LP360 AI Ground Classifier FAQ

What is the LP360 AI Ground Classifier?

The LP360 AI Ground Classifier is a cloud-based AI processing task run through LP360 Cloud and seamlessly integrated within LP360 Desktop. It uses a pre-trained deep learning model to classify individual points in a point cloud as ground or non-ground.

Do I need to train my own AI model to use the classifier?

No. LP360 uses a pre-trained deep learning ground classification model. You do not need to provide training data or know how to build your own machine learning/deep learning model to use the AI classifier in LP360.

Will my data be used to improve the AI model?

No. Your data remains proprietary to you and is not used to improve the AI ground classification model.

How do I use the LP360 AI Ground Classifier?

To access the AI classifier, use the Ground Classification ribbon and the AI Ground Classification button to configure and submit a classification task to Job Manager. Job Manager will submit the job to LP360 Cloud for processing, automatically configuring your data, uploading and downloading from the cloud and adding the resulting classified point cloud as a layer in your LP360 project. Full instructions on the workflow are available here.

How effective is the LP360 AI Ground Classifier?

The AI classifier is very effective at segmenting ground returns from non-ground. Extensive testing and results reported by our beta users show the AI significantly reduces both false positives (non-ground classified as ground) and false negatives (ground not classified as ground) compared to traditional algorithmic techniques such as the adaptive TIN method. However, as with all analytical techniques, the AI results are not 100% accurate. In most cases you should still expect to do some clean-up and editing of the point cloud after AI classification. Most users report this to be significantly less effort than would be needed using the traditional approach.

What are the advantages of the LP360 AI Classifier compared to the traditional ground classification tools in LP360?

Advantages of using the AI classifier include:

• Better overall results with less effort needed for editing your ground surface after classification.
• No pre-processing or set-up required. The AI does not require a seed surface/seed points to be created, significantly reducing your set-up time. You can run the AI ground classifier immediately after geocoding and strip alignment. You can run it on any unclassified point cloud, regardless of the source.
• Better results in the most complex terrain and under canopy. The AI classifier performs much better than the adaptive TIN in complex terrain, under canopy and in areas with significant vertical structure.
• Better discrimination of low objects close to the ground surface. The AI is better at discriminating objects such as cars or vehicles, curbs or low walls, or shrubs and bushes close to the ground. Areas where the adaptive TIN approach is less effective.

What tile size should I set when using the LP360 AI Ground Classifier?

You will be asked to specify a tile size for your AI classification job. There are several factors that impact the optimum tile size for the AI classifier:

• The individual tiles should be between 100 m – 500 m in size.
• The individual LAS file size should be 3 GB or less.
• The minimum number of points in a file/tile should be 1,000.

LP360 will automatically tile your data based on the above restrictions and the tile size you specify. Tiles greater than 3 GB in size will be flagged and prevent the job from proceeding. Tiles with less than 1,000 points will be removed from the job and the job will continue to process. For most TrueView drone datasets flown with less 50% overlap, 300 m is a good default value to use for tile size.

What are the Outlier, Smoothing, and Delete Nosie options in the LP360 AI Ground Classifier?

The AI classifier includes options for outlier detection, smoothing, and noise deletion. These are optional steps that are not required by the AI classifier but are recommended for most datasets. They help to clean the point cloud prior to classification. Unless you are experiencing issues with the AI classifier, we recommend having them ON for most jobs.

Will the LP360 AI Ground Classifier preserve my existing classifications?

No. The AI classifier will undo any existing ground classification or other classification already applied to your point cloud. This restriction will be removed in a future release of

LP360, but for now the AI is intended to be run against unclassified data and will remove any existing classifications.

What are some of the limitations of the LP360 AI Ground Classifier?

The AI classifier has some limitations and areas where it does not perform as well. We are continually looking to improve the AI model’s performance, but areas to look out for with the current model include:

• Water surfaces. To the AI, water surfaces can mimic a ground surface depending on the surface state, turbidity, sediment level, and other factors in the environment. Always check for false positives (water classified as ground) on water bodies if your system is generating a lot of returns from water surfaces.
• Overhanging vegetation. The AI may create more most false positives around streambeds, drainages, and wetlands in the presence of overhanging vegetation or low ground cover. Check for false positives (vegetation classified as ground) in areas with any hydrologic features intermixed with vegetation.
• Steep natural slopes between 85 – 89 degrees. The AI may miss very steep natural surfaces that are less than 90 degrees but greater than 85 degrees. Check for false negatives (unclassified ground) on steep embankments and similar natural structures.
• Changes in point density across the site. The AI is sensitive to major changes in point density. Having a uniform point density across your site is ideal but often hard to achieve. Small variations in point density will not impact the AI, but major differences, for example dropping from 100 points per sq. m to 10 points per sq. m, may result in false negatives (unclassified ground) in that area.

In all cases where the AI results need modification, you can use LP360’s built-in clean-up and editing tools to quickly correct the ground surface.

What other AI classification tools are available in LP360?

We are excited to introduce the AI Ground Classifier in LP360 V2025.1. We prioritized AI ground classification because it is a key area where users have requested improvements. For many users, the ground surface is the primary deliverable that needs to be generated. This initial AI tool will also help us develop our cloud infrastructure to support future AI classification tools in LP360. Our LP360 roadmap includes AI classifiers for highway infrastructure, rail corridors, and power line modeling.