Fig 1.
AP comparisons of single-modal and multimodal models at different point cloud resolutions.
Fig 2.
Delay comparisons of single-modal and multimodal models at different point cloud resolutions.
Table 1.
The value of AP corresponding to different image resolutions.
Table 2.
Difference between our scheme and the most relevant schemes.
Fig 3.
Edge-assist framework of 3D object detection.
Fig 4.
Data size of different point cloud resolutions.
Fig 5.
AP values for single-modal and multimodal models.
Fig 6.
Delay values for single-modal and multimodal models.
Fig 7.
AP values of single-modal.
Fig 8.
Delay values of single-modal.
Table 3.
Main symbols and definitions.
Fig 9.
Performance comparison under different confidence thresholds.
Fig 10.
Performance under fixed point resolution and confidence thresholds.
Fig 11.
The value of terminal single-modal model (PointPillar) of the mapping function between the confidence and AP.
Fig 12.
The value of the edge multimodal model (DeepInteraction) of the mapping function between the confidence and AP.
Table 4.
Algorithm 1: Dynamic threshold updating algorithm.
Table 5.
Algorithm 3: Adaptive offloading algorithm.
Fig 13.
The testbed of system.
Table 6.
Comparison of equipment parameters.
Fig 14.
Transmission rate variation under different time epochs.
Fig 15.
Performance under different values of.
Fig 16.
Delay values of different algorithms.
Fig 17.
AP values of different algorithms.
Fig 18.
Delay values of different algorithms(BEVFusion used for edge inference).
Fig 19.
AP values of different algorithms(BEVFusion used for edge inference).
Fig 20.
Delay values of different algorithms(TransFusion used for edge inference).
Fig 21.
AP values of different algorithms(TransFusion used for edge inference).
Fig 22.
Delay values of different algorithms(BEVFusion_L used for device inference).
Fig 23.
AP values of different algorithms(BEVFusion_L used for device inference).
Fig 24.
Delay values of different algorithms(TransFusion_L used for device inference).
Fig 25.
AP values of different algorithms(TransFusion_L used for device inference).
Fig 26.
Comparison of AP.
Fig 27.
Delay values of different datasets(KITTI).
Fig 28.
AP values of different datasets(KITTI).
Fig 29.
Performance of different point cloud compression algorithms.
Fig 30.
Delay values of different bandwidths.
Fig 31.
AP values of different bandwidths.