一种基于深度图像的自遮挡检测方法

Self-occlusion Detection Approach Based on Depth Image

针对视觉目标存在的自遮挡现象,并为更好地界定、规避自遮挡现象提供依据,提出一种完全基于目标深度图像信息、仅需通过分析深度图像平均曲率变化特征并结合使用二次阈值法进行自遮挡检测的方法.为避免平均曲率计算的复杂性,该方法首先采用改进的离散正交多项式局部曲面拟合法计算深度图像的平均曲率;然后,通过分析图像各点的平均曲率并结合曲率阈值提取与其八邻域点存在曲率异号的点组成自遮挡候选点集;最后,依据候选点与以其为中心的窗口内其它点存在深度值不连续的现象,再次使用阈值法,实现对自遮挡的检测.实验结果表明该方法能够有效地检测出自遮挡现象并获得自遮挡边界.

Considering the self-occlusion phenomenon of visual object and in order to provide better basis for defining and avoiding this phenomenon,an approach based on depth image for detecting self-occlusion is proposed,which only by analyzing the change feature of mean curvature value in depth image and combining with two thresholds judgment. To avoid the calculation complexity of mean curvature value,the improved discrete orthogonal polynomials local surface fitting method is adopted. Then by analyzing the mean curvature value of each image points and combining with curvature threshold,the self-occlusion candidate point set is extracted,in which each point has different curvature sign with its eight neighborhood points. Finally,according to fact that the depth is discontinuous between the candidate point and other points in the window in which the candidate point is its center,the self-occlusion is detected by using threshold method again. Experimental results showed that the proposed approach can detect the self-occlusion phenomenon and obtain occlusion boundary effectively.