Semantic scene understanding is crucial for autonomous driving, robot navigation, and intelligent scene perception tasks. Semantic segmentation can identify and locate targets at the pixel level, providing essential situation awareness for navigation tasks. This chapter focuses on the most commonly used semantic segmentation task in autonomous driving, namely, lane line detection. From the perspective of multimodal fusion, the task of semantic segmentation is revisited. Due to the gap in different modalities, the alignment among of multi-sensor data is challenging for fusion, especially for scene segmentation. Several novel lane line detection algorithms are designed based on attention mechanisms, mutual information, and an end-to-end approach to solve this problem. This chapter first discusses the different fusion architectures and then utilizes the attention mechanism to achieve better results. Further, an adaptive fusion strategy is proposed to address robustness and bias in fusion. After experimental verification, the proposed methods have achieved good performance in segmentation tasks.
Multi-Sensor Scene Segmentation
Multi-sensor Fusion for Autonomous Driving ; Kapitel : 5 ; 87-118
11.05.2023
32 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Focus-aided scene segmentation
British Library Online Contents | 2015
|Perceptual Evaluation of Driving Scene Segmentation
IEEE | 2021
|