An Embedded System with 3D Point Cloud Based Object Detection

Engineering notes

Key topics: autonomous driving, bird's eye view, bev, 3d object detection, object detection, lidar, point cloud, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

This paper presents an embedded system for efficient 3D object detection in robotics and autonomous systems using point cloud data acquired from 3D LiDAR sensors. Due to the large data volume and high computational complexity of point cloud processing, real-time object detection on embedded platforms remains challenging. To address this, the proposed system integrates a voxel feature encoder and a 3D convolution accelerator on an FPGA, combined with an ARM-based embedded module for post-processing and system control. The point cloud data are voxelized and converted into voxel-wise features, which are processed by the 3D convolution accelerator and further refined through a region proposal network (RPN) and post-processing to generate 3D bounding boxes in both bird's eye view (BEV) and image space. The proposed system successfully detects objects in categories such as cars, cyclists, and pedestrians, achieving inference times comparable to those of conventional high-performance CPU/GPU platforms.

Links and sources

• Official / arXiv page

Comments