STAI
A Smart Video Reasoning & Analytics Solution to Real-Time Detect & Track Across A Network of CCTVs.
“Everything we hear is an opinion, not a fact. Everything we see is a perspective, not the truth.”
— Marcus Aureliu
- Widespread deployment of surveillance cameras, but low utilization and inability to analyze/convert into structured image event data.
- CCTV video surveillance heavily relies on manual intervention, making it difficult to promptly detect events.
- It is challenging to identify anomalies from a single camera view, leading to potential misjudgments and false alarms.
- The large volume of data makes real-time detection, alerting, and tracking challenging.
- Post-event investigation involving multiple camera footage is time-consuming and often ineffective.
The mission of STAI is to continuously analyze a sequence of actions from vast real-time video data, identify meaningful complex events, understand the context of these events, and address four key security-related questions pertaining to time and space.”
What Happened? When & Where?
Who Was/Were Involved?
Where Is/Are The Person(s) Now?
What May Happen Next?
Pain Points
Common Pain Points in Recognition
- Facial Recognition: Privacy concerns.
- Non-Facial Recognition: Unable to track beyond the camera’s field of view.
- The security loophole lies in time-consuming & labor-intensive post-event search & tracking.
Re-ID
Integrate video data from large-scale facilities such as amusement parks, smart buildings, factories & shopping centers; automatically detect the occurrence of events, classify them by severity & provide real-time notifications.
Image Inference
Extract valuable insights from a continuous stream of vast, chaotic & partial video footage; automatically infer the cause & consequences of events & promptly identify key frames & associated individuals or objects.
Smart Footprints
Transform critical imagery into actionable real-time data, employ non-biometric recognition techniques for swift cross-zone & cross-floor searches, map ot movement paths & track the locations of related individuals or objects.
Technology Architecture
- Analyze hundreds of CCTV videos in real time, using AI to construct event’s spatial and temporal data for complex events detection.
- Via Live Data Solution, location and time data (spatial / temporal) can be stored and prepared for further analysis.
Video Stream (Unstructured)
- Object Detection
- Action Detection
- Re-ID Featuring
AI
Extraction
Video Frame Database
Video Event Data (Structured)
Spatial-Temporal Information Tagging
GIS
Processing
Video Event Database
Analysis
Association & Causality
Reasoning
Spatial-Temporal Data Search
