MURRIETA, CA - IREX, a global leader in ethical AI and intelligent monitoring based in California, USA, has announced a key update to its Fire and Smoke Detection module. The update v 4.34 delivers enhanced performance, smarter detection, and a streamlined user experience within the IREX video analytics platform. The new model allows to process a critical event in 75-105 milliseconds, marks the release as a significant step forward in both model accuracy and system efficiency, enabling earlier detection of danger, more consistent operation in poor weather or lighting conditions, and simpler setup and configuration for users.
Speed and reliability
The new model has been meticulously developed to offer superior performance and reliability in both indoor and outdoor environments. It’s designed to detect real fire and smoke as early as possible while drastically minimizing false alarms, focusing on events that truly matter rather than reacting to every flash, shadow or reflection. By analyzing how flames and smoke behave over time, the system can distinguish genuine threats from harmless phenomena such as fog, headlights or glare.
Fire, with its brighter and clearer edges, is now especially easy for the system to recognize. Smoke, which is inherently more complex, is handled by teaching the model to prioritize dense smoke rather than faint, drifting wisps, maintaining reliability in rain, fog, glare, vehicle lights, motion blur, and even raindrops on the camera lens.
The new CPU-based model processes a frame in an efficient 75-105 milliseconds and is configured to run once per second, offering fast reaction times while remaining stable and accurate.
Key model advances
To boost accuracy, IREX changed how the system “sees” fire and smoke. Instead of traditional bounding boxes around objects, the updated module uses segmentation, applying a color mask over the exact areas where fire or smoke appears: green for fire and red for smoke, thus better reflecting their irregular shapes. This approach improves the system’s ability to localize hazards precisely within the scene.
- Fire detection is the more straightforward task, since the model can recognize flames reliably due to their sharper, more defined edges.
- Smoke detection is more challenging, so the system focuses on clear, dense smoke and currently operates on 512x512 images to reduce sensitivity to minor lighting or weather fluctuations, and minimize false positives.
- To further reduce false alarms, especially indoors, operators can define Regions of Interest (ROI), so the system monitors only the parts of the frame where a fire could realistically appear, ignoring unimportant areas such as plain walls and subtle lighting changes.
- Each camera now also offers separate Smoke Sensitivity and Fire Sensitivity settings, allowing users to tune detection behavior to specific environments. A camera that monitors a busy factory floor can be configured differently from one watching a quiet hallway, ensuring that each behaves appropriately for its context.
Cleaner, easier interface
Alongside the technical improvements, IREX has simplified the user interface to make daily operations smoother. A key addition is the new “Fire Detected” event type, now fully supported with a standard criticality selector, so operators can clearly classify and escalate high-priority incidents.
The configuration experience has been decluttered by removing several confusing or outdated parameters, such as Object Size, Save Trajectory, Contrast Sensitivity, and Motion Sensitivity. In their place, the module introduces two new settings: Smoke Duration and Fire Duration. These settings define the minimum number of seconds an event must persist before triggering an official alert, helping filter out brief, harmless occurrences while still catching real incidents in time.
Behind the scenes: developer insights
Delivering this update required overcoming substantial technical challenges. To train the system properly, the IREX team generated and labeled large volumes of video realistically depicting different types of fires and smoke across varied environments and weather conditions. The team also completely redesigned one of the tools used to automatically mark smoke and fire in training videos, cutting the processing time for each frame from many seconds to roughly one second and accelerating model development.
The result is a faster, more reliable and more practical Fire and Smoke Detection module that fits with existing cameras and hardware, enabling quicker responses to potential incidents, while simplifying day-to-day operations for security and safety teams. These updates are now live for IREX customers, strengthening early fire and smoke detection as part of the company’s broader ethical AI video analytics platform.
About IREX
IREX is a global Ethical AI-powered video analytics provider committed to enhancing safety and operational efficiency.
The NIST-certified IREX platform offers a user-hosted, highly scalable solution based on secure private cloud infrastructure for public safety agencies, transportation authorities, security and emergency response teams, and critical infrastructure operators worldwide. The platform, powered by unmatched AI-driven computer vision, integrates with existing cameras to deliver real-time alerts, advanced analytics, and robust case logging based on the principles of transparency and accountability.
