Efficient Computation Offloading and Data Transmission Strategy for 3D Object Detection in Edge Computing Networks

Abstract

3D object detection leverages sensors like LiDAR and cameras to capture scene information, enabling precise determination of objects' spatial positions and orientations. This technology finds extensive applications in autonomous driving, smart homes, industrial automation, and intelligent security systems. However, high-precision 3D object detection algorithms often require substantial computational resources, posing limitations for deployment on resource-constrained devices. In this paper, we devise an efficient computation offloading and data transmission framework specifically tailored for edge computing networks to address this challenge. Our framework takes into account both the computing and communication capabilities of terminal devices and network conditions, offloading suitable computation tasks to the edge for processing. This approach mitigates the algorithm's performance requirements on terminal devices. Furthermore, we propose a data transmission scheme that incorporates attention mechanisms and hardware-accelerated coding. This scheme effectively reduces detection time and enhances overall system performance. Experimental results demonstrate that our proposed framework significantly enhances the efficiency of 3D object detection on resource-constrained devices within edge computing networks, while maintaining high detection accuracy.