Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB

3D tracking of vehicles in convoys

3D tracking of vehicles in convoys

For applications involving unmanned vehicles in convoys, the feasibility of identifying and tracking a leading vehicle was demonstrated in complex urban traffic scenarios, including occlusions, turnoffs, parked vehicles, head-on traffic and traffic intersections.

The experimental vehicle used was configured with a scanning LiDAR sensor and on-board processing electronics. A monitor in the vehicle rack visualized the laser range imagery as well as the tracking results. At the start of a test drive, the vehicle to be followed is selected by positioning a marker in the current range image. From the range image, a 3D-model of the leading vehicle is automatically generated. This model is used to discriminate the leading vehicle from other objects and to re-identify it after temporary occlusions. Real-time procedures were implemented to calculate the sensor motion from the range imagery, without the need to use an inertial navigation system.

 

     

Experimental vehicle with laser scanner and on-board processing for automatic vehicle tracking in urban traffic

 

All test drives using a large variety of leading vehicles were successfully carried out without track loss.

Typical tracking problems such as the presence of confusers, partial occlusions, object rotations, and highly variable cluttered backgrounds, which often induce track loss for 2D-sensors, were effectively resolved. Automatic re-identification after occlusions was successful, provided the surface geometry of the leading vehicle was distinguishable (for the given range accuracy) from other vehicles within the field of view.

 

Further information

  • Armbruster, W., 2008. Bayesian hypothesis generation and verification. Pattern Recognition and Image Analysis, Vol. 18, No. 2, pp. 269-274, ISSN: 1054-6618. [DOI: 10.1134/S1054661808020120] [publica]

  • Armbruster, W., 2005. Comparison of deterministic and probabilistic model matching techniques for laser radar target recognition. Proceedings of SPIE 5807, Automatic Target Recognition XV, 233. [DOI: 10.1117/12.602044] [publica]