视觉图像与三维点云融合的障碍物主动识别与距离感知研究

Research on Active Recognition and Distance Perception of Obstacles Based on Fusion of Visual Image and 3D Point Cloud

针对无人机配电线自动巡检及绝缘层涂覆维护过程中障碍物主动识别和距离感知的问题,提出视觉图像与三维点云相结合的障碍物识别方法。对图像进行增广预处理来丰富数据集,引入基于特征提取的深度学习进行模型训练,获取障碍物目标的类别和方位,结合三维点云信息得到目标的距离信息。实验结果表明:三维点云与视觉图像融合的障碍物主动识别与距离感知算法可以兼顾实时与精准测距的需求,提高了系统预警的精确度,最大识别误差为2.356%,有助于提高无人机及线缆涂覆机器人的障碍感知能力,保障作业安全。

To address the challenges related to obstacle active recognition and distance perception in the context of unmanned aerial vehicle automatic inspection of distribution lines and maintenance involving insulation layer coating,a method combining visual images and 3D point clouds for obstacle recognition was proposed.Data augmentation preprocessing was applied to enhance the dataset of images.A deep learning approach based on feature extraction was introduced for model training,enabling the recognition of obstacle categories and orientations.Distance measurements for the detected targets were obtained by integrating 3D point cloud information.The experimental results demonstrate that the obstacle active recognition and distance perception algorithm integrated with 3D point cloud and visual image can take into account the needs of real-time and accurate ranging,the accuracy of the system′s early warning is improved,the maximum recognition error is 2.356%,which can help to improve the obstacle perception ability of unmanned aerial vehicle and cable-coated robot,thus ensuring operational safety.