Phoenix Mobile Experimental Platform

Research Group Robotics & Process Control

Neural Scan Classification:

The ScanTrainer software package applies a neural classification technique originally introduced by G. von Wichert for image processing. This technique is successfully used for image classification/scene recognition in the CAROL project and modified in the ScanTrainer component to process laserscans instead of camera images.

One the method's key features is its ability to autonomously adapt to the environmental specifics in an unsupervised training phase. This significantly reduces the manual effort to model relevant environment aspects.

The results of both, visual scene recognition and scan classification are especially valuable for place recognition in topological or hybrid topological/metric self-localization architectures.

Implementation: Alexander Spieß

Publications: