Dan Strother is sharing code with you
Bitbucket is a code hosting site. Unlimited public and private repositories. Free for small teams.
Don't show this again1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 | #include <iostream>
#include <deque>
#include <algorithm>
#include <numeric>
#include "dlsc_stereobm_models.h"
void dlsc_xsobel(
const cv::Mat &in,
cv::Mat &out,
const dlsc_stereobm_params ¶ms
) {
for(int y=0;y<in.rows;++y) {
const uint8_t *r0 = y > 0 ? in.ptr<uint8_t>(y-1) : in.ptr<uint8_t>(y+1);
const uint8_t *r1 = in.ptr<uint8_t>(y);
const uint8_t *r2 = y < (in.rows-1) ? in.ptr<uint8_t>(y+1) : in.ptr<uint8_t>(y-1);
uint8_t *d = out.ptr<uint8_t>(y);
d[0] = (uint8_t)(params.data_max/2);
d[in.cols-1] = (uint8_t)(params.data_max/2);
for(int x=1;x<(in.cols-1);++x) {
int d0 = r0[x+1] - r0[x-1];
int d1 = r1[x+1] - r1[x-1];
int d2 = r2[x+1] - r2[x-1];
int v = d0 + 2*d1 + d2 + (params.data_max/2);
if(v < 0) v = 0;
else if(v > params.data_max) v = params.data_max;
d[x] = (uint8_t)v;
}
}
}
void dlsc_stereobm(
cv::Mat &il,
cv::Mat &ir,
cv::Mat &id,
cv::Mat &valid,
cv::Mat &filtered,
const dlsc_stereobm_params ¶ms
) {
std::deque<uint8_t*> rowsl; // pointers to rows within current window
std::deque<uint8_t*> rowsr; // ""
std::deque<int> sad_delay; // delay-line for accumulating column sums into a window
int *disps = new int[il.cols];
int *sads = new int[il.cols];
int *sads_thresh= new int[il.cols];
int *sads_prev = new int[il.cols]; // sad[d-1]
int *sads_lo = new int[il.cols]; // sad[mind-1]
int *sads_hi = new int[il.cols]; // sad[mind+1]
// initialize to zero, so values outside of usable area are zeroed
id = cv::Mat::zeros(il.rows,il.cols,CV_16S);
valid = cv::Mat::zeros(il.rows,il.cols,CV_8UC1);
filtered = cv::Mat::zeros(il.rows,il.cols,CV_8UC1);
// process all rows
for(int y = 0; y < il.rows; ++y) {
// accumulate pointers to rows within the params.sad_window window
rowsl.push_back(il.ptr<uint8_t>(y));
rowsr.push_back(ir.ptr<uint8_t>(y));
// can only compute disparities once we have enough rows for a complete params.sad_window window
if(y >= (params.sad_window-1)) {
short *dptr = id.ptr<short>(y-(params.sad_window/2));
uint8_t *vptr = valid.ptr<uint8_t>(y-(params.sad_window/2));
uint8_t *fptr = filtered.ptr<uint8_t>(y-(params.sad_window/2));
// initialize row disparity buffer
for(int x=0;x<il.cols;++x) {
disps[x] = 0;
sads[x] = INT_MAX/256;
sads_thresh[x] = INT_MAX/256;
sads_prev[x] = INT_MAX/256;
sads_lo[x] = INT_MAX/256;
sads_hi[x] = INT_MAX/256;
}
// process one disparity level at a time
for(int d=0;d<params.disparities;++d) {
int sad_accum = 0;
sad_delay.clear();
// process whole row at this disparity
// for(int x=params.disparities-1;x<il.cols;++x) {
for(int x=0;x<il.cols;++x) {
if(d>x) continue;
// sum column
int sad = 0;
for(int ys=0;ys<params.sad_window;++ys)
sad += abs((int)(rowsl[ys][x]) - (int)(rowsr[ys][x-d]));
// accumulate window
sad_accum += sad;
sad_delay.push_back(sad);
// once window is filled, produce output
if(sad_delay.size()==(unsigned int)params.sad_window) {
int xd = x - (params.sad_window/2);
// ** keep track of best sad **
if(sad_accum <= sads[xd]) { // favors newer/higher disparities
// update thresh
if(disps[xd] != (d-1)) {
// previous disp is not within exclusion window, so we can use it
sads_thresh[xd] = sads[xd];
} else {
// find best of previous disps's thresh and lo
if(sads_lo[xd] < sads_thresh[xd]) {
sads_thresh[xd] = sads_lo[xd];
}
}
// update disp/sad
disps[xd] = d;
sads[xd] = sad_accum;
} else if(sad_accum < sads_thresh[xd] && disps[xd] != (d-1)) {
sads_thresh[xd] = sad_accum;
}
// ** keep track of adjacent sads **
if(disps[xd] == d) {
// capture sad[mind-1]
sads_lo[xd] = sads_prev[xd];
}
if(disps[xd] == (d-1)) {
// capture sad[mind+1]
sads_hi[xd] = sad_accum;
}
sads_prev[xd] = sad_accum;
// subtract column sums falling outside of window
sad_accum -= sad_delay.front(); sad_delay.pop_front();
}
} // for(x..
} // for(d..
if(params.texture) {
// texture filtering
// (reuse params.sad_window logic)
int sad_accum = 0;
sad_delay.clear();
// process whole row
for(int x=0;x<il.cols;++x) {
// sum column
int sad = 0;
for(int ys=0;ys<params.sad_window;++ys)
sad += abs((int)(rowsl[ys][x]) - params.data_max/2);
// accumulate window
sad_accum += sad;
sad_delay.push_back(sad);
// once window is filled, produce output
if(sad_delay.size()==(unsigned int)params.sad_window) {
int xd = x - (params.sad_window/2);
if(sad_accum < params.texture) {
// below threshold; filtered
fptr[xd] = UCHAR_MAX;
}
// subtract column sums falling outside of window
sad_accum -= sad_delay.front(); sad_delay.pop_front();
}
} // for(x..
}
// ** post-process **
for(int x=(params.disparities-1+(params.sad_window/2));x<(il.cols-(params.sad_window/2));++x) {
// for(int x=(params.sad_window/2);x<(il.cols-(params.sad_window/2));++x) {
vptr[x] = UCHAR_MAX;
dptr[x] = (short)(disps[x] << params.sub_bits);
if(params.sub_bits) {
// ** sub-pixel approximation **
if(disps[x] > 0 && disps[x] < (params.disparities-1)) {
int lo = sads_lo[x] - sads[x];
int hi = sads_hi[x] - sads[x];
if( lo != hi ) {
int t = (lo>hi) ? (lo-hi) : (hi-lo);
int b = (lo>hi) ? lo : hi;
int d = (t<<(params.sub_bits+params.sub_bits_extra-1))/b;
if(lo > hi) {
dptr[x] += (short)( (d + ((1<<params.sub_bits_extra)-1)) >> params.sub_bits_extra );
} else {
dptr[x] += (short)( (((1<<params.sub_bits_extra)-1) - d) >> params.sub_bits_extra );
}
}
}
}
if(params.unique_mul) {
// ** uniqueness filtering **
int thresh = (sads[x] * (params.unique_mul+params.unique_div))/params.unique_div;
if(sads_thresh[x] <= thresh) {
fptr[x] = UCHAR_MAX;
}
}
}
// remove rows falling outside of window
rowsl.pop_front();
rowsr.pop_front();
} // if(y..
} // for(y..
// clean up
delete disps;
delete sads;
delete sads_thresh;
delete sads_prev;
delete sads_lo;
delete sads_hi;
}
void dlsc_stereobm_invoker(
cv::Mat &il,
cv::Mat &ir,
cv::Mat &ilf,
cv::Mat &irf,
cv::Mat &id,
cv::Mat &valid,
cv::Mat &filtered,
const dlsc_stereobm_params ¶ms
) {
int width = params.width;
int height = params.height;
if(width <=0) width = il.cols;
if(height<=0) height = il.rows;
if( (params.scale || il.cols < width || il.rows < height) && (il.cols != width || il.rows != height) ) {
// perform aspect-ratio-preserving scale
cv::Mat ils = il.clone();
cv::Mat irs = ir.clone();
int w,h;
float src_aspect = 1.0*il.cols/il.rows;
float aspect = 1.0*width/height;
if( src_aspect > aspect ) {
h = height;
w = (int)(height*src_aspect);
} else {
w = width;
h = (int)(width/src_aspect);
}
cv::resize(ils,il,cv::Size(w,h));
cv::resize(irs,ir,cv::Size(w,h));
}
if(il.cols != width || il.rows != height) {
// take central crop
int x = (il.cols/2) - (width/2);
int dx = x + width;
int y = (il.rows/2) - (height/2);
int dy = y + height;
il = il(cv::Range(y,dy),cv::Range(x,dx));
ir = ir(cv::Range(y,dy),cv::Range(x,dx));
}
assert(il.cols == width && il.rows == height);
ilf = il.clone();
irf = ir.clone();
if(params.xsobel) {
dlsc_xsobel(il,ilf,params);
dlsc_xsobel(ir,irf,params);
}
dlsc_stereobm(ilf,irf,id,valid,filtered,params);
filtered &= valid;
}
|