Disparity computation

Theoretical basics

The disparity computed by Pandora2D is such that:

$I_{L}(x, y) = I_{R}(x + dx, y + dy)$

with $I_{L}$ , $I_{R}$ the left image (left image) and the right image (right image), and $dx$ the column disparity and $dy$ the row disparity.

At this stage, a 4D (dims: row, col, disp_row, disp_col) cost_volumes is stored. We use the Winner-Takes-All strategy to find the right disparity for each pixel. That means we are looking for the min (resp. max for zncc and mutual information measures). For column disparities (resp. row disparities) we search the min or max in disp_row (resp. disp_col). To conclude, we extract the disparity of min (or max) from the 4D cost_volumes and we obtain two disparity maps for row and col.

Configuration and parameters

Table 8 Parameters
Name	Description	Type	Default value	Available value	Required
disparity _method	Disparity method	string	None	“wta”	Yes
invalid_disparity	Invalid disparity value	string, int, float	“NaN”	“NaN”, “inf”, int	No

Example

{
    "input" :
    {
        // input content
    },
    "pipeline" :
    {
        // ...
        "disparity":
        {
            "disparity _method": "wta",
            "invalid_disparity": "NaN"
        },
        // ...
    },
    "output":
    {
        // ...
    }
}