Multimodal 3D object detection is crucial for autonomous systems but suffers from high delay due to significant computational demands. To address this, we propose an edge computing-assisted framework that balances load between terminal devices and edge servers. We introduce dynamic threshold tuning and resolution-adaptive offloading algorithms to optimize performance. Experimental results demonstrate that our approach significantly reduces delay by minimizing offloading frequency while maintaining high accuracy, achieving a superior delay-accuracy trade-off. Furthermore, the framework exhibits robust adaptability across various models and bandwidth conditions, ensuring effectiveness in dynamic environments.
Zhao et al. (Thu,) studied this question.