结合彩色图像局部分割的Kinect深度图修复算法

Local segmentation of color image for Kinect depth recovery

针对Kinect深度图空洞和噪声存在降低三维目标检测和识别精度的问题,提出一种结合彩色图像局部分割的深度图修复算法。首先,利用区域连通性对空洞区域进行分析和标记;其次,根据空洞区域领域的深度直方图峰值数目判断空洞区域类型;再次,分割膨胀区域对应的彩色区域,获取空洞像素所在分割区域与其领域的交域(有效支撑区域),利用交域内的有效深度信息计算空洞像素;最后,利用引导滤波对深度图进行平滑、去噪。实验结果表明,经该算法处理后的深度图边缘细节更加清晰,具备微小梯度不变性,且测量精确度更高。

Considering the problem that hole and noise in the depth map captured by Kinect decreased the accuracy of 3 D object detection and recognition,this paper proposed a depth recovery method utilizing valid support region. Firstly,the proposed method labeled depth holes based on 8-connectivity. Secondly,for each labeled depth hole,the proposed method determined depth hole's type using the number of dominant peaks of depth histogram of neighboring pixels of depth hole. Thirdly,for a pixel in depth map hole,it calculated its depth value using its known depth values of valid support region. Finally,the algorithm used fast guided filter to smooth the depth map. The experimental results show that the proposed algorithm can repair the noise pixels while remaining the object profiles unchanged and keeping the depth gradient. Therefore,it improves the recovery accuracy of depth map.