Navigation and Content

Non-Logistics Use Cases: Disaster Response and Recovery

Computer Vision in Disaster Response and Recovery

From droughts and wildfires to floods and landslides, earthquakes, hurricanes, pandemics, and more, computer vision and machine-learning algorithms can predict certain occurrences before they happen and provide essential warnings. In the immediate aftermath of an uncontrollable event, this technology can assist first responders and recovery experts on the ground, helping them make urgent assessments, scope the damage, and strategize rescue efforts. And over time it can help to coordinate effective reconstruction.

Current Trends

COVID-19 showed everyone the serious implications of being unprepared, from shortages of emergency supplies and other vital resources to the risks of disease transmission. Swift responses are required to the increase in uncontrollable events which can impact communities for months and even years afterwards. Effective event prediction and early warning, supported by computer vision are saving lives.

If a heatwave is predicted 24 hours in advance, this can help reduce ensuing damage by 30%.

Here we explore three important computer vision disaster and response recovery applications.

Monitoring Environmental Conditions

Weather models help to improve the accuracy of weather forecasting and prediction. By combining visual data from multiple sources, such as satellite images, weather stations, and ground-level observations, computer vision algorithms can vastly improve and increase how comprehensive and effective these weather models are. 

Computer vision can help detect patterns in visual weather data, such as changes in cloud cover, temperature trends, and changes in water levels. These patterns can enable meteorologists to predict future weather conditions and help governments and communities provide early warnings to keep everyone safe. For example., IBM and NASA have launched the first open-source geospatial AI foundation model for Earth observation data. Its wide range of potential applications include tracking changes in land use, monitoring natural events, and predicting crop yields.

Based in France, the startup TENEVIA uses computer vision image analysis and numerical modelling to measure, monitor, and forecast environmental conditions. With camera hardware and simulator software, the solution helps forecast high water levels and snow – the cameras record flows, and the simulator analyzes this visual data. Together they create a virtual fence around any flow to analyze water and snow dynamics – and can indicate irregular water accumulation as well as melting glaciers and snow to give advanced flood warnings.  

Betterview in California specializes in providing risk information to the insurance industry. To get an immediate, complete picture of wildfire risk, for example, the startup combines aerial imagery, computer vision, and third-party property intelligence. Insurers use this information to streamline efficiency, as well as predict and prevent wildfire-related losses.

Optimizing Relief Resources

Before each response demanding event, it is extremely important to pinpoint where it will occur and predict the likely damage. Computer vision technology can help detect and track the path of hurricanes, volcanic lava flows, and wildfires so that a plan can be made to promptly optimize relief resources. Of course, it also enables rapid and effective evacuation planning.

AiDash, based in India, provides pre-event computer vision solutions to help predict, prepare, and deploy accurately using satellite imagery, real-time weather data, and variations in vegetation data. This detailed intelligence enables organizations to estimate resource requirements in advance.

Measuring Natural Disaster Impact 

Computer vision technology can help identify and measure the impact of uncontrollable events. Large satellite image datasets of vegetation and land cover can be sourced from specialist companies. By applying deep-learning methods and high performance-based models, affected areas can be assessed and monitored over time, incorporating recent imagery to recognize any changes. To train the AI model, pre-event images are used, and AI algorithms are applied to identify damage and estimate repair costs.

During Japan’s typhoon season, homeowners have been taking and submitting smartphone pictures of any property damage caused by a storm, using an AI solution from Tractable, a New York technology startup. With this simple app-based method, there’s no need to send out an appraiser in the wake of a damaging event. Instead, using computer vision technology and machine-learning techniques to enable visual assessment, the insurance company can use these photos for fast, accurate damage appraisal to accelerate each customer’s recovery from impact.

Challenges of Implementation/Realization

Although these use cases are drawn from various industries and operations, for the most part, they share some common challenges. 

A key challenge is the persistent perception that artificial intelligence generally, and computer vision specifically, may be less accurate and less capable than a human. The right balance must be found between extracting meaningful insights, ensuring adequate defenses against hacking and manipulation that can cause AI systems to misclassify objects, and protecting individual privacy.

Wherever cameras are involved, there will always be valid concerns about data privacy and security. In some applications, particularly in the area of healthcare, there is the issue of false positives – a patient may be diagnosed with a disease they don’t in fact have. 

Implementation barriers may arise due to the trustworthiness of AI, if experts cannot explain decisions based on machine learning. These and other challenges, including the investment needs of technology implementation, are tending to slow the uptake of computer vision technology.