Lightweight 3D Object Detection Based on Bridging Structure

The study of 3D object detection based on deep learning has increasingly grown in importance in the realm of autonomous driving technology. In this study, we developed a lightweight 3D object detection model based on bridge structure while maintaining the model's accuracy. The model has a much smaller number of parameters. The SGE attention mechanism and the cross-attention mechanism module are first integrated into a backbone network that enhances the lightweight CNN's global semantic information. Second, the loss of target details is prevented thanks to a new bridging structure. Finally, the model's parameter count is decreased using the weight fusion recovery training-based model pruning technique. According to the experimental findings, the model's calculation amount is decreased by 50% while its detection accuracy is raised by 1% to 2%.