Person detection in LIDAR range and intensity data

The ability to detect and recognize persons is an essential part of human-robot interaction. In our department we research methods and algorithms for person detection using 3D LIDAR (Light Detection And Ranging) sensors. These sensors offer, compared to camera-based solutions, precise distance measurements while being insensitive to changes in lighting conditions. This is especially beneficial for robot applications in outdoor environments.
We focus on two designs of 3D-LIDAR: a low-resolution true 3D laserscanner based on an oscillating MEMS mirror, and a 2D-laser ranger mounted on a Tilt-Roll-Unit.


Fig. 1: Person detection in a low-resolution
3D-LIDAR scan using segment blocks

Fig. 2: Person detection using a decision tree
in a scan generated by a tilting 2D-laser ranger

Autonomous person following in outdoor environments

We present a system for autonomous following of a walking person in outdoor environments while avoiding static and dynamic obstacles. The principal sensor is a 3D LIDAR with a resolution of 59x29 points. The system is implemented and demonstrated on one of our outdoor robots, which have been developed and built at our department.
At the final stage of our experiments we conducted two test drives within different environments and as final experiment a long drive with a length of 2083.3m three times around the faculty building. In all three test drives together the robot followed a person for a total distance of 3.5km. During the long track the robot followed the person continuously for about 30minutes and lost its target only three times. The video below shows the first round of this test drive.



Contact

Karsten Bohlmann, Tel.: (07071) 29-77176, karsten.bohlmann at uni-tuebingen.de