Modern real-time dense stereo vision provides precise depth information for nearly every pixel of an image, indicating stereo cameras as a key sensor for future vehicle safety systems. Efficient analysis of this large amount of data by different tasks running in parallel asks for a medium level representation that decouples application specific analysis from low-level vision. Recently, the so called “Stixel World” has been proposed. It models the objects in the scene, implicitly separates them from the ground plane, encodes the freespace to maneuver and thus represents the scene in a highly compact manner that supports different recognition tasks efficiently. The potential of this new representation depends on the accuracy that can be achieved. Therefore, this paper analyzes the precision of this representation using a high performance laser scanner as reference sensor. The statistical analysis confirms the high accuracy as expected from visual inspection.
Ground truth evaluation of the Stixel representation using laser scanners
2010-09-01
1611713 byte
Conference paper
Electronic Resource
English
Learning Stixel-based Instance Segmentation
IEEE | 2021
|SPIDER-BASED STIXEL OBJECT SEGMENTATION
British Library Conference Proceedings | 2014
|Efficient Stixel-based object recognition
IEEE | 2012
|LEARNING STIXEL-BASED INSTANCE SEGMENTATION
British Library Conference Proceedings | 2021
|STIXEL ESTIMATION AND ROAD SCENE SEGMENTATION USING DEEP LEARNING
European Patent Office | 2016
|