NOVA99x

Overview

NOVA99x is an intelligent false-alarm filtering module for GCXONE surveillance systems. It processes motion-triggered events in real time and uses computer vision to classify each event as a true alarm or a false alarm before it reaches an operator.

By filtering up to 99% of non-threatening activity before human review, NOVA99x reduces alarm noise, limits operator fatigue, and helps security teams scale monitoring operations without increasing headcount — freeing up 4–5 hours per operator shift and enabling up to 2x more site coverage per operator.

One-Line Definition NOVA99x is a noise-suppression capability that filters up to 99% of false alarms, allowing security operations to scale without increasing headcount.

NOVA99x operates inline on incoming alarms, controlling what passes through the system before it reaches an operator. ZenMode then organizes and prioritizes those alarms — live and historical — into a single view operators can review and act on.

Key Features and Algorithms

Core Capabilities

Algorithmic Features

• Advanced Object Detector: Uses an object-detection algorithm that achieved state-of-the-art performance.
• Dynamic Motion Detection: Pinpoints relevant movement within an event so analysis can ignore unrelated regions and reduce false alarms.
• Tracking and Re-Identification (ReID): Adds temporal context tailored to surveillance events and helps filter objects that are not moving, especially parked vehicles.

Open NOVA99x from the dashboard to review performance:

Adjust the time range, compare true and false alarms, and filter events by camera type (thermal or optical) to understand activity across the system.

Performance Requirements: Optical (RGB) Cameras

NOVA99x performance depends on sensor quality, sensor configuration, the object being detected, and model performance.

Image Resolution and Quality

• Use an image resolution of at least 640 × 480 pixels.
• Provide clear lighting of 300–1,000 lux, or an average image intensity of at least 30%.
• Do not send frames with detection boxes or zone polygons burned into the image. These overlays change scene statistics and can drastically reduce detection quality.
• Repair or replace noisy and unhealthy cameras. Salt-and-pepper noise, heavy JPEG compression, motion blur, and damaged-camera artifacts degrade edges and increase both false positives and false negatives.

Person Detection

• A detected person should occupy at least 300 px², or at least 0.3% of the total image area.
• NOVA99x can detect smaller people, but detection probability decreases below 300 px².
• Detection probability decreases as a person moves farther away within the camera field of view.
• Low-light and low-contrast regions reduce accuracy. Illuminate the area of interest according to the image-intensity requirements above.
• Cameras below 336 × 190 pixels may contain blind spots and produce lower accuracy in those regions.
• Occlusion, motion blur, glare, rain, and lens contamination distort object edges and textures and significantly reduce reliability.
• Detection is less reliable when a person is at the edge or corner of the frame, or when less than 50% of the body is visible.
• Distortion from low-resolution panoramic or 180-degree cameras can alter a person's dimensions enough to prevent detection.
• IR cameras may not detect people correctly when they are close to bright lights.
• IR-reflective vests can over-reflect infrared light, oversaturate a person's features, and reduce the detection rate.
• IR imagery requires good IR illumination; people outside the illuminated area may not be detected.

Vehicle Detection

• Vehicle types outside the supported list have not been tested or trained.
• Fast roads and sharp turns: Set carBypassDisplacementCheck = true when vehicles appear briefly, pass very quickly, or make sharp turns. This bypasses ReID and relies on motion-box intersection with detected objects.
• Parking lots: Do not enable carBypassDisplacementCheck. Keeping displacement checks active filters as much vehicle jitter as possible.
• Device-provided motion boxes, including those supplied by supported devices such as ADPRO, take priority over generated motion boxes.
• Adaptive frame differencing is the default motion detector. Optical flow is more sensitive for subtle motion, thermal imagery, very fast vehicles, or vehicles visible in only one frame, but it carries a higher false-alarm risk.
• Spider webs, insects, reflections, automatic gain control (AGC) shifts, and water droplets can trigger false motion alarms, especially in parking lots. Clean lenses regularly and reduce frame intervals to limit these alarms.

Night Detection

• Cameras without IR rely on ambient or built-in illumination. They suit indoor use but may require additional outdoor lighting.
• IR night vision has a reduced effective field of view. A person should occupy at least 600 px²; detection probability decreases below this size.

Detection Confidence

detection_confidence controls the detection and filtering threshold. Set it below 0.05 when sensitive detection is more important and operators can tolerate more false alarms. Set it to 0.4 or higher when stronger filtering is preferred and objects are expected to be close to the camera.

Performance Requirements: Thermal Cameras

• Configure the camera as Thermal in GCXONE so NOVA99x uses the correct analysis model.
• A detected object should occupy at least 600 px²; detection probability decreases below this size.
• Thermal imagery has lower contrast and smoother textures than optical imagery. Excessive CLAHE or AGC enhancement can introduce artifacts and false detections and can be disabled without preventing detection.
• If a thermal camera produces false positives because of poor preprocessing, set useContrastEnhancement = true for that sensor to test whether contrast adjustment resolves the issue.
• Humans generally appear brighter than surrounding objects and are detected more sensitively than vehicles because thermal detection relies on temperature gradients.
• Small or distant heat sources can become uniform white blobs in low-resolution imagery, especially when the thermal gradient is large, reducing detection precision.
• Chimneys, vents, heated pipes, and bright CLAHE artifacts can resemble a person's vertical profile. Reposition the camera or tune thresholds to reduce false person detections.
• Cats and other animals can resemble humans in thermal imagery and may be classified as people.
• Vehicle ReID still requires the vehicle in at least 2 of 3 frames. For fast, sharp turns on curved roads, set carBypassDisplacementCheck = true.
• Use a white-hot, black-cold configuration. Other thermal color mappings can cause false negatives.

Unsupported Cameras and Configurations

• Fisheye cameras, because of edge distortion and very high resolution requirements.
• PTZ or dome cameras that change the scene; NOVA99x requires a static scene to estimate motion correctly.
• Dual-sensor thermal/optical fusion. NOVA99x treats each channel as either thermal or optical and does not stack the channels.
• Thermal color palettes or heatmaps other than white-hot/black-cold.
• Cameras that zoom onto moving objects, because the motion will not be detected correctly.

Entity-Level Operation

NOVA99x can be applied at multiple levels of the GCXONE hierarchy so filtering matches the way an operation is managed.

Inheritance Lower-level entities inherit NOVA99x behavior from higher levels unless explicitly overridden.

Activation and Configuration

Prerequisites

• Administrator access to the GCXONE tenant.
• Permission to manage the selected entity level.

Subscribe and Enable NOVA99x

1. Log in to GCXONE as an administrator.
2. Open Marketplace from the main menu.
3. Select Apps.
4. Find NOVA99x and select Explore.
5. Select Proceed.
6. When asked to save the subscription, select Yes.

NOVA99x appears under My Subscriptions and is enabled for the tenant.

Configure Filtering Scope

Select the entity where filtering should apply. Administrators can apply it globally or selectively. Higher-level configuration cascades downward unless a lower-level entity overrides it. Changes take effect immediately for incoming alarms.

Fail-Safe Alarm Handling

NOVA99x Unavailable or Delayed If NOVA99x is unavailable, cannot complete verification, or does not respond within the agreed service-level agreement (SLA), GCXONE treats the alarm as a true alarm and presents it to the operator.

This fail-safe prevents potentially critical events from being missed. Alarm-verification capability depends on NOVA99x availability and timely response.

Business Value and Use Cases

Common Use Cases

• Service Provider-Wide Noise Control: Enforce baseline filtering across all customers so a noisy deployment cannot overwhelm shared teams.
• Customer-Level Alarm Policies: Match one customer's environment and risk tolerance without affecting others.
• Site-Level Environmental Filtering: Address weather, lighting changes, or routine activity unique to a location.
• Device and Sensor Precision Control: Tune a problematic camera or sensor without changing the rest of the site.
• Large-Scale Expansion: Stabilize alarm volume while onboarding customers or sites to prevent early overload.

Roles, Scope, and Limitations

Roles and Responsibilities

Scope and Limitations

• NOVA99x filters incoming, motion-triggered alarms within GCXONE and supports entity-level control.
• It does not replace detection analytics, camera hardware configuration, or physical security measures.
• It does not guarantee that every false alarm will be removed.
• It does not replace human judgment, live monitoring, or compliance rules.
• It is not a reporting, investigation, or predictive-analytics tool.

Relationship to ZenMode NOVA99x controls alarm noise before alarms reach operators. ZenMode then organizes and prioritizes those alarms — live and historical — into a single view operators can review and act on. Together they support real-time scale and long-term optimization.