Supper & Supper GmbH has developed a Software as a Service called Pointly, which allows to process 3D point clouds with the help of artificial intelligence (AI).
The kick-off is scheduled with an AI-supported labelling tool for mid-2020.
3D point clouds are a key technology in the digital transformation of many industries, from construction and forestry to urban planning and robotics. Where CAD models guide the real-world implementation of digital designs, point clouds enable the (re)capture of the physical environment for digital analysis. Current developments in the field of AI make it possible to automatically recognize objects such as houses or grounds in point clouds. This opens up new possibilities for the automatic and area-wide evaluation of 3D data.
The key to a successful 3D AI strategy is data that can be used to train the AI. For 3D point clouds these are classified point clouds. Classified means that a specific label or object class has been assigned to each point within the cloud. However, it is the training data that is the crux of the matter for companies, because large amounts of training data are needed to train a robust AI and classifying is very time-consuming. This is where Pointly comes in with intelligent editing tools that allow training data to be generated faster and easier than before.
Why companies should rely on AI when dealing with 3D point clouds
The digital representation of the physical environment holds diverse potential. More and more companies are focusing on mapping their assets as digital twins. The goal is to achieve a more accurate and faster inventory and condition assessment, analysis and evaluation of the recorded data – preferably automatically. This is where AI can provide active support to accelerate and automate processes.
Point clouds are currently mainly evaluated manually and used for virtual site inspections. Individual objects are easy for humans to recognize. However, a point cloud initially has no information about the objects it contains. Only with classified point clouds it is possible to carry out further analyses, to filter according to certain points and to recognize new connections.
Some point clouds are already classified automatically, but there are major limitations with regards to the recognizable classes and the required data quality. AI in the form of neural networks can be used to automatically classify point clouds for any type of object.
The advantages of such automatically classified point clouds are enormous. For example, the progress on a construction site can be monitored automatically or the urban infrastructure can be inventoried quickly. Limited 3D analyses can be turned into 3D Big Data analyses in the future.
Training data – the new gold
The generation of training data is an often neglected but essential component in AI strategies, as it is needed for the development of powerful neural networks. Already classified point clouds are initially shown to the neural network. On this basis, it learns to dynamically assign the relationships between patterns in the raw data and for example the class “tree”.
With enough training data, the performance of the network becomes so advanced that it can classify new point clouds fully automatically in the future. The larger the data set and the better the quality, the better the performance of the AI that can be trained on it. The conclusion is that companies that start collecting training data early will be pioneers in the future. On the one hand, because they can make more data available for the AI. On the other hand, because they have been training their AI for a longer period of time.
Supper & Supper has developed a range of intelligent tools to lower the entry barrier, the creation of training data. They will simplify working with point clouds. This includes tools such as a magic wand or automatic selection. The magic wand reduces time-consuming marking of specific objects to a single click. With automatic selection, similar objects to the one already selected are displayed. In combination with the simple management of object classes and point cloud projects, a large amount of future training data can be created in a very short time.
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.