基于改进Criminisi算法的自动驾驶仿真道路场景图像修复

Image Restoration of Automatic Driving Simulation Road Scene based on Improved Criminisi Algorithm

利用真实图像生成虚拟驾驶场景视图时,首先要从图像中移除车辆、行人等前景目标,以便获得不包含移动目标的背景图像。针对原始Criminisi算法存在的图像破损区域在修复后会出现模糊效应、边界不够平滑的弊端,提出了一种在处理掩膜图像环节使用开运算,并优化原始优先权函数的改进算法。运用MATLAB仿真平台进行实验分析,结果表明改进后算法修复时间平均缩短了16.5 s,峰值信噪比平均高出原算法1.65 dB。该算法还原了图像中更多的道路信息,弥补了Criminisi算法的不足。

When generating virtual driving scene view from real images,the foreground objects such as vehicles and pedestrians should be removed from the image first,so to obtain a background image without moving objects.Aiming at the shortcomings of the original Criminisi algorithm that the damaged area of the image will have fuzzy effect and the boundary is not smooth enough after repairing,this paper proposes an improved algorithm,which uses open operation and optimizes the priority function in the process of mask image.MATLAB Simulation platform is used for experimental analysis.The results show that the restoring time of the improved algorithm is shortened by 16.5 seconds on average,and the peak signal-to-noise ratio is 1.65 dB higher than the original algorithm on average.The improved algorithm restores more road information in the image and makes up for the deficiency of Criminisi algorithm.