SPS-RCNN: Semantic-Guided Proposal Sampling for 3D Object Detection from LiDAR Point Clouds

Sensors, Год журнала: 2025, Номер 25(4), С. 1064 - 1064

Опубликована: Фев. 11, 2025

Three-dimensional object detection using LiDAR has attracted significant attention due to its resilience lighting conditions and ability capture detailed geometric information. However, existing methods still face challenges, such as a high proportion of background points in the sampled point set limited accuracy detecting distant objects. To address these issues, we propose semantic-guided proposal sampling-RCNN (SPS-RCNN), multi-stage framework based on point–voxel fusion. The comprises three components: voxel-based region network (RPN), keypoint sampling stream (KSS), progressive refinement (PRN). In KSS, novel (SPS) method, which increases foreground enhances sensitivity outliers through multilevel that integrates proposal-based local global sampling. PRN, cascade module (CAM) is employed aggregate features from multiple subnets, progressively refining proposals improve for medium- long-range Comprehensive experiments widely used KITTI dataset demonstrate SPS-RCNN improves exhibits enhanced robustness across categories compared baseline.

Язык: Английский

---

Instance-Aware Sampling and Voxel-Transformer Encoding for Single-Stage 3D Object Detection

Digital Signal Processing, Год журнала: 2025, Номер unknown, С. 105171 - 105171

Опубликована: Март 1, 2025

Язык: Английский

---

Towards a Data‐Driven Digital Twin AI‐Based Architecture for Self‐Driving Vehicles

IET Intelligent Transport Systems, Год журнала: 2025, Номер 19(1)

Опубликована: Янв. 1, 2025

ABSTRACT Recent advancements on digital technologies, particularly artificial intelligence, have been resulted into remarkable transformations in automobile industry. One of these technologies is intelligence (AI). AI plays a key role the development autonomous vehicles. In this paper, vehicle (AV) platform layers studied. The focus paper indexed papers Scopus database. most relevant keywords are selected and searched. 628 articles, between 2014 2024 were for analysing reviewing. Articles analysed based source type, topics, algorithms. Text mining content analysis articles revealed that 233 journals published top 185 to assess. topics classified perception, localization mapping, planning, decision making, control, communication, security, data management, general topics. Each areas consisted many roles, or tasks use realize their tasks. Convolutional neural network mapping more used. Deep reinforcement learning had application planning decision‐making areas. main result recognition AVs classification, designing data‐driven twin AI‐based model vehicles architecture, containing physical world, virtual communication space, applied algorithms each layer, which aid researchers choose suitable methods field This study provided comprehensive map research projects related from 1985 2022. Finally, some directions suggested.

Язык: Английский

---

ME-ODAL: Mixture-of-Experts Ensemble of CNN Models for 3D Object Detection from Automotive LiDAR Point Clouds

Communications in computer and information science, Год журнала: 2024, Номер unknown, С. 279 - 300

Опубликована: Янв. 1, 2024

Язык: Английский

---