Knowledge reserve of FPGA binocular matching algorithm

Approximate step flow

1, camera calibration, image correction

2, local matching algorithm :

A, Ad-census cost initialization: Calculates the difference between the pairs of binocular images corresponding to all parallax levels, and the resulting result is called the initial cost.

B, Cost statistics

(Fixed support domain and variable support domain of local matching algorithm, but the cost statistic based on fixed support domain is easy to implement, but it often causes the depth graph to blur at the edge of object.)

variable support domain for semi-global optimization )

C, Parallax Selection

D, post-processing

Knowledge reserve of FPGA binocular matching algorithm