基于权值优化分块自适应灰度-距离Markov随机场的无扫描3D激光雷达距离图像重构

Reconstruction of Scannerless 3D LIDAR Range Image Based on the Adaptive Block Grayscale-Range Markov Random Filed Model with Optimizing Weights

针对无扫描三维(3D)激光雷达距离图像分辨率低、随机噪声大的问题,以高分辨率单目相机作为辅助,提出基于权值优化分块自适应灰度-距离马尔科夫随机场(MRF)的稀疏距离图像重构方法。在构建灰度-距离MRF多层次相关图基础上,采用分块快速插值策略,克服纹理拷贝并提高重构速度;采用简单线性迭代聚类(SLIC)超像素分割边缘惩罚方法,有效保护图像结构细节;采用空域距离核函数和灰度相似核函数双重引导,并针对不同邻域系统自适应调整核函数标准差,保证算法稳健性;针对各邻域系统,采用共轭梯度法实现了全局能量函数快速优化。标准图像数据和真实图像实验表明所提出方法较双线性插值、双边滤波和标准MRF方法具有更优的综合性能,可有效实现无扫描3D激光雷达距离图像重构。

Aim to improve the low resolution and noisy range image from scannerless three-dimensional （3D） LIDAR, a reconstruction approach of sparse range image based on adaptive block grayscale-range Markov random filed （MRF） with optimizing weights is proposed through integrating a monocular camera with high resolution. A grayscale-range MRF multilevel correlogram is established. On this basis a fast interpolation is obtained without the texture copying by using block processing and the resconstruciton speed is improved. The edge penalty factor based on simple linear iterative clustering （SLIC） superpixels segmentation is applied to preserve the image structure details. In order to get a robust performance, both the spatial depth kernel function and grayscale similarity kernel function with adaptive adjustment of standard deviation of kernel funciton for different neighborhood systems are used as guided map. The conjugate gradient algorithm is performed for each neighborhood system to fast optimize the global energy function. The experiments with standard image datasets and real images show that proposed method have better performance than bilinear interpolation, bilateral filter and standard MRF, so that it is effective for realizing the image reconstruction of scannerless 3D LIDAR.