一种联合散焦与大气透视线索的室外场景深度估计方法

An Outdoor Scene Depth Estimation Method Combining Defocus and Atmospheric Perspective Cues

传统基于散焦测距的单目图像深度估计方法存在焦平面及模糊纹理的二义性,且对室外场景的深度估计鲁棒性较差。针对此问题,提出一种联合散焦模糊与大气透视线索的深度估计方法。该方法分别利用散焦及透视线索恢复场景深度信息,利用最小二乘优化的方式融合两种深度线索的深度值,消除或减小图像歧义对深度估计带来的负面影响,得到高精度的场景深度图。使用多种场景拍摄的真实数据进行对比实验,实验结果表明,该方法在获得完整场景深度估计的同时,有效避免焦面二义性及模糊纹理二义性造成的估计误差,在几乎不增加算法复杂度的情况下,精度提高10%~13%,提高深度估计的鲁棒性。

Most current monocular image depth estimation methods are based on depth from defocus,and because of the ambiguity of focal plane and blur texture,an accurate robust depth map is hard to obtain.To improve the accuracy and robustness of the depth map,a new depth estima⁃tion algorithm which combines both defocus and atmospheric perspective cues was proposed.In this method,defocusing and perspective cues are used respectively to recover scene depth information,and by fusing the depth values from both cues using least squares optimiza⁃tion,the proposed algorithm can eliminate the negative impact of the image ambiguity on depth estimation,and the final depth estimation is obtained.Experimental results are conducted on a large variety of real-scene images to demonstrate that the proposed method can make a reliable robust depth map of a scene while effectively avoid the estimation error caused by the ambiguity of the focal plane and the blur tex⁃ture.And the quantitative analysis shows that the accuracy is improved by 10%~13%without increasing the complexity of the algorithm.