SDG 13: Climate Action
SDG 15: Life on Land
2. Project Details
Company or Institution
Pano Rapid Detect
General description of the AI solution
Pano is a San Francisco-based company that is building wildfire situational awareness tools to help first responders and fire agencies safeguard communities. Pano’s solution, Pano Rapid Detect, combines mountaintop cameras, artificial intelligence, and intuitive software to detect the first wisps of smoke and put real-time fire images in the hands of first responders, all with the goal of enabling responders to reach fires faster and contain them while they are still small.
Wildfire cameras have been introduced in the past 5-10 years as a modern day fire lookout tower. However, these wildfire cameras have typically been used to monitor ongoing major fire incidents, rather than detecting wildfires in their early stages, because manual monitoring of camera feeds is a costly, labor intensive effort. Instead, most wildfires are detected via 911 calls, which can result in meaningful delays in the containment of wildfire incidents. This is because a large share of 911 calls about wildfires are incorrect observations, and individuals who do correctly identify wildfire smoke often struggle to provide an accurate location for the fire. As a result, smoke detection AI that can correctly identify smoke, and enable first responders to triangulate a fire’s location, will enable more rapid response to active wildfire incidents.
To address this challenge, Pano is pioneering the use of AI, to rapidly detect smoke in imagery from live camera feeds, and software tools, to quickly confirm and share incident information with emergency responders. Pano uses an object detection model for smoke identification and is iterating on this model with the goal of reliably detecting 95%+ of fires within 15 minutes, with a manageable number of false positives. The AI model is currently running live inference camera feeds across high fire risk areas in the United States.
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
There are other companies with fire observation cameras throughout the US, but Pano has made the improvement of running inference on the camera images in real-time in order to report detected incidents to customers and other stakeholders. Pano uses a computer vision object detection model to detect smoke in the cameras at its stations. Previously, other companies relied on crowd-sourcing humans to watch cameras in order to detect fires.
One of the largest challenges to creating the AI model was gathering and labeling the appropriate data. Smoke is a rare event, and it has variable characteristics, depending on the environmental conditions when it occurred. It appears as a gradient, which often does not have distinct boundaries. Additionally, there is a natural time component to the object identification, where humans have difficulty identifying smoke in a single image in early stages of a fire. To adjust to these constraints, Pano has developed a novel method of labeling smoke and incorporates a time component into detections. The training set must constantly be updated when new cameras are installed, but the holdout set to measure time to detection was randomly generated and is held constant between each new model iteration.
As the AI Algorithm is tuned for performance, Pano is actively partnering with utilities and local fire agencies for the 2021 fire season. Pilot participants include PG&E, PGE, Aspen FPD, South Lake County FPD, and Big Sky Fire Department. Pano is also participating in the EPRI Incubate Energy Labs Challenge, meeting weekly with representatives from EPRI, PG&E, PGE, Xcel, SCE, and others. Pano’s pilot in Aspen has been written about in the Aspen Times multiple times (article 1, article 2) as well as in the Wall Street Journal, Forbes and GovTech.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
In the context of the climate action SDG, wildfires are one of the biggest threats of the decade. Climate change has led to increasingly frequent and large fires, which in turn release massive quantities of carbon dioxide into the atmosphere. In water-constrained areas, the web of climate-related problems further entangles, as the water required to suppress wildfires strains local communities. While wildfires will continue to ignite, earlier detection impacts the response time and ultimately the final size of the fire and resources used to combat it.
One of the most exciting parts of the Pano solution is the wide-reaching nature of the technology. Earlier wildfire detection and better incident information can positively impact millions, whether by mitigating direct threats to their homes and lives, reducing wildfire smoke that can travel thousands of miles from a fire’s origin with serious health impacts, protecting agricultural/land-based livelihoods, or preventing massive releases of carbon dioxide into the atmosphere. Wildfire is an issue in every continent, and Pano’s rapid detection solution can be applied anywhere.
Pano is closely tracking the efficacy of its technology through user feedback and hardware and software metrics. For example, every positive detection is disseminated to relevant stakeholders, and the time-to-detection (how long it takes the AI model to detect smoke once it’s visible) is being tracked across all wildfire incidents this season. The software engineering team is also regularly updating the AI model and testing time-to-detection against the holdout data set, balancing detection time with the number of false positive detections.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
Pano is piloting with multiple utilities and fire agencies in its pilot year. The over twenty 2021 station installations were carefully selected to represent a variety of mounting and connectivity configurations, so that the learnings can be applied to the 2022 installations. As part of the pilot, Pano is closely tracking key AI metrics, such as time to smoke detection and false positives per day per camera.
Pano’s goal is to reduce wildfire risk worldwide, scaling from four states in the US this summer to multiple international locations in the next year. Pano is in conversations with its first international customers and installation partners.
Pano hopes to support a wide range of wildfire technologies, particularly AI-based ones. For example, Pano’s initial detection of a fire could be shared with a company using drones for aerial fire suppression, streamlining the end-to-end fire response process. The long-term goal is to create an ecosystem for wildfire technologies, ranging from weather and fuel monitoring, to early wildfire detection, to rapid response.
Through Pano’s 2021 pilots, the company is already interfacing with a diverse ecosystem of partners. The company is working with Stanford University, utilities, EPRI, insurance companies, private firefighters, and municipalities. While Pano’s mission is closely aligned with the climate action SDG, the company’s partners and technology directly impact many of the other SDGs. Wildfire impacts on health, clean water, infrastructure, and cities show the importance of the Pano solution and partnerships across multiple SDGs.
In addition to notifying customers of the first sign of fire, Pano will post wildfire incident feeds to Twitter, providing situational awareness information to community members. This expands Pano’s reach from the 50+ first responders using the technology this summer to thousands of community members who turn to social media for up-to-date information about nearby threats.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
Pano’s international deployment goals and social media strategy will serve to put wildfire AI in the hands of many people worldwide. While Pano can be deployed anywhere there is sufficient power, connectivity, and line-of-sight, the company is focusing on the highest fire risk areas for its earlier deployments.
The model has been trained on publicly-available data and images from Pano’s own cameras, and the majority of these smoke images were taken on the West Coast of the United States. When Pano AI is deployed to a new area with a geography that the model has not yet trained on, there will be a period of time where training data is accumulated before the efficacy of the model is on-par with its performance at known geographies.
The system is compliant with local laws and takes added precautions for privacy. All cameras have the option for physical or digital privacy filters to ensure that the cameras do not show private residences or areas where people have a reasonable expectation of privacy.
The holdout dataset that is used to measure model performance was randomly generated. While the team could have chosen fire sequences for this set that are easiest to identify, the team did not want to misrepresent the model’s accuracy.