design a spatio‑temporal auto‑encoder and introduce a weight‑decay prediction error metric. Evaluations on real traffic scenes show the method surpasses previous best results on key indicators.

Paper 2: Deep Siamese Network with Multi‑level Similarity Perception for Person Re‑identification

Person re‑identification aims to match a pedestrian’s image across multiple cameras, a key technology for smart‑city surveillance. The proposed approach combines deep siamese and classification networks, extending similarity perception to multiple levels. Compared with major public datasets, it achieves the highest retrieval accuracy currently reported.

Paper 3: Stylized Adversarial Autoencoder for Image Generation

License‑plate recognition is a foundational model for smart cities, requiring diverse, accurately labeled data. To address data scarcity, the authors adopt a stylized adversarial auto‑encoder that merges content and style features via separate extraction networks, enabling controlled generation of realistic license‑plate images and improving GAN‑based data synthesis.

All three papers are compiled into a mini‑e‑book that can be read or downloaded for free by following the QR‑code instructions.