"As the global video analytics market scales rapidly—projected to grow from $12 billion in 2024 to over $38 billion by 2030—organizations face a major bottleneck: processing massive volumes of video and physical camera streams alongside enterprise cyber data without driving up latency, bandwidth, or cloud GPU infrastructure costs"
Equitus Expert Labs (EEL) Assesses Video Analytics Solution Fit
" ARCXA's Semantic Control Plane deployed on IBM Power 11, transforms IBM EVS video feeds and core enterprise data into unified 3-column RDF Triples. Backed by KGNN reasoning, this unlocks true cyber-physical fusion—empowering your core systems to automatically reason over physical security, logistics, and sales flows with zero compromise to enterprise performance or compliance."
EEL Proposes: IBM EVS, combining Equitus intelligent data ingestion products bridging the expanding cyber and physical infrastructure intelligence gap into a unified core operational stack. IBM EVS: Bridges and Extends, not replaces your current CYBER/PHYSICAL Infrastructure. IBM EVS’s GPU interface, expands Power10/11’s on-chip MMA acceleration to GPU-enabled processing, rather than MMA being solely an alternative to GPUs. Contact Us to start Evaluating your Project Scope, Goals and Timeline.
Section I. Accelerated Deployment via Equitus Expert Labs (EEL)
EEL helps organizations rapidly capitalize on the expanding video analytics market—without suffering from protracted integration cycles or complex IT/OT friction—Equitus Expert Labs provides a streamlined, turnkey pathway from initial architecture to operational readiness.
1. Turnkey Delivery & Services
End-to-End Execution: Equitus Expert Labs manages consulting, architecture mapping, container deployment, integration, and performance testing.
Rapid Readiness: Pre-configured deployment blueprints significantly reduce installation and validation time, allowing organizations to go live in days rather than months.
2. Pre-Configured Hardware & Platform Bundles
IBM Power Hardware Optimization: Specifically engineered for deployment on IBM Power IC922 and IBM Power S1022 servers. By harnessing IBM’s native Matrix Math Acceleration (MMA), the platform achieves real-time video inferencing without necessitating separate GPU accelerators.
Cloud-Native & Hybrid Orchestration: Standardized on Red Hat OpenShift (as well as SUSE Rancher), enabling consistent container management across on-premises edge nodes, private data centers, and multi-cloud environments.
3. Seamless Infrastructure Integration
Identity & Access Management: Direct integration with Active Directory / LDAP for unified role-based access control (RBAC), multi-factor authentication, and enterprise governance.
Video Ecosystem Compatibility: Connects natively with existing IP camera networks and enterprise Video Management Systems (VMS) via standard protocols (ONVIF, RTSP), preserving prior hardware investments.
Enterprise-Scale Ingestion: Scales intelligently from local multi-camera setups to thousands of high-definition video streams across distributed enterprise, defense, transit, and public safety environments.
________________________________________
III. IBM EVS - Video Ai Inter-connection
Equitus Video Sentinel gives IBM Power10/11 customers a practical entry point into video intelligence using built-in MMA acceleration, with a path to expand into GPU-accelerated multimodal analytics through IBM’s forthcoming GPU interface. Customers can start with existing Power infrastructure and extend processing capacity as their cyber/physical security requirements grow.
IBM EVS (Ingestion & Stream Orchestration): Acts as the foundational video bridge. It captures live feeds from disparate cameras and VMS architectures, standardizes stream formatting, and allocates frame chunks to the appropriate vision pipeline.
Ultralytics YOLO (Fast Localized Detection): Handles first-pass, low-latency bounding-box detection (identifying humans, backpacks, vehicles, or weapons) at frame rates required for real-time alerting.
Meta SAM 2 (Precision Segmentation & Tracking): Takes object prompts from YOLO and performs zero-shot spatial-temporal segmentation. SAM 2 tracks exact pixel masks across video frames even through occlusions, cluttered environments, or erratic movements.
Axelera RelateAnything (Open-Vocabulary Relational Mapping): Evaluates image regions from SAM 2/YOLO and applies open-vocabulary predicates to map real-time interactions (“Person X is concealing Object Y” or “Person A is handing item to Person B”). Accelerated via hardware like Axelera's Metis/Europa AIPUs, this executes edge intelligence with high energy efficiency.
Equitus KGNN & SPARQL (Semantic Fusion): KGNN can translate 2 column data into triple store RDF Knowledge Graphs and Ingests the structured relational metadata produced by the vision models and binds it to enterprise data ( core financial logs, access control records, and asset inventories) using schema-less knowledge graphs queried via SPARQL.
IV. Tailored Cutting-Edge Use Cases
1. Advanced Theft Protection (Retail Loss Prevention & Asset Tracking)
The Fusion: EVS streams store feeds to YOLO to spot localized concealment behavior. When a suspicious motion or item handling is flagged, SAM 2 segments the exact object (e.g., high-value retail electronics) and tracks its boundary as the suspect moves through the store.
Relational Reasoning: RelateAnything computes dynamic predicates between the individual and the inventory item ("holding", "concealing inside bag").
Enterprise Action: Equitus KGNN instantly cross-references this visual event with POS (Point of Sale) logs and inventory tracking databases. If no payment event matches the timestamp via a SPARQL query, an automated, evidence-backed alert is sent directly to security personnel.
2. Anti-Money Laundering (AML) & Physical-Digital Fraud Fusion
The Fusion: AML is traditionally limited to database text logs (wire transfers, account flags). This architecture bridges physical branch video with banking systems. EVS monitors high-value teller lines or ATM vestibules.
Relational Reasoning: YOLO and RelateAnything detect behavioral anomalies—such as an individual repeatedly switching physical burner phones, handling multiple unassociated debit cards, or coercing a customer—and map the physical actors involved.
Enterprise Action: The KGNN binds the visual identity/biometric vector of the suspect to core banking ledger data, flagging accounts showing sudden structuring or rapid cash-out behaviors that correlate precisely with physical surveillance triggers.
3. Missing Persons & Vulnerable Subject Searches
The Fusion: Rapidly locating a missing child or vulnerable adult across a sprawling enterprise campus, smart city, or transit hub.
Precision Tracking: EVS ingests feeds across multiple disparate cameras. When an operator inputs a description or reference image, SAM 2 provides robust video segmentation to maintain tracking continuity as the subject moves across different camera angles with changing lighting or clothing partial occlusions.
Relational Reasoning: RelateAnything evaluates contextual relationships ("accompanied by unfamiliar adult", "entering restricted maintenance zone").
Enterprise Action: The system updates the central Knowledge Graph in real-time, executing a SPARQL query to map the subject's projected path against access control logs and historical foot-traffic data to guide first responders directly to their current location.\
Equitus Video Sentinel (EVS) provides large‑scale, on‑premises video analytics on IBM Power11 systems using AMD Instinct GPUs (MI210 today, MI350 expected). It turns live camera feeds into searchable events and real‑time alerts without relying on the cloud.
Validated Performance
~100–200 active cameras per AMD MI210 GPU in baseline testing.
2× expected capacity on MI350 (not yet benchmarked).
Performance varies by resolution, frame rate, and analytic profile.
Key Capabilities
Real‑time analytics:
Person & vehicle detection
Classification & re‑identification
Cross‑camera tracking
Behavioral analytics (tripwire, loitering, left‑object, parked‑vehicle, weapons detection)
Alerts:
Real‑time notifications with keyframes and metadata
Forensic search:
Search across live and recorded video
Data layer:
Sentinel API
PostgreSQL
S3‑compatible storage
Architecture Overview
Input: Fixed, PTZ, mobile cameras; RTSP; VMS integration; uploaded video files
Compute: IBM Power11 + AMD GPUs
EVS ONNX inference engine optimized for MI210
Containerized stack running in a Power LPAR using ROCm
Deployable via Helm on OpenShift or Kubernetes
Operators:
Alert feed, search, playback
Keycloak SSO and role‑based access
Why IBM Power
Run analytics next to systems of record
Hardened platform with GPU acceleration for large camera counts
Power11 MMA enables smaller or edge deployments
Fully on‑prem, no cloud dependency
No comments:
Post a Comment