Why AI in reality capture does less than you think – but matters more than you realize

The capabilities of artificial intelligence (AI) have been overhyped in several areas, leading to unrealistic expectations of the technology. Curiously, in other areas, its accomplishments are overlooked. While AI isn’t doing everything in reality capture, the part it plays is truly transformational.

No less than 80% of surveying and geospatial professionals believe that AI, machine learning and real-time analytics will significantly shape the industry in the coming years, according to the latest GIM International Business Guide survey. Many view AI as a game-changer for the geospatial industry, with promises of transforming workflows and redefining reality capture. In some areas, that is true. In others, it isn’t.

I see AI as a targeted tool that, when applied correctly, can drive levels of efficiency which were not previously possible. However, deploying AI everywhere is not realistic. So, evaluating where it adds value is essential. We must develop a clear understanding of what AI can achieve, both right now and in the foreseeable future.

AI is rarely the whole answer

AI is currently best used in select parts of the reality capture workflow. Reality capture still relies heavily on traditional algorithms, established processing techniques and human expertise. Take point cloud classification, for example. AI can now automatically recognize and categorize objects or features within point cloud data, increasing efficiency by reducing the long hours needed to clean data in the office. However, someone still needs to check the data for accuracy, because AI can make mistakes and misclassify objects. I remember an occasion when we trained an early AI model on vegetation data in Zurich, yet when we then used it in Las Vegas, it thought some of the palm trees were columns on buildings. This wasn’t the fault of the AI; it just hadn’t been trained on that specific vegetation or environment type. I would advise users to understand where AI makes decisions on their behalf and consider the impact of these decisions on the accuracy or the end results.

Rather than replacing existing workflows in reality capture, AI enhances them. AI algorithms can process huge datasets from large geospatial scans, providing useful insights or patterns that a human would miss. Additionally, AI can save users time by removing unwanted elements from point clouds or distinguishing between different material types, which usually takes a lot of manual effort. In reality capture, AI can significantly improve core workflows, optimize efficiency and unlock new possibilities in data analysis and visualization. But it needs human guidance.

Smaller, smarter AI is the way forward

Knowing AI’s limitations means recognizing that a one-size-fits-all approach isn’t the best way forward. AI is often mistakenly seen as a universal solution that should be able to solve any problem. However, this misconception has resulted in 3D models that require optimization to fully deliver on their promises. Instead of trying to do everything, AI is most valuable when it is used to focus on doing one thing exceptionally well. In reality capture, targeted, specialized AI models are far more effective than large, general-purpose ones. Take point cloud cleaning, for example. Instead of relying on a single AI model to classify an entire dataset, targeted AI models have been developed that tackle distinct challenges: one efficiently removes stray points from dust or sensor noise, another helps separate land types even in complex environments, and another detects and eliminates moving objects like cars and pedestrians while keeping fixed objects in the scene.

Developing specialized AI solutions that focus on one, specific task is starting to add a lot of value. This narrow AI approach not only improves precision and efficiency, but also reduces processing demands and leads to the faster creation of more accurate 3D models. For geospatial users, this means focusing on specialized AI tools that streamline time-consuming processes, rather than expecting a single system to do everything. In other words, it’s about picking the right AI tool for the job at hand.

Seeing the unseen: AI's role in overcoming data gaps

AI is also increasingly being used to fill in missing data – but can we trust the outputs? In reality capture, occlusions are areas that cannot be directly scanned because they are blocked by physical barriers like walls, furniture or vehicles. This leaves gaps in the model and can reduce its accuracy. Modern AI, however, has the potential to complete the model, using pattern recognition to predict what is behind an obstacle.

During a scan, AI may assume that a wall continues behind an obstructing pillar based on context clues in the surrounding area. However, this can also lead to incorrect assumptions or inaccurate models. While this can be useful for visualization purposes during the early stages of a project – like site planning, urban design or interior layouts in construction – it would likely be too risky in situations when accuracy is essential, such as compliance surveys or structural assessments. For applications that demand high levels of accuracy, inferred data must be clearly distinguished from real, verified data. AI’s role in these instances is not to fabricate reality, but to provide additional insights while clearly showing what has been captured and what AI has inferred. As AI becomes more integrated into reality capture workflows, geospatial professionals will need to adapt, learning new skills to interpret and validate AI-driven results.

Will AI steal your job?

One of the biggest concerns about AI in reality capture is its potential to replace human jobs. According to another recent industry survey, 48.9% of companies report turning to automation and advanced technologies to reduce reliance on manual labour. But while there is no doubt that AI is changing the job and skills landscape, it is also creating opportunities. AI can automate complex workflows and optimize processes, allowing professionals to get more work done and to focus on more strategic tasks.

As with any new technology, the most successful geospatial experts have adapted quickly and learnt new skills to add AI to their toolbelt – using it to save time and uncover deeper insights from vast datasets. I’m also seeing AI work well as a co-pilot, guiding users to the best location to use their laser scanner and even alerting them to issues with data quality in real time. The key takeaway is that AI will not take over. It will enhance workflows while keeping a human in the loop.

Unlocking AI's potential

AI in reality capture may be doing less than you expect, but its impact is more transformative than you realize. It’s neither an all-encompassing solution nor a replacement for human expertise. Instead, it is a powerful tool that, when applied thoughtfully, can unlock new levels of efficiency, accuracy and scalability.

The next three to five years will see AI becoming even more deeply integrated into reality capture. AI-driven automation will continue to optimize workflows, allowing geospatial professionals to achieve more in less time. As such, AI is becoming a core theme in research and development. This evolution will lead to a more personalized and efficient reality capture experience from field to finish. The real challenge ahead is not simply adopting AI, but understanding where it truly adds value and where it doesn’t. By familiarizing themselves with AI’s potential, geospatial professionals can identify areas where AI offers the most significant impact for them.