SC-MII: Infrastructure LiDAR-based 3D Object Detection on Edge Devices for Split Computing with Multiple Intermediate Outputs Integration
Taisuke Noguchi, Takayuki Nishio, Takuya Azumi
TL;DR
SC-MII addresses the latency and privacy challenges of infrastructure-based 3D object detection by offloading heavy tail-model inference to an edge server while edge devices process early layers. It uses Split Computing to transmit compact intermediate features from multiple infrastructure LiDARs, which are spatially aligned with NDT-derived transformations and integrated via max-pooling or convolution-based fusion. Compared with single-LiDAR and full on-device inference, SC-MII delivers up to a $2.19\times$ speed-up and up to $71.6\%$ reduction in edge processing time, with only a $1.09\%$ drop in AP@0.5, demonstrating effective privacy-preserving multi-LiDAR fusion. The approach offers practical benefits for real-time distributed perception in autonomous systems, with potential extensions in data compression and resilience to network variability.
Abstract
3D object detection using LiDAR-based point cloud data and deep neural networks is essential in autonomous driving technology. However, deploying state-of-the-art models on edge devices present challenges due to high computational demands and energy consumption. Additionally, single LiDAR setups suffer from blind spots. This paper proposes SC-MII, multiple infrastructure LiDAR-based 3D object detection on edge devices for Split Computing with Multiple Intermediate outputs Integration. In SC-MII, edge devices process local point clouds through the initial DNN layers and send intermediate outputs to an edge server. The server integrates these features and completes inference, reducing both latency and device load while improving privacy. Experimental results on a real-world dataset show a 2.19x speed-up and a 71.6% reduction in edge device processing time, with at most a 1.09% drop in accuracy.