门控激光相机雷达三边滤波平滑去噪算法研究

Trilateral filter smoothing and denoising algorithm for gated lightranging and imaging

基于距离能量相关选通三维成像技术的门控激光相机雷达可以同时获取高水平分辨率的强度图像和三维图像,不存在激光雷达和摄像机复合技术方案中的异源数据融合问题。然而,与现有激光雷达一样,门控激光相机雷达输出的选通三维图像依然会受到器件噪声、时域特性、目标反射率和背景环境等因素影响,导致三维图像质量降低,最终影响实际应用,特别是高水平分辨率特点下的小尺度噪声问题。针对门控激光相机雷达小尺度噪声问题,文中提出三边滤波平滑去噪算法,该算法同时利用选通图像和深度图像对三维图像进行平滑。相比传统的双边滤波,该算法引入选通图像的灰度作为第三个权重,当物体边缘与其他物体或背景存在灰度差异时,该算法有很好的边缘保持效果。仿真实验使用两种不同噪声的金字塔靶,并通过双边滤波和三边滤波对噪声进行平滑,对于噪声标准差为0.2 m和0.3 m的金字塔,三边滤波后三维图像的峰值信噪比分别平均提高10.81 dB和6.76 dB。在物理实验中,对于含有噪声的标准靶,三边滤波后峰值信噪比平均提高8.47 dB。最后采集了真实复杂场景的选通图像,完成小尺度噪声的平滑和大尺度噪声的去除,验证了三边滤波结合统计学滤波方法的实用性。

Objective Three-dimensional(3D)images provide a stereoscopic view enabling observers to perceive shapes,sizes,and spatial relationships.They play an important role in a variety of applications including computer vision,medical imaging,robotics,autonomous driving,graphics,virtual reality(VR)and augmented reality(AR).3D range gated imaging(Gated3D)is a light ranging and imaging(LiRAI)technique,which can obtain highresolution 3D spatial images as well as 2D intensity images,avoiding the heterogeneous data fusion problem in the combined technology of LiDAR and camera.Gated3D can utilize at least two gated images to recover depth information.Similar to existing LiDAR,Gated3D is also influenced by device noise,temporal characteristics,target reflectivity and background environment,which results in a low peak signal-to-noise ratio(PSNR)of 3D images,ultimately affecting practical applications.To improve the PSNR of the LiRAI based on Gated3D,we propose a trilateral filter smoothing and denoising algorithm.Methods For the gated LiRAI,the smoothing and denoising algorithm of trilateral filter utilizes spatial and range filtering in 3D images,as well as the grayscale value filtering in gated images.The trilateral filter is explained with the Eq.(1),and the spatial,range and gray-scale filtering are shown in Eq.(2)-(5).Compared to the traditional bilateral filtering algorithm,the trilateral filtering algorithm introduces grayscale value filtering.For depth maps with high noise recovered by the gated LiRAI,the proposed algorithm can preserve the object edge information while smoothing small-scale noise,even if two objects with different reflectance are relatively close together.After smoothing noise,the depth maps are transformed to point cloud,and the statistical outlier removal(SOR)algorithm is used to remove outliers in the point cloud.Results and Discussions The trilateral filter algorithm is tested on both simulated data and real-world data.In the simulation experiment,we designed a pyramid-shaped depth map to simulate the smooth regions and edge regions.The original pyramid-shaped depth maps and their depth maps with different noise are shown in Fig.3.We test bilateral filter and trilateral filter algorithms by applying them to smooth the noise in the pyramid-shaped depth maps,and the PSNR of the whole depth map as well as the edge regions of the depth map are analyzed.The results are shown(Fig.5-6,Tab.1-2).Two diffuse targets with different reflectivity are used as the real-world test data in Fig.7,and a part of the data are selected for quantitative analysis in Fig.9.The results are shown in Tab.3.In both test datasets,the trilateral filter algorithm performs better than the bilateral filter algorithm especially in the edge regions.The result of real scene by the proposed algorithm are shown in Fig.10.Conclusions We propose a noise smoothing algorithm of trilateral filter to smooth noise in 3D images from gated LiRAI,and further combine the SOR filter algorithm to remove outliers in point cloud.The 3D images recovered by the LiRAI based on Gated3D are affected by noise,in which case the traditional noise smoothing algorithm such as the bilateral filter is unable to smooth the noise and preserve edges.The proposed algorithm can preserve the edges when the targets have different reflectivity,and performs better than the bilateral filter algorithm on both synthetic data and real-scene data.