EUROPA2

Our robot navigating through the populated downtown area of Freiburg (EUROPA/EUROPA2).

Obelix localizing itself in the 3D map provided by GeoAutomation.

This video shows our particle filter based localization using 3D laser scans relative to a street network from OpenStreetMap.

These videos shows our novel method for Visual localization of autonomous robots across extreme perceptual changes e.g seasons. We use dense image description using HOG features instead of keypoint based methods. We also leverage the sequential information to handle different vehicle speeds, new routes, and do not rely only on global descriptor based matches.

A comparison between the laser scaner setup of our robot in the EUROPA1 project and the EUROPA2 project. The new sensor setup provides high frequency 3D measurements.

Our robot can detect pedestrians from its camera input in real-time, while only using a single CPU core for this task. Accuracy will be further improved with a tracking system and by exploiting depth data from other sensors.

The traffic sign recognition system uses the same computer vision framework as the pedestrian detector. It detects a number of general traffic sign classes (mandatory, prohibitive, danger, ...) and recognizes the specific sign (e.g. speed limit of 50km/h) with a one-vs-all classifier.