Uploaded September 2016 | Updated November 2026, 1 minute ago
We present SegMatch, a technique for enabling autonomous vehicles to recognize previously visited areas based on the extraction and matching of 3D segments. This important capability of autonomous vehicles to localize themselves in their environment is used for constructing accurate 3D maps and for performing different tasks such as driving to goals or exploring areas of interest.
The technique is based on detecting segments (eg. parts of vehicles, trees and buildings) in the vicinity of the vehicle (shown in colours) and on matching these segments with the ones extracted from a target map (shown in white). These segment matches can be directly translated into precise localization information which enables accurate 3D map construction.
In this specific example, the target map is created from real-world data recorded by a road vehicle equipped with a 3D Velodyne laser range-finder, in Karlsruhe, Germany (KITTI dataset). This target map can be either loaded previously or built online as respectively shown in the first and second demonstrations.
The detailed SegMatch description is available here: arxiv.org/abs/1609.07720 and its open source implementation is available here: github.com/ethz-asl/segmap
For more information on the TRADR project: http://www.tradr-project.eu/
Related videos: youtube.com/watch?v=JJhEkIA1xSE and youtu.be/CMk4w4eRobg
Contact: renaudube@gmail.com
We present SegMatch, a technique for enabling autonomous vehicles to recognize previously visited areas based on the extraction and matching of 3D segments. This important capability of autonomous vehicles to localize themselves in their environment is used for constructing accurate 3D maps and for performing different tasks such as driving to goals or exploring areas of interest.
The technique is based on detecting segments (eg. parts of vehicles, trees and buildings) in the vicinity of the vehicle (shown in colours) and on matching these segments with the ones extracted from a target map (shown in white). These segment matches can be directly translated into precise localization information which enables accurate 3D map construction.
In this specific example, the target map is created from real-world data recorded by a road vehicle equipped with a 3D Velodyne laser range-finder, in Karlsruhe, Germany (KITTI dataset). This target map can be either loaded previously or built online as respectively shown in the first and second demonstrations.
The detailed SegMatch description is available here: arxiv.org/abs/1609.07720 and its open source implementation is available here: github.com/ethz-asl/segmap
For more information on the TRADR project: http://www.tradr-project.eu/
Related videos: youtube.com/watch?v=JJhEkIA1xSE and youtu.be/CMk4w4eRobg
Contact: renaudube@gmail.com










