目标检测在机器人、自动驾驶等实际应用领域中具有广泛的应用。在这些场景下,目标检测任务需要在资源有限的平台上实时执行,对目标检测算法的参数量和检测速度有着较高的要求,因此需要实现目标检测算法的轻量化和高效化。然而传统的卷积神经网络(convolutional neural networks,CNNs) 由于网络结构复杂、对算力要求较高,难以满足移动端的应用需求。为解决以上问题,本文提出了一种基于点云数据的一阶段轻量化目标检测算法CYM-Net模型。该模型融合了MobileNetV3的bneck模块设计思想和YOLOv4目标检测思想,并对特征金字塔进行了改进,从而显著减少了模型的参数量。本文在KITTI数据集上对CYM-Net模型进行了训练和验证。实验结果表明,CYM-Net模型在鸟瞰图和3D检测两个任务上均展现出更优异的性能,并且其检测速度也优于其他方法。通过本研究,本文为机器人、自动驾驶等领域的目标检测问题提供了一种高效轻量化的解决方案。

Object detection plays a crucial role in practical applications such as robotics and autonomous driving.In these scenarios,real-time execution of object detection tasks on resource-constrained platforms is essential,demanding highly on parameters and detection speed of object detection algorithm,realizing lightweight and efficient object detection algorithms.However,traditional convolutional neural networks (CNNs) with complex network structures and high computational requirements are not suitable for deployment on mobile devices.To address these challenges,this paper proposes a one-stage lightweight object detection algorithm,named CYM-Net,based on point cloud data.The CYM-Net model cleverly integrates the design principles of MobileNetV3′s bneck module and YOLOv4′s object detection concept while improving the feature pyramid,resulting in a significant reduction in model parameters.The CYM-Net model is trained and validated on the KITTI dataset.Experimental results demonstrate that the CYM-Net model outperforms other methods in both bird′s-eye view and 3D detection tasks,and it also exhibits superior detection speed.This research provides an efficient and lightweight solution for object detection in fields like robotics and autonomous driving.

TP391