This paper focuses on the effective utilization of data augmentation techniques for 3Dlidar point clouds to enhance the performance of neural network models.These point clouds,which represent spatial information throu...This paper focuses on the effective utilization of data augmentation techniques for 3Dlidar point clouds to enhance the performance of neural network models.These point clouds,which represent spatial information through a collection of 3D coordinates,have found wide-ranging applications.Data augmentation has emerged as a potent solution to the challenges posed by limited labeled data and the need to enhance model generalization capabilities.Much of the existing research is devoted to crafting novel data augmentation methods specifically for 3D lidar point clouds.However,there has been a lack of focus on making the most of the numerous existing augmentation techniques.Addressing this deficiency,this research investigates the possibility of combining two fundamental data augmentation strategies.The paper introduces PolarMix andMix3D,two commonly employed augmentation techniques,and presents a new approach,named RandomFusion.Instead of using a fixed or predetermined combination of augmentation methods,RandomFusion randomly chooses one method from a pool of options for each instance or sample.This innovative data augmentation technique randomly augments each point in the point cloud with either PolarMix or Mix3D.The crux of this strategy is the random choice between PolarMix and Mix3Dfor the augmentation of each point within the point cloud data set.The results of the experiments conducted validate the efficacy of the RandomFusion strategy in enhancing the performance of neural network models for 3D lidar point cloud semantic segmentation tasks.This is achieved without compromising computational efficiency.By examining the potential of merging different augmentation techniques,the research contributes significantly to a more comprehensive understanding of how to utilize existing augmentation methods for 3D lidar point clouds.RandomFusion data augmentation technique offers a simple yet effective method to leverage the diversity of augmentation techniques and boost the robustness of models.The insights gained from this research can pave the way for future work aimed at developing more advanced and efficient data augmentation strategies for 3D lidar point cloud analysis.展开更多
Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildi...Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildings, etc. Non-ground objects need to be removed for creation of a Digital Terrain Model (DTM) which is a continuous surface representing only ground surface points. This study aimed at comparative analysis of three main filtering approaches for stripping off non-ground objects namely;Gaussian low pass filter, focal analysis mean filter and DTM slope-based filter of varying window sizes in creation of a reliable DTM from airborne LiDAR point clouds. A sample of LiDAR data provided by the ISPRS WG III/4 captured at Vaihingen in Germany over a pure residential area has been used in the analysis. Visual analysis has indicated that Gaussian low pass filter has given blurred DTMs of attenuated high-frequency objects and emphasized low-frequency objects while it has achieved improved removal of non-ground object at larger window sizes. Focal analysis mean filter has shown better removal of nonground objects compared to Gaussian low pass filter especially at large window sizes where details of non-ground objects almost have diminished in the DTMs from window sizes of 25 × 25 and greater. DTM slope-based filter has created bare earth models that have been full of gabs at the positions of the non-ground objects where the sizes and numbers of that gabs have increased with increasing the window sizes of filter. Those gaps have been closed through exploitation of the spline interpolation method in order to get continuous surface representing bare earth landscape. Comparative analysis has shown that the minimum elevations of the DTMs increase with increasing the filter widow sizes till 21 × 21 and 31 × 31 for the Gaussian low pass filter and the focal analysis mean filter respectively. On the other hand, the DTM slope-based filter has kept the minimum elevation of the original data, that could be due to noise in the LiDAR data unchanged. Alternatively, the three approaches have produced DTMs of decreasing maximum elevation values and consequently decreasing ranges of elevations due to increases in the filter window sizes. Moreover, the standard deviations of the created DTMs from the three filters have decreased with increasing the filter window sizes however, the decreases have been continuous and steady in the cases of the Gaussian low pass filter and the focal analysis mean filters while in the case of the DTM slope-based filter the standard deviations of the created DTMs have decreased with high rates till window size of 31 × 31 then they have kept unchanged due to more increases in the filter window sizes.展开更多
为了提高基于车载LiDAR点云数据的地面点提取精度,最大限度发挥车载LiDAR点云数据的利用效率。本文提出了一种基于Otsu算法与最远点采样(Farthest Point Sampling,FPS)算法的地面点提取方法。首先,根据Otsu算法自适应计算高程分割阈值,...为了提高基于车载LiDAR点云数据的地面点提取精度,最大限度发挥车载LiDAR点云数据的利用效率。本文提出了一种基于Otsu算法与最远点采样(Farthest Point Sampling,FPS)算法的地面点提取方法。首先,根据Otsu算法自适应计算高程分割阈值,滤除大部分地面点,实现地面点的粗提取;其次从地面点粗提取结果中随机选取一个种子点,并使用FPS算法选取剩余种子点进行最优平面模型拟合提取精确地面点。通过两组实验数据进行地面点提取实验,结果表明,两组实验数据地面点提取结果误差均在8%以内,验证了本文方法的有效性与适用性。展开更多
基金funded in part by the Key Project of Nature Science Research for Universities of Anhui Province of China(No.2022AH051720)in part by the Science and Technology Development Fund,Macao SAR(Grant Nos.0093/2022/A2,0076/2022/A2 and 0008/2022/AGJ)in part by the China University Industry-University-Research Collaborative Innovation Fund(No.2021FNA04017).
文摘This paper focuses on the effective utilization of data augmentation techniques for 3Dlidar point clouds to enhance the performance of neural network models.These point clouds,which represent spatial information through a collection of 3D coordinates,have found wide-ranging applications.Data augmentation has emerged as a potent solution to the challenges posed by limited labeled data and the need to enhance model generalization capabilities.Much of the existing research is devoted to crafting novel data augmentation methods specifically for 3D lidar point clouds.However,there has been a lack of focus on making the most of the numerous existing augmentation techniques.Addressing this deficiency,this research investigates the possibility of combining two fundamental data augmentation strategies.The paper introduces PolarMix andMix3D,two commonly employed augmentation techniques,and presents a new approach,named RandomFusion.Instead of using a fixed or predetermined combination of augmentation methods,RandomFusion randomly chooses one method from a pool of options for each instance or sample.This innovative data augmentation technique randomly augments each point in the point cloud with either PolarMix or Mix3D.The crux of this strategy is the random choice between PolarMix and Mix3Dfor the augmentation of each point within the point cloud data set.The results of the experiments conducted validate the efficacy of the RandomFusion strategy in enhancing the performance of neural network models for 3D lidar point cloud semantic segmentation tasks.This is achieved without compromising computational efficiency.By examining the potential of merging different augmentation techniques,the research contributes significantly to a more comprehensive understanding of how to utilize existing augmentation methods for 3D lidar point clouds.RandomFusion data augmentation technique offers a simple yet effective method to leverage the diversity of augmentation techniques and boost the robustness of models.The insights gained from this research can pave the way for future work aimed at developing more advanced and efficient data augmentation strategies for 3D lidar point cloud analysis.
文摘Light Detection And Ranging (LiDAR) is a well-established active remote sensing technology that can provide accurate digital elevation measurements for the terrain and non-ground objects such as vegetations and buildings, etc. Non-ground objects need to be removed for creation of a Digital Terrain Model (DTM) which is a continuous surface representing only ground surface points. This study aimed at comparative analysis of three main filtering approaches for stripping off non-ground objects namely;Gaussian low pass filter, focal analysis mean filter and DTM slope-based filter of varying window sizes in creation of a reliable DTM from airborne LiDAR point clouds. A sample of LiDAR data provided by the ISPRS WG III/4 captured at Vaihingen in Germany over a pure residential area has been used in the analysis. Visual analysis has indicated that Gaussian low pass filter has given blurred DTMs of attenuated high-frequency objects and emphasized low-frequency objects while it has achieved improved removal of non-ground object at larger window sizes. Focal analysis mean filter has shown better removal of nonground objects compared to Gaussian low pass filter especially at large window sizes where details of non-ground objects almost have diminished in the DTMs from window sizes of 25 × 25 and greater. DTM slope-based filter has created bare earth models that have been full of gabs at the positions of the non-ground objects where the sizes and numbers of that gabs have increased with increasing the window sizes of filter. Those gaps have been closed through exploitation of the spline interpolation method in order to get continuous surface representing bare earth landscape. Comparative analysis has shown that the minimum elevations of the DTMs increase with increasing the filter widow sizes till 21 × 21 and 31 × 31 for the Gaussian low pass filter and the focal analysis mean filter respectively. On the other hand, the DTM slope-based filter has kept the minimum elevation of the original data, that could be due to noise in the LiDAR data unchanged. Alternatively, the three approaches have produced DTMs of decreasing maximum elevation values and consequently decreasing ranges of elevations due to increases in the filter window sizes. Moreover, the standard deviations of the created DTMs from the three filters have decreased with increasing the filter window sizes however, the decreases have been continuous and steady in the cases of the Gaussian low pass filter and the focal analysis mean filters while in the case of the DTM slope-based filter the standard deviations of the created DTMs have decreased with high rates till window size of 31 × 31 then they have kept unchanged due to more increases in the filter window sizes.
文摘为了提高基于车载LiDAR点云数据的地面点提取精度,最大限度发挥车载LiDAR点云数据的利用效率。本文提出了一种基于Otsu算法与最远点采样(Farthest Point Sampling,FPS)算法的地面点提取方法。首先,根据Otsu算法自适应计算高程分割阈值,滤除大部分地面点,实现地面点的粗提取;其次从地面点粗提取结果中随机选取一个种子点,并使用FPS算法选取剩余种子点进行最优平面模型拟合提取精确地面点。通过两组实验数据进行地面点提取实验,结果表明,两组实验数据地面点提取结果误差均在8%以内,验证了本文方法的有效性与适用性。
