The AI model automatically identifies and separates rails, overhead line infrastructure, poles, signals, signs, vegetation, ground and surrounding infrastructure.

Both the class catalog and processing workflow are customizable. Models can be trained and optimized for specific railway networks, acquisition systems and project requirements. Dedicated processing steps can further improve geometrically challenging objects and correct common classification errors.

For the Digitale Schiene Deutschland project, the optimized workflow achieved classification accuracies of up to 95 to 98 percent for well represented target classes.

 From Classified Points to Railway Vector Data

Classification provides the basis for the next processing stage. Pointly can automatically transform detected railway objects into structured points, lines and 3D geometries.

For railway applications, this can include rail tops, rail centerlines, platform edges, overhead line masts, catenary infrastructure, signals, signs and other trackside assets.

The extraction workflow can also identify object positions, heights, key points and geometric relationships depending on the required output.

Vector Objects and Attribute Extraction

Pointly not only extracts vector objects and 3D geometries from the classified point cloud but also derives object specific attributes such as position, height, length, orientation and geometric relationships. The extracted information can be customized to match specific BIM, GIS and railway asset requirements.