Deep Learning NVR Series

AI-Powered Network Video Recorder and Video Analysis Platforms

ABOUT THE SERIES

The IBBVision Deep Learning NVR Series is a family of AI-powered video analysis platforms that go far beyond traditional network video recorder (NVR) solutions. These systems process and analyze video streams from the field, delivering meaningful data to operational processes. They simultaneously interpret live camera feeds and enable AI-powered querying of historical recordings—eliminating the need for manual replay—thereby accelerating post-event analysis and significantly reducing operator dependency.

This series comprises models capable of real-time analysis ranging from 8 to 64 channels, addressing security needs of varying scales. Each model offers a scalable structure tailored to the size and requirements of the organization.

Models in the Series:

• DLN 08 – 8 Channels

• DLN 16 – 16 Channels

• DLN 24 – 24 Channels

• DLN 32 – 32 Channels

• DLN 40 – 40 Channels

• DLN 48 – 48 Channels

• DLN 56 – 56 Channels

• DLN 64 – 64 Channels

Purpose of Use: To deliver compact, cost-effective, enterprise-grade analytical infrastructure for security and operational needs of all scales—from small and medium-sized businesses to corporate facilities, branch offices to large campuses, showrooms to production sites, warehouses to shopping malls, residential entrances to critical infrastructure—enabling intelligent, data-driven management of security and operational processes.

Remember: The IBBVision Deep Learning NVR Series is not just a passive video recording device. It is an intelligent platform that actively processes images, derives meaning from them, and generates actionable insights. The system analyzes camera feeds in real time, generates alarms, enables AI-powered queries on historical recordings, and integrates with operational decision-support mechanisms.

Traditional surveillance systems are only capable of passive recording. They suffer from critical vulnerabilities such as hours of manual video review after security incidents, missed detections due to operator fatigue, and the inability to intervene at the moment of an event. The Deep Learning NVR Series is designed to eliminate these problems.

SYSTEM ARCHITECTURE

The Deep Learning NVR Series consists of integrated layers, each performing a specific function. This layered architecture guarantees system reliability, efficiency, and scalability.

A. Input Layer (Video Stream)

• The platform receives video streams from IP cameras via the RTSP protocol.

• Channel Capacity: Support for 8 to 64 IP cameras, depending on the model.

• Protocol: RTSP compatibility, independent of camera brand.

• Stream Management: Independent analysis queue and resource allocation for each camera.

• Extendability: Modular structure allows the number of channels to be increased as needed.

B. Processing and Analysis Layer (AI Engine)

• This layer constitutes the core intelligence of the system, simultaneously processing and analyzing incoming video streams.

• Real-Time Analysis: Instantaneous object detection and behavior analysis on live streams.

• Retrospective Analysis: AI-powered target search on historical recordings.

• Adaptive Processing: Dynamically balances frame processing frequency based on scene density.

• Parallel Inference: Simultaneous, lag-free processing capacity for up to 64 channels.

• Performance Modes: Dynamic switching between low-latency, accuracy-focused, or resource-saving operating modes.

C. Alarm and Event Management Layer

• This layer interprets analysis results and delivers them to the operator.

• Alarm Engine: Evaluates events based on time, zone, and camera.

• Prioritization: Critical events are prioritized and delivered to the operator.

• Notification Mechanisms: Visual marking, instant operator notification, webhook integration, audible alerts.

• Quick Rewind: Direct access to the moment of the alarm.

• Task Assignment: Distribution of alarms among operators.

D. Storage and Archiving Layer

• This layer securely stores video recordings and analysis metadata.

• Video Storage: Continuous recording for all channels.

• Metadata Storage: Object feature vectors, timestamps, event logs.

• Fast Access: Target querying in seconds via feature vectors.

• Data Security: Encryption and backup mechanisms.

E. Management and Reporting Layer

• This layer is where the system is configured, monitored, and generates data for business decisions.

• Management Dashboard: Camera-based analysis densities, alarm distributions, model performance metrics.

• Reporting: Risk density maps, event analyses, operational planning data.

• Integration: REST service outputs, webhook, third-party VMS/CMS integration.

• User Management: Role and permission-based access control.

HOW IT WORKS

The working principle of the Deep Learning NVR Series is built upon a continuous and automated data flow cycle:

1. Video Ingestion: Video streams are received from IP cameras via RTSP. Each channel is processed independently.

2. Preprocessing: Incoming video frames are timestamped and forwarded to the analysis engine. The system dynamically balances frame processing frequency based on scene density.

3. Real-Time AI Analysis: Independent analysis engines running on each camera perform operations such as object detection, human and vehicle detection, multi-target tracking, zone intrusion detection, density and motion analysis.

4. Alarm Generation: Detected anomalies are compared against predefined rules and Regions of Interest (ROI). When thresholds are exceeded, alarms are generated and a prioritized alarm flow is created.

5. Notification and Intervention: Alarms are delivered via visual marking, instant operator notification, and webhook integration with third-party systems. Operators can directly access the moment of the alarm for rapid intervention.

6. Metadata Generation and Storage: Object feature vectors, timestamps, and event logs produced during analysis are securely stored alongside video.

7. Retrospective Querying: Users can perform AI-powered target searches on historical recordings. The system can determine where a specific person appeared, which cameras a vehicle passed through, or at what times a face was detected—all in seconds.

8. Reporting: The system produces measurable data that supports decision-making for security and operations teams. Alarm distributions, model performance metrics, and risk density maps can be reported.

All these processes run 24/7 without requiring operator intervention.

KEY FEATURES AND CAPABILITIES

Channel Capacity and Optimized Processing Architecture

• The platform processes and records video streams from 8 to 64 IP cameras, depending on the model.

• Independent Analysis Queues: Each camera has its own analysis queue.

• Dynamic Resource Allocation: Processing resources are allocated per camera.

• Adaptive Frame Processing: Frame processing frequency is balanced based on scene density.

• Low Latency: Lag-free, continuous analysis flow.

• Timely Alarms: Alarm generation without delay.

• Recording Continuity: Recording continues uninterrupted, regardless of analysis load.

• Modular Expansion: Gradual and cost-effective scaling from 8 to 64 channels.

Real-Time AI-Based Video Analysis

• Independent analysis engines running on each camera provide the following capabilities:

  • Object Detection: Detection of people, vehicles, and other objects.

  • Multi-Target Tracking: Simultaneous tracking of multiple objects.

  • Person-Based Tracking: Tracking specific individuals within the camera's field of view.

  • Zone Intrusion Detection: Detection of unauthorized entry/exit.

  • Density Analysis: Measurement of people/vehicle density in specific areas.

  • Motion Analysis: Detection of abnormal or unexpected movement patterns.

  • Face Analysis: Face detection and recognition (optional).

  • Vehicle-Pedestrian Differentiation: Detailed traffic flow analysis.

  • Performance Mode: Automatic switching between performance-focused and resource-saving modes based on resource usage and scene complexity.

AI Analysis on Historical Recordings (Retrospective AI)

• The Deep Learning NVR Series is not limited to live stream analysis:

  • Feature Vector-Based Querying: Fast search using embeddings generated during recording.

  • Target Search: Identifying where a specific person appeared.

  • Vehicle Tracking: Determining which cameras a vehicle passed through.

  • Temporal Querying: Identifying faces or objects detected within specific time ranges.

  • Reduced Manual Review: Results in seconds instead of hours of manual footage review.

  • Post-Event Analysis Speed: Instant access and rapid review for critical incidents.

Multi-Layered Security and Behavior Analysis

• The platform supports a wide range of security-focused analysis scenarios:

  • Suspicious Object Detection: Detection of abandoned or suspicious objects.

  • Abnormal Movement Detection: Detection of unusual running, fleeing, or directional changes.

  • Unauthorized Area Access: Detection of entry/exit to restricted zones.

  • PPE Compliance Check: Monitoring of safety equipment usage (helmets, vests, etc.).

  • Wrong Direction Movement: Detection of objects moving in the wrong direction in one-way areas.

  • Vehicle-Pedestrian Differentiation: Separate analysis of vehicle and pedestrian traffic.

  • Density Analysis: Instant and periodic density measurements in specific areas.

  • Risky Movement Detection: Detection of falls, fights, sudden movements, and other risky situations.

  • Queue and Waiting Analysis: Measurement of wait times at service points.

Organization-Specific Model Development and Integration

• The Deep Learning NVR platform can be customized according to operational needs:

  • Custom Model Capacity: Up to 10 custom AI models on DLN 08 and up to 20 on DLN 64 can be added to the system.

  • Model Lifecycle: Encompasses data collection, label validation, performance measurement, and version-controlled deployment.

  • Sector-Specific Customization: Secure and controlled integration of industry-specific risk scenarios.

  • Secure Deployment: Safe model deployment with controlled version management.

• Example Scenarios:

  • Field deformation detection in municipal applications.

  • Incorrect part placement on production lines.

  • Hazardous area approach detection in energy sites.

  • Suspicious package type identification in shopping malls.

ROI (Region of Interest)-Based Prioritization

• Region of Interest Definition: Critical and non-critical areas can be defined for each camera.

• Resource Optimization: Analysis density is increased at critical points, while processing load is reduced in less important areas.

• False Alarm Rate Reduction: Movements in non-critical areas do not trigger alarms.

• Sensitivity Settings: Different sensitivity levels can be defined per region.

Alarm and Event Management Layer

• Alarms generated by the system are supported by the following capabilities:

  • Visual Marking: Visual marking of the moment the event occurred.

  • Instant Operator Notification: Sending notifications to operators at the moment of an event.

  • Audible Alerts: Sound alarms for critical events.

  • Event Logging: Recording of all alarms and associated metadata.

  • Quick Rewind: Direct access to the moment of the alarm.

  • Task Assignment: Distribution of alarms among operators.

  • Automatic Recording Trigger: Initiating automatic recording upon alarm.

  • Alarm History Reporting: All alarms are reportable and integrable into task-based operational processes.

Reporting and Analytics Dashboard

• The platform produces measurable data that supports decision-making:

  • Camera-Based Analysis Densities: Analysis intensity for each camera.

  • Alarm Distributions: Time-, zone-, and camera-based alarm analyses.

  • Model Performance Metrics: Accuracy and performance measurements of AI models.

  • Risk Density Maps: Visualization of risk distribution across the facility.

  • Operational Planning: Use of generated data in business decisions.

  • Preventive Security Planning: Taking preventive measures based on detected risks.

  • Camera Placement Optimization: Improvements to camera positioning based on analytical data.

TECHNICAL ARCHITECTURE AND PROCESSING PIPELINE

The system architecture is built upon the following components:

INTEGRATION AND SYSTEM COMPATIBILITY

The Deep Learning NVR Series offers the following integration capabilities:

• Camera Independence: RTSP compatibility independent of IP camera brand.

• VMS Integration: Integration with third-party Video Management Systems.

• CMS Integration: Integration with third-party Central Management Systems.

• REST Service Outputs: Exposure of system data via REST-based service outputs.

• Webhook Support: Webhook-enabled alarm integrations.

• Custom Integration: Possibility for custom integration development if required.

• Standard Protocol Support: Compliance with industry standards.

KEY USE CASES

The Deep Learning NVR Series enables AI-powered security and operational processes in the following areas:

Small and Medium-Scale Solutions (DLN 08–24)

• Small and Medium-Sized Businesses: Cost-effective, enterprise-grade security analysis.

• Retail Stores and Boutiques: Customer behavior analysis, theft prevention.

• Branch Offices and Bank Branches: Personnel and visitor safety, unauthorized area supervision.

• Warehouses and Logistics Units: Personnel and vehicle tracking, occupational safety inspection.

• Residential and Apartment Entrances: Access control, suspicious person detection.

• Schools and Educational Institutions: Student and visitor safety, unauthorized area access control.

• Showrooms and Exhibition Areas: Visitor density analysis, valuable product security.

• Parking Entry/Exit Points: Vehicle tracking, license plate recognition (optional), wrong direction detection.

Corporate and Large-Scale Solutions (DLN 32–64)

• Corporate Campuses and Large Facilities: Comprehensive security and operational analysis.

• Municipalities and Public Areas: City security, traffic analysis, crowd management.

• Multi-Entrance Buildings and Shopping Malls: Visitor management, security optimization.

• Production and Quality Control Lines: Production process monitoring, quality control.

• Energy Sites and Critical Infrastructure Areas: Hazardous area supervision, security monitoring.

• Airports and Ports: Terminal security, vehicle and pedestrian traffic management.

• Sports Facilities and Concert Venues: Crowd density, security risk detection.

• Hospitals and Healthcare Campuses: Personnel and patient monitoring, security supervision.

TECHNICAL SPECIFICATIONS

MODEL COMPARISON

FREQUENTLY ASKED QUESTIONS

Q1: Does the Deep Learning NVR Series only record video?

A: No. Unlike traditional NVRs, this series is an intelligent video analysis platform that offers real-time AI analysis, AI-powered querying on historical recordings, and alarm/event management in addition to video recording.

Q2: Which camera brands does it support?

A: The Deep Learning NVR Series supports all IP cameras that can stream video via the RTSP protocol. It operates independently of camera brand.

Q3: Which model should I choose?

A: The choice depends on the size of your organization, the number of cameras, and your analysis needs. For an 8-camera system, DLN 08 is sufficient, while DLN 64 is suitable for a 64-camera corporate facility. You can scale up gradually based on your needs.

Q4: Does AI analysis run on live streams or recorded footage?

A: Both. The series performs real-time analysis on live camera streams while also enabling AI-powered target searches (retrospective analysis) on historical recordings.

Q5: Can I add a custom analysis model specific to my organization?

A: Yes. The platform allows the addition of up to 10 custom AI models on DLN 08 and up to 20 on DLN 64, tailored to your operational requirements.

Q6: How are alarms managed and notified to operators?

A: Alarms generated by the system are delivered via visual marking, instant operator notification, audible alerts, and webhook integration with third-party systems. Alarms are prioritized based on time, zone, and camera.

Q7: Does the system experience performance degradation with long-term use?

A: No. Thanks to its camera-based resource allocation and adaptive frame processing approach, the series maintains stable performance during prolonged operations, even as scene density varies.

Q8: Can it integrate with third-party security software?

A: Yes. The series can integrate with third-party VMS (Video Management System) and CMS (Central Management System) solutions via REST-based service outputs, webhook-enabled alarm integrations, and standard protocol support.

Q9: Is it possible to upgrade between models?

A: Yes. Thanks to its modular structure, the series can be upgraded to a model with higher channel capacity if needed. This enables organizations to scale cost-effectively during growth phases.

Q10: What occupational safety scenarios does the system support?

A: The system supports a wide range of health and safety scenarios, including PPE compliance checks (helmets, vests, goggles, etc.), risky movement detection (falls, sudden movements), hazardous area approach detection, unauthorized area access detection, and many other health and safety-focused applications.

CONCLUSION

The IBBVision Deep Learning NVR Series is an AI-powered, scalable family of video analysis platforms that transforms the traditional passive surveillance approach. It actively processes images, extracts meaning, and adds value to operational processes. With models ranging from 8 to 64 channels, it addresses security and operational needs at every scale—from small businesses to corporate facilities. Through real-time analysis, retrospective querying, custom model integration, multi-layered security analysis, and comprehensive reporting features, it provides security teams with proactive intervention capabilities while empowering business managers with data-driven decision-making.