Abstract:

An algorithm was proposed to extract the valid depth information and correct the values of flying pixels with depth map and confidence map for time-of-flight camera.Firstly,the depth map was segmented adaptively based on kernel density estimation and connected component labeling.Then edge detection was performed by using a modified structure tensor to recognize the invalid pixels and the flying pixels.At last,the values of flying pixels were corrected with the bi-cubic interpolation and that of the invalid pixels were deleted by voting.Meanwhile,using augmented confidence,the pixels with wrong depth values were removed.Experimental results show the effectiveness of the proposed algorithm.Comparing with the conventional methods,the proposed algorithm can remove more invalid pixels and remain more valid depth data.Also,the proposed method is more robust to noise.

Key words: ToF camera, valid depth data, flying pixels, adaptive segmentation, edge detection, augmented confidence