Kinect Based Depth Map Restoration And Saliency Detection

In recent years,Depth map based research has gained its popularity thanks to the appearance of Kinect.However,limited by the performance of the equipment,the depth map of Kinect will show some defects such as dismatching with the RGB image and missing information in some area,which has seriously limited its application performance.Meanwhile,the depth map is generally used to capture and track the dynamic object but studies incorporating depth and RGB information synergistically have merely been proposed.Based on the above background,this paper starts from the depth map restoration,then study and design the depth feature and apply the depth information to detect the saliency object combined with the RGB image.The main works and innovations of this article can be summarized as followed:1.Neighborhood-based Support for Smooth Depth Map Restoration: The model calculates the smoothness of the pixel via the corresponding RGB image and can well capture the local structure of the object.Combined with the joint bilateral filter,this model is able to not only restore the depth map but also preserve local depth discontinuities.Experimental results on benchmark databases demonstrate the better performance of our proposed approach compared with several state-of-art methods.2.Depth Map based Superpixel Segmentation: We design the depth feature of depth map to measure the similarity of the depth pixels and then improve the SLIC algorithm to make better performance in depth map superpixel segmentation.The algorithm is also used in the following saliency detection model.3.Saliency Detection Combining RGB Image and Depth Map: In this model,the surface consistency and RGB information are fused to make a better performance of saliency object detection.Experimental results indicate that our model achieves better precision and integrity of the saliency object.