Drone Imagery: From Capture to Automated GeoAI Analysis
Key takeaway
Reality mapping enables us to create 2D and 3D digital representations of the physical world to use with geographic information system (GIS) technology and gain real-world context for spatial data. To take it a step further, we can apply geospatial artificial intelligence (GeoAI) to help extract information from these representations more efficiently, accurately, and at scale. Through a set of demonstration videos, you’ll learn how it all works, including training and using a deep learning model, leveraging pretrained AI models, and exploring real-world applications in a variety of industries.
Reality mapping is defined as a process of creating a spatially accurate 2D and 3D digital representation of the physical world using images or lidar. It’s very beneficial when using geographic information system (GIS) technology for establishing foundational content—where it isn’t available yet—to gain a realistic view of critical sites and begin problem-solving. So by using drones to capture current imagery, we gain real-world context for spatial data. We can then layer additional data points over the digital representation for deeper analysis and to facilitate a holistic view for a wide variety of stakeholders.
Drone imagery analysis with GeoAI
Geospatial artificial intelligence (GeoAI) is the integration of artificial intelligence (AI) with spatial data, science, and technology to increase understanding, solve spatial problems, and automate information extraction.
With ArcGIS, you can analyze a wide variety of imagery—anything from a point cloud to oriented imagery to video. Perform a variety of tasks with imagery including change detection, site suitability, vegetation monitoring, object detection, image redaction, or feature extraction and classification.
For public and commercial property management, organizations can use still or motion imagery to detect, classify, and count vehicles to learn when parking lots reach capacity. For utility right-of-way monitoring, point clouds can be used to classify utility assets and identify vegetation encroachments in the network to help prevent wildfires. In environmental research, orthoimagery or video imagery and classification algorithms can detect changes in animal populations over time to understand how they are evolving.
Drone imagery analysis can be performed on a variety of data types.
Workflow: Training and using a deep learning model
The time-saver: Pretrained AI models
Esri now provides over 40 pretrained deep learning models that are ready to use in ArcGIS Living Atlas of the World. These pretrained deep learning models eliminate the need for huge volumes of training data, massive compute resources, and extensive artificial intelligence (AI) knowledge. They enable you to accelerate your geospatial workflows with built-in expertise and resources designed specifically for image feature extraction, land-cover classification, image redaction, and object detection. Automate the way you extract meaningful insights from imagery, point clouds, and video.
Pretrained deep learning models from Esri include car detection, building detection, land-cover classification, power line detection, and pavement crack detection. Simply grab a pretrained model of your choice and run it on your imagery using ArcGIS Image for ArcGIS Online or ArcGIS Pro. You can then fine-tune it to fit your needs and location. No training is needed.
Follow these steps to begin training and using a deep learning model.
Watch the following demo videos to learn how AI and drone imagery are being used in different industries to solve complex problems.
Demo 1: San Bernardino International Airport uses AI to identify pavement failure
The San Bernardino International Airport flies drones at its new Unmanned Aircraft Systems Center. The airport’s miles of concrete taxiways and runways need to be evaluated regularly for safety. While historically this has been done in person, requiring hundreds of hours of on-the-ground survey work, today it can apply drones and artificial intelligence algorithms to expedite the process. Watch this demo to see how AI and deep learning are used to identify pavement failures and quickly examine runways and taxiways for cracks.
Demo 2: Drone imagery and machine learning help identify flood risk in Belize
Coastal areas in Belize are situated on low-lying terrain, which enhances its risk for floods. Using drone imagery and pretrained deep learning models, we can look at the vulnerability of structures—not only their proximity to the ocean but also their elevation, building materials, and a variety of other features. Using Belize as an example, this demo walks through the process of uploading imagery into ArcGIS Online using new capabilities within ArcGIS Image for ArcGIS Online to quickly begin analysis in the cloud—no coding required.
Demo 3: Automate fire damage assessment with deep learning
In 2018, the Woolsey Fire burned nearly 97,000 acres of land around Los Angeles, California, in 15 days. Surveying the scale of damage was essential to disaster relief. Previously it would take days to manually classify aerial imagery, but now it only takes hours thanks to pretrained machine learning models and automated AI. With the Woolsey Fire example, this demo walks you through the process for training, creating, and applying your own deep learning model.
Need help finding the right solution?
Schedule a conversation with one of our experienced sales consultants. Tell us how you're using imagery and remote sensing data now, and we'll show you where a comprehensive geospatial system can take your work next.