Authors:
Andrei Vatavu
and
Sergiu Nedevschi
Affiliation:
Technical University of Cluj-Napoca, Romania
Keyword(s):
Environment Representation, Polyline, Object Contour, Iterative Closest Point, Driving Assistance, Stereo-vision, Object Delimiters, Motion Detection.
Related
Ontology
Subjects/Areas/Topics:
Image Processing
;
Informatics in Control, Automation and Robotics
;
Intelligent Transportation Technologies and Systems
;
Perception and Awareness
;
Robotics and Automation
;
Vehicle Control Applications
;
Vision, Recognition and Reconstruction
Abstract:
The detection of moving traffic participants is an essential intermediate step for higher level driving technology tasks. Regardless of the type of used sensors, dynamic environment modeling becomes even more difficult when the surrounding world is unstructured and heterogeneous. In such complex environments the representation system can be affected by noisy measurements, occlusions, wrong data association or unpredictable nature of the traffic participants. We propose a solution of representing the dynamic environment in real-time by using the pairwise alignment of free-form models and considering the advantages provided by a dense stereovision system. Instead of registering the whole 3D point cloud, our method is based on extracting and registering a more compact model of the environment taking into consideration the most visible object cells from the ego car. The proposed method is based on information provided by a Digital Elevation-Map, but can be easily adapted for other types
of intermediate representations.
(More)