AI Maps - How is AI being used in Mapping in 2024?

2023 may very well end up being remembered as the year that AI went mainstream. Of course, AI itself as a concept has existed for decades, and many companies - including Pointr - have been making use of AI for years in various forms. However, with the rise of consumer-friendly tools like ChatGPT and the race between tech giants such as Microsoft and Google to incorporate AI into their popular services, AI is clearly here to stay.

With interest in AI and its potential uses at an all-time high, we wanted to take a look at how this cutting edge technology is being combined with one of humankind’s oldest innovations, the map.

Maps have been around for thousands of years, but they’re still continuing to evolve at a rapid pace. In recent years, we’ve become accustomed to tools such as Google Maps which can update rapidly and be overlaid with real-time information such as traffic patterns. Digital internet-based maps enable us to quickly scan vast areas of the other side of the globe and zoom into incredible levels of detail, looking at road layouts, topographical information, and more. 

Indoor maps are another relatively recent innovation. While maps of indoor locations have been possible for as long as maps have existed, they’ve lagged behind traditional outdoor maps for a number of reasons. In the early history of cartography, there was a natural focus upon charting the previously uncharted, which meant an emphasis on outdoor exploration. In more recent times, the challenges, cost and time involved in producing accurate indoor maps was often outweighed by how often indoor locations such as retail stores and airports changed layouts and rendered maps out of date and obsolete. With AI able to better adapt maps without having to start them from scratch again, as well as interact with other systems to anticipate and understand layout changes, these barriers are now being surmounted.

This is just one example of how AI is impacting modern mapping practices. In this post, we’ll explain some of the most prominent uses of AI in mapping at the moment, as well as looking at how things may develop in the future.

Google Maps - using AI to deliver a new Immersive View mapping experience

Google Maps has been without question one of the greatest achievements in mapping history. In just a few years, Google have delivered maps of incredible quality to users across the world, and then built numerous services upon them, including navigation, user reviews and perhaps most impressively, Street View, which has necessitated a fleet of specialized vehicles driving around the globe repeatedly, capturing photographs as they go to build up a visual, street-level map of a huge number of locations. 

But even with such a beloved core product developed before AI truly came to the fore in the past couple of years, Google are managing to find new and exciting ways to use AI in Google Maps to deliver exciting new features for users. One such example is their new Immersive View program, which is designed to take their classic Street View photography project to the next level.

Traditionally, Google Maps has offered both classic satellite imagery (taken from a birds eye view) and images taken from specialized cameras mounted on top of vehicles, which offer what Google calls a Street View of areas. What Immersive View seeks to do is bridge the gap between these two image datasets and provide the best of both. AI will stitch aerial satellite photography together with Street View images to enable users to truly immerse themselves in a location, all via the Google Maps interface. Users will be able to swoop inside restaurants, see predictions for traffic and weather visualized, get a more comprehensive idea of the size and scale of buildings, and get a true feel for a location before ever setting foot there.

Disaster relief - predicting wildfires before they happen & mapping historic flood levels

Wildfires are becoming a greater and greater problem for much of the world as global temperatures continue to rise, and enterprising scientists at the World Economic Forum and the Turkish Ministry for Agriculture and Forestry are using AI-powered maps to help them predict where ecological disaster could strike next.

Source: World Economic Forum

Using hundreds of different variables and AI trained upon using historical datasets, the FireAid system can help officials predict wildfires before they begin. So far, the system has achieved an 80% accuracy rating in anticipating fires 24 hours before an outbreak, a crucial window of time in which resources can be allocated and firefighters put on alert to help quell a fire before it burns beyond control.

Another side effect of climate change is the increased risk of flooding. Floods and high water levels can be inherently hard to predict, with rivers bursting their banks at different points depending on a huge range of factors that are constantly shifting over time. An example can be an area particularly prone to flooding erecting a highly effective floor defense system, which then protects that area, but creates new issues further down the same river for an area that previously hadn’t experienced flooding.

Aecom is attempting to counter this issue of unpredictability and poorly kept historical records of past floods by building a new AI-powered system that is capable of viewing photos of old floods, making an estimate of the water level depicted in the photo, figuring out where the photo was taken, and mapping it. Without AI, conducting this process on a single photo could take a human researcher multiple hours. However, Aecom’s tool is estimated to take just 15 seconds per photo. 

By building a more comprehensive database of historic flooding flashpoints, the hope is that government services will be able to take a more proactive and tactical approach to flood defenses.

Ecological maps for protecting the future of the planet

AI still has a big role to play in maps designed to tackle more long-term ecological problems. 

In order to help protect endangered species, the Spatial Planning for Area Conservation in Response to Climate Change has developed a system which leverages AI to allow conservationists to predict where different animals and fish are migrating. Migration patterns are shifting due to rising temperatures, and scientists are hoping an AI model of how migration patterns continue to develop will enable them to proactively plan conservation efforts in areas before the animals begin migrating through them.

Plant life is also critically important. Microsoft are supporting efforts to effectively monitor and map underwater plant species and organisms to ensure that the ocean ecosystem continues to function as intended. Normally, monitoring underwater plant life requires extensive and expensive manned diving expeditions, where photographs are taken. These photos are then painstakingly classified manually by a team of researchers, a process that can take hours and is open to mistakes.

However, through a combination of a smaller number of samples and high quality satellite photography, Microsoft’s Azure architecture is able to leverage AI to create a comprehensive underwater map by categorizing the sample photos and finding how they correspond to the satellite images. These maps will then help to identify shifts in patterns and pinpoint areas where conservation efforts need to be enhanced to stop destruction of ocean-dwelling organisms before the damage is irreparable. 

Fighting illegal fishing activity in the Earth's oceans

Fishing quotas are an important part of the fight to keep the delicate underwater ecosystems of the Earth's oceans balanced. However, quotas and limits are only effective when they're adhered to, which makes undocumented and unsanctioned fishing activity a major threat to keeping the oceans healthy and productive.

 Source: Global Fishing Watch

The challenge is spotting vessels operating illegally, which is akin to finding a needle in a haystack when considering the vastness of even comparatively small areas of ocean, such as the British Channel and North Sea. This is where AI comes in.

Per Nature, the Global Fishing Watch initiative was able to use various AI models to comb through thousands of terabytes' worth of satellite imagery to identify thousands of images of boats in the world's waterways. Armed with this data, the AI was then able to create dynamic models of the busiest and most frequented routes boats were taking, and cross-reference these routes and areas with the expected volume of ocean traffic and help determine how much illegal activity was taking place. The net result? A finding that around 75% of all fishing vessels are undocumented.

AI mapping models such as this will be critical in helping the authorities work with environmental charities in future to stop the worst of the illegal activity and protect the oceans for future generations.

AI boldly going where no maps have gone before - space

It’s not just on our own planet where AI is revolutionizing mapping processes. Maps of Earth are still complicated, time-consuming objects to create, having to take into account everything from topography to conflicting data across multiple sources to information stored in a range of formats. Take those challenges, and then extrapolate them by attempting to map objects in our solar system, hundreds of thousands of miles away, and you get an idea of the challenge facing scientists attempting to map other planetary and extraplanetary bodies.

Various different companies and foundations are now using AI trained on existing datasets to more accurately identify features such as craters and other landforms on planets and moons that not only inform our understanding of worlds beyond our own, but may someday form the basis of how spacecraft are designed and where they aim to land as we expand our exploratory boundaries to new worlds.

Helping to solve the riddle of indoor maps

As we touched upon in our introduction, indoor maps have proved enormously tricky over the years to get right. While they have some advantages over outdoor maps - such as the ability to work certain measurements out using geometry - these tend to be outweighed by the sheer level of detail required to make an indoor map effective.

Imagine a normal office building. On a standard outdoor map, this building would be very easy to account for; simply measuring the exterior walls, as well as the building’s distance from landmarks and roads, would allow outdoor cartography software to incorporate the building quickly. What’s more, the shape and layout of the building’s exterior are unlikely to change significantly - once it’s added onto a map, it may remain for decades without ever needing an alteration.

Compare this relatively simple task to the legion of issues and challenges faced by someone trying to map the inside of the same building. Instead of a simple geometric shape dropped onto a map, an indoor map may have to incorporate everything from seating areas to desks to walls and doors. Many of these features - such as seats, desks, and cubicle arrangements - may move regularly, to the point where an indoor map is quickly out of date. This process will then need to be followed for every single floor of the building.

There’s also the challenge of what details ought to be included, which can be ignored, and how different details populate on different zoom levels of the map. For example, someone with a far zoomed version of the map may be looking for the quickest way to get from one end of the workplace to another, and thus won’t need much in the way of information about rooms and areas along the way. Someone zoomed in to an extremely low level may be looking for features and descriptions of an individual room, such as the facilities within it (like screens or projectors) or how accessible it is. 

One Pointr client was, at one stage, having to conduct a cycle through thousands of different locations, manually charting and updating changes to the indoor maps for their retail stores one by one. This process took months and, due to the constantly shifting in-store layouts, meant that maps were quickly out of date, and remained that way until they revolved back to the front of the queue. It was this client who inspired us to create our revolutionary MapScale® tool.

Whereas most indoor maps require laborious manual work, often including on-site visits, MapScale® instead seeks to convert CAD files of a building’s layout into a fully interactive map that can then have features such as geofencing and blue dot navigation layered upon them. One of the many processes that goes into MapScale® is using AI and machine-learning to be able to accurately predict what elements of a CAD file should be included in the end product map and what can confidently be discarded, vastly reducing the amount of timely human intervention required. Different models apply to different types of buildings - for example, the elements on the CAD file that need incorporating on an office map will differ vastly from those required in a hospital map.

The future of AI mapping

As you will have seen from the examples above, AI is already having a demonstrable impact upon the way we create and use maps in a huge variety of different contexts, from outdoors to indoors and from underwater to outer space. As such, predicting just what impact AI could be having on cartography in 5, 10, or 50 years from now is almost impossible. As astounding as AI’s growth into an easy-to-use, consumer-friendly tool has been in recent years, there’s still plenty of room for improvement when it comes to AI maps:

Many of the current uses of artificial intelligence boil down to essentially being able to take tasks that humans can do, but that may take hundreds or even thousands of hours, and work through them using datasets or training data. This can not only save huge amounts of time, but also money, and means that tasks that were either acting as huge impediments to accuracy (such as the example above of the company mapping each indoor retail location one by one, resulting in maps that would be months out of date before they were next updated) or were simply too large to tackle at all (such as reading, contextualizing and mapping the images of historic floods) are now possible.

In some respects, it’s the possibility of what AI will layer on top of existing maps that represents the next quantum leap forwards. Let’s use the trusty example of Google Maps once again. Currently, Google calculates traffic volume by the number of devices with the app installed in a particular place, and relies upon user reports to show specific accidents. When traffic is particularly heavy, Google may advise users navigating through the area of alternate routes they can take to avoid traffic and save time. However, this method requires an accumulation of traffic, which takes time, and can therefore mean drivers still get caught up in queues.

With AI, systems such as Google Maps could examine years of traffic data, potentially anticipate when queues and even accidents are most likely to occur, and route drivers around them before they’ve even happened. What’s more, with systems like Google Maps enjoying such popularity, it’s possible that an AI driven model could help to ease overall congestion on the roads by deliberately routing different portions of users to different routes, more effectively balancing the traffic levels and meaning each route has a more even split cars on it.

This is just one example, of which there are almost limitless possibilities.