可变形特征融合的无人驾驶系统三维车辆检测  

Deformable feature fusion 3D vehicle detection of unmanned vehicle system

作　　者：伍锡如[1] 林钰睿 WU Xiru;LIN Yurui(School of Electronic Engineering and Automation,Guilin University of Electronic Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林电子科技大学电子工程与自动化学院,广西桂林541004

出　　处：《激光杂志》2024年第8期69-75,共7页Laser Journal

基　　金：国家自然科学基金地区科学基金项目(No.62263005);广西自然科学基金重点项目(No.2020GXNSFDA238029);广西高校人工智能与信息处理重点实验室开放基金重点项目(No.2022GXZDSY004);广西研究生教育创新计划项目(No.YCSW2023298)。

摘　　要：针对无人驾驶系统环境感知中的车辆检测精度低的问题,提出一种基于可变形特征融合的三维车辆检测算法。首先,通过路面实况增强算法,提高收敛速度和性能;去除地面点云,减少无关点云的干扰。接着,构造可变形特征融合模块,自适应对齐不同模态数据之间的姿态和位置信息,提升多模态数据的利用效率;优化损失函数,添加对抗损失判断车辆运动的真实性,提高网络对小目标的检测精度。最后,通过训练得到网络模型的最佳权重,使用KITTI数据集进行测试,能达到较好的车辆识别效果。实验结果表明:其平均精度值为83.26%,平均检测时间为0.15 s。该算法能够快速、准确地在无人驾驶系统中对车辆进行识别。A vehicle detection algorithm based on deformable feature fusion is proposed to solve the problem of low vehicle detection accuracy in environment perception of unmanned driving system.We propose a 3D vehicle detection algorithm based on deformable feature fusion.Firstly,the convergence speed and performance were improved by the road scene enhancement algorithm.The ground point cloud is removed to reduce the interference of irrelevant point cloud.Then,a deformable feature fusion module was constructed to adaptively align the pose and position information between different modal data to improve the utilization efficiency of multi-modal data.The loss function was optimized,and the adversarial loss was added to judge the authenticity of vehicle motion,so as to improve the detection accuracy of the network for small targets.Finally,the best weight of the network model is obtained by training,and the KITTI data set is used for testing,which can achieve better vehicle recognition effect.The experimental results show that the average precision value is 83.26%,and the average detection time is 0.15 s.The algorithm can quickly and accurately identify the vehicle in the unmanned driving system.

关 键 词：环境感知 车辆检测 三维点云 可变形特征融合 无人驾驶 

分 类 号：TN211[电子电信—物理电子学]