Supper & Supper is a full-service data science service provider in Germany that develops tailor-made solutions for machine learning, artificial intelligence, process optimization and digital transformation. Pointly is their intelligent, cloud-based software solution for managing, classifying and analyzing big data in 3D point clouds, using innovative AI techniques. Pointly is planned for release in Q2 of 2020.
Currently, Pointly offers automated measurement of objects on point clouds (e.g. for logistics), creation of tree cadastre, 3D modeling of buildings, recording of road boundaries and damages, measurement of roof areas and checking for possible applications (e.g. solar cell). All these tasks are currently done manually and can be approached via Pointly as an AI strategy.
The first software release is scheduled for Q2 of 2020 and will provide intelligent selection tools to classify point clouds faster and easier among other possibilities like to upload, manage and view huge point clouds. This release manages .las and .laz file formats, while in the future most of the common formats for managing point clouds will be supported.
The importance of collecting training data for AI
For Pointly, the companies who start collecting training data early will be pioneers in the future. The more training data a company has, the better the AI can detect what matters to them. One of the exciting advantages of AI-based solutions compared to rule-based automation is that the neural network learns to abstract from the training data. What seems like bad training data to the human user can increase the robustness of the algorithm and make it applicable to an even broader range of use cases.
The companies that start collecting training data will be pioneers because they can make more data available for the AI, and also because they have been training their AI for a longer time. Although companies with more data and training experience will have a head start over other companies, Pointly offers that same data training experience more easily, as a result of its labeling process. Also, Pointly can help companies who don’t have data with its Acquisition Service, acquiring data for customers through its Partner network, by generating point clouds via drone images for example.
Pointly workflow
Assuming that a customer has already a point cloud in the .las format and wants to classify all houses in this point cloud, this can be uploaded to the cloud after which all houses can be classified. Pointly accelerates his classifying process, as the user only has to click into one point of the house and every similar point is selected automatically. In the end, all similar points will reveal the shape of the house. In the past, every single point had to be selected before the shape of the same house could be created.
Besides accelerating manual classification, Pointly creates high-quality training data that can be used for automatic classifications in the future. Supper & Supper offers 3D deep learning know-how for tailoring AI solutions to specific needs and helps clients get scalable data products out of their point cloud data.
Pointly tools that simplify working with point cloud data
Pointly’s intelligent selection tools offer a simpler and more effective way to train AI using training classifiers to detect custom objects from point clouds. Launching in Q2 2020, Pointly’s most important tool will be the segment selector. This tool allows for selecting coherent segments with only one click, so there’s no need to zoom around, change views and draw a perfect border. Pointly has already figured out the most likely breaks between distinct objects in the data and allows for making selections based on these. That way, complex objects with difficult outlines can be quickly selected and assigned with an object class.
To support different levels of detail, the segment tool allows choosing between different degrees of segment granularity. These range from “fine” over “medium” to “coarse” and are designed to cover different use-cases across all possible point cloud types and resolutions. That way, point cloud data users can select large swathes of ground with one click and street signs with the next. A polygon lasso tool provides an alternative to adapting the segments manually, although the accuracy of Pointly’s segment finder already satisfies most of all objects and does the heavy lifting.
Both selection tools directly integrate with the assignment of object classes so there is no need to assign those. The segment tool is only the beginning of many more intelligent selection tools that will speed up working with point clouds: in the pipeline is a dynamic Magic Wand selector, as well as a tool for selecting similar objects to already selected ones. Similar to the well-known tool from paint programs, with the Magic Wand tool users will be able to select a point and dynamically set a threshold for the desired similarity. With this tool, even extremely detailed selection in massive point clouds are possible in just a few clicks
Extending current functionality
In the future, Pointly will be usable as a user-friendly end-to-end platform solution to not only manage and label but also analyze big data from 3D point clouds. Supper & Supper plans to extend current functionality in the future by implementing standard detectors, offering storage integration capabilities to standard storage solutions, having data acquisition capabilities with partners, and exporting not only point clouds all kinds of data products like 3D CAD models, map layers and more. AI-based detectors will be added to Pointly for a wide range of common use cases as well as analytics services generating high-level data products from these automatic detection results, such as reports, statistics, 3d models and map layers.
