IT Operations Automation · Risk Management
Shenzhen Gas had extensive camera coverage across plants and business halls, but the systems were isolated, intelligence was limited and manual patrols still left blind spots and delayed responses to safety and service risks.
Reported impact
Shenzhen Gas transformed passive surveillance into active safety and service governance, with faster detection, far more events found per day and a repeatable shared platform for multi-site operations.
Shenzhen Gas and Baidu AI Cloud built an intelligent video middle platform that turns existing surveillance assets into an operational AI layer for safety and service management. The design combines a unified video platform for collection, processing, services and administration with a unified AI management platform for data, model services, application orchestration and an AI marketplace. On top of those shared services sits an algorithm library trained for common gas-utility scenarios such as flame, smoke, clothing and helmet compliance, staff absence and intrusion detection. Video streams from stations and customer-service halls are fed into the platform, which performs real-time intelligent analysis, surfaces alerts, opens work orders and distributes events to web and mobile administrators. Instead of relying on human patrols to notice issues after the fact, Shenzhen Gas now uses a continuously learning, centrally governed video stack that shares data, models, compute and event handling across many operating sites.
Video streams from gas stations and service halls are collected by the video middle platform, analyzed by the AI algorithm stack in real time, and then emitted as alerts and work orders to administrators on large screens, web interfaces and mobile endpoints.
Baidu built an intelligent video middle platform for Shenzhen Gas with unified video, AI-management and algorithm layers (original: 为此，百度为深圳燃气打造了智能视频中台：统一的视频平台……统一的AI算法管理平台……统一的AI智能算法模型库).
Manual video inspection used to find only 1-2 issues per day, while the platform can automatically identify and alert on more than 30 events (original: 视频巡查效率从日均人工发现1-2个问题，提升为中台自动识别30多个事件并实时告警).
The platform now has 17 AI algorithms and connects 68 stations plus 58 business halls across 1,023 video streams (original: 目前视频中台已配置了17种AI算法，接入了68个场站、58个营业厅共计1023路视频数据).
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Aug 22, 2025 Siemens Energy Siemens Energy needed to automate leak and abnormal-noise inspections across complex power plants so sites would not depend on constant expert walk-throughs or full-time staffing for every inspection scenario. Hundreds of inspection types Model ensembles · NVIDIA GPUs on AWS · NVIDIA Triton Inference Server 
Microsoft Sample Aug 21, 2024 Uniper The challenge: Managing the energy transition when skilled workers are in short supply Copilot for Microsoft 365 had already demonstrated its potential for adding value as part of an early adopter… Azure · Microsoft 365 
Huawei Cloud Sample CNOOC CNOOC needed to connect previously separate offshore-oilfield systems, standardize development, and add intelligent monitoring and safety capabilities to core production and supply-chain workflows. 8 core technologies CodeArts · Huawei Cloud Stack · central control system 
