.png){kind=icon}
.png){kind=icon}
Return
ShellNet++: A Dynamic Shell Convolution-Based Framework for LIDAR Point Cloud Pedestrian Detection
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2026.3661235.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Designing accurate pedestrian detection systems is a crucial task for many intelligent applications. This paper proposes a novel approach to detect pedestrians using 3D point clouds acquired by a Light Detection and Ranging (LIDAR) sensor. The proposed method is achieved in two major stages: 1) extraction of points of interest (POIs), where LIDAR data is preprocessed and clustered to generate possible pedestrian point subsets; and 2) POIs classification, where tests are conducted to classify the POIs extracted from the first stage into pedestrian and non-pedestrian categories. Here, an advanced ShellNet network version is introduced. The so-called ShellNet++ overcomes the limitations and drawbacks of the traditional version through its new Dynamic Contextual ShellConv (DCS) operator that integrates Dynamic Shell Parameter Network (DSPN) and Contextual Feature Pyramid Network (CFPN) networks to adjust shell parameters dynamically and improve contextual understanding, respectively. The proposed approach has been tested in non-controlled environments using a collection of common LIDAR datasets well known in the road scenes research community (KITTI, LIPD, Nuscenes, and Waymo Open Dataset). The obtained results highlight the model’s capabilities and justify its favorable standing compared to recent state-of-the-art pedestrian detection methods.
Keywords:
Pedestrian detection
intelligent transportation systems
light detection and ranging (LIDAR)
point clouds
3D data
ShellNet
deep learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3.6
Papers:
9.8W
Citations:
29.4W