Abstrakt

Reconstruction of kinect sensor’s depth maps based on inpainting and CS theory

Qian Zhang, Bin Cui, ShaoMin Li, Pei Wang, Jifeng Huang

The Microsoft’s Kinect sensor can generate a real-time dense depth map of a indoor scene and relatively commercial available. However, there are some artifacts like holes, instability of the raw input data which seriously affect the application. In this paper,by using inpainting algorithm and CS theory which enable the kinect generate a dense depth map, the large holes be inpainted without blurring between foreground and background and the edges of the objects are sharpened. Experiments on captured data including complex foreground objects and large areas of missing values demonstrate the effectiveness of the method.