The field of autonomous driving perception is rapidly advancing, with a focus on improving the accuracy and robustness of 3D object detection, semantic segmentation, and other related tasks. Recent developments have highlighted the importance of multimodal fusion, particularly the combination of camera, LiDAR, and radar sensors, to enhance perception capabilities in various environmental conditions. Notably, researchers have explored innovative architectures and techniques, such as fine-tuning vision foundation models, multi-frame feature fusion, and geometry-aware point drop, to address challenges like adverse weather, occlusions, and sparse data. These advancements have led to significant improvements in detection accuracy, efficiency, and scalability, paving the way for more reliable and widespread adoption of autonomous driving technologies. Noteworthy papers include AD-SAM, which fine-tunes the Segment Anything Model for autonomous driving perception, achieving state-of-the-art segmentation accuracy on Cityscapes and BDD100K benchmarks. M^3Detection proposes a unified multi-frame 3D object detection framework, demonstrating superior performance on VoD and TJ4DRadSet datasets. These innovative approaches underscore the field's progress toward developing more effective and efficient autonomous driving perception systems.
Advancements in Autonomous Driving Perception
Sources
AD-SAM: Fine-Tuning the Segment Anything Vision Foundation Model for Autonomous Driving Perception
MLPerf Automotive
M^3Detection: Multi-Frame Multi-Level Feature Fusion for Multi-Modal 3D Object Detection with Camera and 4D Imaging Radar
Which LiDAR scanning pattern is better for roadside perception: Repetitive or Non-repetitive?
GEDICorrect: A Scalable Python Tool for Orbit-, Beam-, and Footprint-Level GEDI Geolocation Correction
Benchmarking individual tree segmentation using multispectral airborne laser scanning data: the FGI-EMIT dataset
Towards classification-based representation learning for place recognition on LiDAR scans
LiDAR-VGGT: Cross-Modal Coarse-to-Fine Fusion for Globally Consistent and Metric-Scale Dense Mapping
Saliency-Guided Domain Adaptation for Left-Hand Driving in Autonomous Steering
Source-Only Cross-Weather LiDAR via Geometry-Aware Point Drop
CaRLi-V: Camera-RADAR-LiDAR Point-Wise 3D Velocity Estimation
Terrain-Enhanced Resolution-aware Refinement Attention for Off-Road Segmentation
UniLION: Towards Unified Autonomous Driving Model with Linear Group RNNs
Self-Supervised Moving Object Segmentation of Sparse and Noisy Radar Point Clouds
The Urban Vision Hackathon Dataset and Models: Towards Image Annotations and Accurate Vision Models for Indian Traffic
Evaluating the Impact of Weather-Induced Sensor Occlusion on BEVFusion for 3D Object Detection