3DTV depth map reconstruction based on structured light scheme

Abstract

In this paper, we proposed an algorithm of 3DTV depth map reconstruction based on structured light scheme. We used a projector to project the structured light scheme pattern and captured the pattern by digital camera. After a series of edge extraction algorithm which contains bi-direction Sobel operator, morphology filter, edge thinning algorithm, pattern matching, and gradient calculation, we built the gradient database to convert gradient to physical distance to replace traditional complex triangulation algorithm. That is more efficient and accurate. In addition, we took humans' perception into consideration. It not only output the physical distance depth map but also quantified it. Moreover, through DIBR technology, we perceived three-dimensional effects on 3D display.

VLSI Design of a Depth Map Estimation Circuit Based on Structured Light Algorithm

“…Nonformal codification [14]- [16], as the name suggests, is based on self-defined sample design, where all colors in the transform hue-saturation-value (HSV) color space will be represented. Fan et al [17] proposed an algorithm of 3-D television (3-DTV) depth map reconstruction based on structured light scheme. The coding for the patterns encoded with De Bruijn sequences [18]- [24] is based on the De Bruijn sequence, and the color is selected in accordance with HSV color space.…”

“…The structuring element S used in this system is a square structuring element with the size of 5 × 5 [17]. After the opening operation of the morphological filter, nonedge noises are found because the noises going through the erosion are eliminated, and voids in the information are filled to form a smooth edge by the dilation edge after the erosion takes place.…”

“…When a pixel satisfies the aforementioned three conditions, the pixel value will be set to 255, or else the pixel value will be set to 0. Where P(x, y) is the executing pixel, and the threshold is set according to the statistics of the experiment and has a numeric value of 75 to obtain a mask for a better object segmentation [17]. After the determination for red, green, and blue pixel values, we can separate the image from the black background to form a binary picture.…”

“…Then, we start establishing a structure light pattern for projection, and adopt black-and-white stripe sample patterns arranged by a spatial neighborhood method. We generate the alternately arranged black-and-white sample patterns according to [17] Z (x, y) = 255 × Mod(y ÷ width) (1) where x and y represent the x-axis pixel value and y-axis pixel value of the projection, respectively, width is equal to the pixel width required for producing the lines of the structure light pattern, and Z (x, y) represents the pixel value of the x and y pixels of the produced sample pattern and only has two possible values, respectively, 0 and 255.…”

“…Therefore, a masking method is used to reduce the complexity of the computation significantly. We employ a Gaussian low pass filter with a mask size of 3 × 3 to obtain eight adjacent pixel and original position pixel according to [17] g(P 0 ) = p 1 16…”