How to improve Depth map Accuracy generated from Stereo Images?

Hey

How can I improve the Depth Map Quality??

I am following Depth Estimation From Stereo Video Documentation.

Below is my two L&R images.

And, Below is what I achieved from

points3D(:, :, 3)

Below, is Ideal Depth Map

%%This is my entire code :)
clc
clear all
close all
load('stereoParams.mat')
frameLeft = imread('Left.png');
frameRight = imread('Right.png');
[frameLeftRect, frameRightRect] = rectifyStereoImages(frameLeft, frameRight, stereoParams);
frameLeftGray  = rgb2gray(frameLeftRect);
frameRightGray = rgb2gray(frameRightRect);
disparityMap = disparity(frameLeftGray, frameRightGray, 'BlockSize', 15,'DisparityRange', [0 64]);
disparityMap = medfilt2(disparityMap);
points3D = reconstructScene(disparityMap, stereoParams);
Z = points3D(:, :, 3)./ 1000;
imtool(Z, [0, 100]);

[Madhusudan Govindraju]

have you tried different block sizes ?

Answer 1

[Sukhendu]

What is the level of reprojection error in stereo calibration ?

What is the dataset (details, samples) used to compute stereo calibration pararmeters, on your street view (assuming you have not used, matlab default) ?