Smoke Detection Cameras
Risk Reduction Category
Technology Description
The technology consists of a network of long-range, geo-located cameras that are mounted to existing or dedicated structures and are coupled with image processing and AI to alert on early detection of smoke. This technology has been field-proven to alert fire responders before receiving the first 911 call. Rapid confirmation of fire, size approximation, and geolocation are additional benefits of the technology compared to eyewitness reports via 911.
While many of such cameras are mounted to existing mountaintop towers, having long range views, others are designed for more remote locations where power supply and communications may be scarce. Additionally, some manufacturers couple their visual cameras with other detection means such as infrared sensing and chemical (smoke) detection.
Technical Readiness (Commercial Availability)
A few vendors are currently supplying equipment and/or software commercially with small-scale deployments. Two broad categories of vendors exist in this space: first, those who supply turnkey systems – both hardware and software, consisting of purpose-built camera/sensor systems, paired with software including an AI engine, visualization, alerting system, and web portal. A second category of vendors provides the software components only, utilizing a camera-agnostic platform that can accept any standard source of video. These vendor platforms can utilize, for example, publicly available imagery from California’s network of ALERT Wildfire cameras.
Leaders in early fire detection can support the intake of additional data such as temperature, humidity, wind speed, etc. to improve the AI and to predict severity and spread of fire.
The figure below shows a typical view from a fixed, tower-mounted camera. If desired, panoramic views can be provided by an array of fixed cameras or by motorized PTZ (pan, tilt, zoom) cameras. Image processing and AI discern smoke from fog and clouds. Although the view is long range, the typical distance for accurate detection of small fires, measuring 3’ x 3’, is about 300 yards. With increasing distance, the fire must be larger to be detected. Accuracy of detection is expressed in terms of false positives and false negatives. These systems can detect smoke or flame during daylight hours and flame during periods of low light. AI analyzes the image for smoke and produces bounding rectangles (simulated in the figure below) indicating a positive detection.
Related Technology
Vendors with Hardware and Software
Pano https://www.pano.ai/
Delphire, Inc. https://delphiretech.com/ www.wildfireai.com viewing portal (experimental)
Delphire offers a multi-sensor platform with built-in infrared detection, chemical smoke sensor. Additionally, external inputs, such as wireless soil moisture sensors, can be connected via Wi-Fi. Delphire also performs edge computing on the unit, reducing the communications requirements. Low power requirements allow it to operate from a solar panel and battery.
IQ FireWatch (Germany) https://www.iq-firewatch.com/
SmokeD
www.smokedSystem.com
Vendors with Software Only
FireScout - uses ALERT Wildfire camera network https://firescout.ai/
Chooch https://www.chooch.com/solutions/wildfire-detection/
Exci www.exci.ai
Implementations / Deployments
Pano: 33 cameras in CA, approximately 100 total (TechCrunch July 2023) [1]
Delphire: approximately 10 cameras installed during various pilots and grant-funded installations. [2], [3], [4]. Commercially available with production capability of approximately 30 units per month (phone interview with Gilberto DeSalvo 1/29/24).
SmokeD, a Polish company, has deployed cameras in Austin, TX and in California.
ALERT Wildfire
https://www.alertwildfire.org/
Publicly available imagery from approximately 175 cameras online as of 9/5/2023 across parts of California, Nevada, Oregon, and reaching into Arizona, Colorado, Washington, and Montana (see map below).
ALERT California
Originally part of ALERT Wildfire, but became a state-specific program focused on covering California. Publicly available imagery from a network of 1000+ “cameras and sensor arrays” including partner equipment such as Caltrans traffic cameras.
High Performance Wireless Research and Education Network (HPWREN)
Approximately 280 cameras reporting publicly available imagery as of 9/5/2023 with locations in southern California (see locations below). Update rate is approximately one frame per minute. Some older cameras are monochrome and fixed. Newer cameras are full color and PTZ. Archived data accessible.
Innovations as of Mid 2023
- AI layers filter out false positives like clouds and fog
- AI obfuscates private residences, humans and other non-public imagery
- Real time imagery is viewable via web portal
- Flexible Machine Learning platform that can add physical security (people detection), for example
Potential Enrichment Work Opportunity
Added data layers and data fusion such as: Wind, Weather, Vegetation Condition, Power Line Proximity to a Fire, Emergency Response Agent Locations, etc.