Commit fabadb74 authored by Andrey Filippov's avatar Andrey Filippov

committing remained files accidentally not commited

parent 222e8bbe
...@@ -48,7 +48,7 @@ extern "C" __global__ void convert_direct( // called with a single block, single ...@@ -48,7 +48,7 @@ extern "C" __global__ void convert_direct( // called with a single block, single
float ** gpu_kernels, // [NUM_CAMS], float ** gpu_kernels, // [NUM_CAMS],
float ** gpu_images, // [NUM_CAMS], float ** gpu_images, // [NUM_CAMS],
struct tp_task * gpu_tasks, struct tp_task * gpu_tasks,
float ** gpu_clt, // [NUM_CAMS][TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float ** gpu_clt, // [NUM_CAMS][TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
size_t dstride, // in floats (pixels) size_t dstride, // in floats (pixels)
int num_tiles, // number of tiles in task int num_tiles, // number of tiles in task
int lpf_mask, // apply lpf to colors : bit 0 - red, bit 1 - blue, bit2 - green. Now - always 0 ! int lpf_mask, // apply lpf to colors : bit 0 - red, bit 1 - blue, bit2 - green. Now - always 0 !
...@@ -57,10 +57,12 @@ extern "C" __global__ void convert_direct( // called with a single block, single ...@@ -57,10 +57,12 @@ extern "C" __global__ void convert_direct( // called with a single block, single
int kernels_hor, int kernels_hor,
int kernels_vert, int kernels_vert,
int * gpu_active_tiles, // pointer to the calculated number of non-zero tiles int * gpu_active_tiles, // pointer to the calculated number of non-zero tiles
int * pnum_active_tiles); // indices to gpu_tasks int * pnum_active_tiles, // indices to gpu_tasks
int tilesx);
extern "C" __global__ void correlate2D( extern "C" __global__ void correlate2D(
float ** gpu_clt, // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float ** gpu_clt, // [NUM_CAMS] ->[TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
int colors, // number of colors (3/1) int colors, // number of colors (3/1)
float scale0, // scale for R float scale0, // scale for R
float scale1, // scale for B float scale1, // scale for B
...@@ -68,41 +70,58 @@ extern "C" __global__ void correlate2D( ...@@ -68,41 +70,58 @@ extern "C" __global__ void correlate2D(
float fat_zero, // here - absolute float fat_zero, // here - absolute
struct tp_task * gpu_tasks, // array of per-tile tasks (now bits 4..9 - correlation pairs) struct tp_task * gpu_tasks, // array of per-tile tasks (now bits 4..9 - correlation pairs)
int num_tiles, // number of tiles in task int num_tiles, // number of tiles in task
int tilesx, // number of tile rows
int * gpu_corr_indices, // packed tile+pair int * gpu_corr_indices, // packed tile+pair
int * pnum_corr_tiles, // pointer to a number of correlation tiles to process int * pnum_corr_tiles, // pointer to a number of correlation tiles to process
const size_t corr_stride, // in floats const size_t corr_stride, // in floats
int corr_radius, // radius of the output correlation (7 for 15x15) int corr_radius, // radius of the output correlation (7 for 15x15)
float * gpu_corrs); // correlation output data float * gpu_corrs); // correlation output data
extern "C" __global__ void corr2D_normalize(
int num_corr_tiles, // number of correlation tiles to process
const size_t corr_stride_td, // in floats
float * gpu_corrs_td, // correlation tiles in transform domain
const size_t corr_stride, // in floats
float * gpu_corrs, // correlation output data (either pixel domain or transform domain
float fat_zero, // here - absolute
int corr_radius); // radius of the output correlation (7 for 15x15)
extern "C" __global__ void corr2D_combine(
int num_tiles, // number of tiles to process (each with num_pairs)
int num_pairs, // num pairs per tile (should be the same)
int init_output, // !=0 - reset output tiles to zero before accumulating
int pairs_mask, // selected pairs (0x3 - horizontal, 0xc - vertical, 0xf - quad, 0x30 - cross)
int * gpu_corr_indices, // packed tile+pair
int * gpu_combo_indices, // output if noty null: packed tile+pairs_mask (will point to the first used pair
const size_t corr_stride, // (in floats) stride for the input TD correlations
float * gpu_corrs, // input correlation tiles
const size_t corr_stride_combo, // (in floats) stride for the output TD correlations (same as input)
float * gpu_corrs_combo); // combined correlation output (one per tile)
extern "C" __global__ void textures_nonoverlap( extern "C" __global__ void textures_nonoverlap(
struct tp_task * gpu_tasks, struct tp_task * gpu_tasks,
int num_tiles, // number of tiles in task list int num_tiles, // number of tiles in task list
// int num_tilesx, // number of tiles in a row
// declare arrays in device code? // declare arrays in device code?
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7) int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
int * pnum_texture_tiles, // returns total number of elements in gpu_texture_indices array int * pnum_texture_tiles, // returns total number of elements in gpu_texture_indices array
float ** gpu_clt, // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float ** gpu_clt, // [NUM_CAMS] ->[TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
// TODO: use geometry_correction rXY ! // TODO: use geometry_correction rXY !
struct gc * gpu_geometry_correction, struct gc * gpu_geometry_correction,
int colors, // number of colors (3/1) int colors, // number of colors (3/1)
int is_lwir, // do not perform shot correction int is_lwir, // do not perform shot correction
float params[5], float params[5],
// float min_shot, // 10.0
// float scale_shot, // 3.0
// float diff_sigma, // pixel value/pixel change
// float diff_threshold, // pixel value/pixel change
// float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
float weights[3], // scale for R,B,G float weights[3], // scale for R,B,G
int dust_remove, // Do not reduce average weight when only one image differs much from the average int dust_remove, // Do not reduce average weight when only one image differs much from the average
// int keep_weights, // return channel weights after A in RGBA (was removed) (should be 0 if gpu_texture_rbg)?
// combining both non-overlap and overlap (each calculated if pointer is not null ) // combining both non-overlap and overlap (each calculated if pointer is not null )
size_t texture_stride, // in floats (now 256*4 = 1024) // may be 0 if not needed size_t texture_stride, // in floats (now 256*4 = 1024) // may be 0 if not needed
float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles // may be 0 if not needed float * gpu_texture_tiles, // (number of colors +1 + ?)*16*16 rgba texture tiles // may be 0 if not needed
float * gpu_diff_rgb_combo); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS] // may be 0 if not needed float * gpu_diff_rgb_combo, //); // diff[NUM_CAMS], R[NUM_CAMS], B[NUM_CAMS],G[NUM_CAMS] // may be 0 if not needed
int num_tilesx);
extern "C" extern "C"
__global__ void imclt_rbg_all( __global__ void imclt_rbg_all(
float ** gpu_clt, // [NUM_CAMS][TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float ** gpu_clt, // [NUM_CAMS][TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
float ** gpu_corr_images, // [NUM_CAMS][WIDTH, 3 * HEIGHT] float ** gpu_corr_images, // [NUM_CAMS][WIDTH, 3 * HEIGHT]
int apply_lpf, int apply_lpf,
int colors, int colors,
...@@ -111,7 +130,7 @@ __global__ void imclt_rbg_all( ...@@ -111,7 +130,7 @@ __global__ void imclt_rbg_all(
const size_t dstride); // in floats (pixels) const size_t dstride); // in floats (pixels)
extern "C" __global__ void imclt_rbg( extern "C" __global__ void imclt_rbg(
float * gpu_clt, // [TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float * gpu_clt, // [TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
float * gpu_rbg, // WIDTH, 3 * HEIGHT float * gpu_rbg, // WIDTH, 3 * HEIGHT
int apply_lpf, int apply_lpf,
int mono, // defines lpf filter int mono, // defines lpf filter
...@@ -130,22 +149,15 @@ extern "C" __global__ void generate_RBGA( ...@@ -130,22 +149,15 @@ extern "C" __global__ void generate_RBGA(
int * gpu_texture_indices,// packed tile + bits (now only (1 << 7) int * gpu_texture_indices,// packed tile + bits (now only (1 << 7)
int * num_texture_tiles, // number of texture tiles to process (8 separate elements for accumulation) int * num_texture_tiles, // number of texture tiles to process (8 separate elements for accumulation)
int * woi, // x,y,width,height of the woi int * woi, // x,y,width,height of the woi
int width, // <= TILESX, use for faster processing of LWIR images (should be actual + 1) int width, // <= TILES-X, use for faster processing of LWIR images (should be actual + 1)
int height, // <= TILESY, use for faster processing of LWIR images int height, // <= TILES-Y, use for faster processing of LWIR images
// Parameters for the texture generation // Parameters for the texture generation
float ** gpu_clt, // [NUM_CAMS] ->[TILESY][TILESX][NUM_COLORS][DTT_SIZE*DTT_SIZE] float ** gpu_clt, // [NUM_CAMS] ->[TILES-Y][TILES-X][NUM_COLORS][DTT_SIZE*DTT_SIZE]
// TODO: use geometry_correction rXY ! // TODO: use geometry_correction rXY !
struct gc * gpu_geometry_correction, struct gc * gpu_geometry_correction,
int colors, // number of colors (3/1) int colors, // number of colors (3/1)
int is_lwir, // do not perform shot correction int is_lwir, // do not perform shot correction
float params[5], // mitigating CUDA_ERROR_INVALID_PTX float params[5], // mitigating CUDA_ERROR_INVALID_PTX
/*
float min_shot, // 10.0
float scale_shot, // 3.0
float diff_sigma, // pixel value/pixel change
float diff_threshold, // pixel value/pixel change
float min_agree, // minimal number of channels to agree on a point (real number to work with fuzzy averages)
*/
float weights[3], // scale for R,B,G float weights[3], // scale for R,B,G
int dust_remove, // Do not reduce average weight when only one image differs much from the average int dust_remove, // Do not reduce average weight when only one image differs much from the average
int keep_weights, // return channel weights after A in RGBA (was removed) int keep_weights, // return channel weights after A in RGBA (was removed)
......
...@@ -294,7 +294,6 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -294,7 +294,6 @@ extern "C" __global__ void get_tiles_offsets(
float * gpu_rByRDist, // length should match RBYRDIST_LEN float * gpu_rByRDist, // length should match RBYRDIST_LEN
trot_deriv * gpu_rot_deriv) trot_deriv * gpu_rot_deriv)
{ {
// int task_num = blockIdx.x * blockDim.x + threadIdx.x; // blockIdx.x * TILES_PER_BLOCK_GEOM + threadIdx.x
int task_num = blockIdx.x * blockDim.y + threadIdx.y; // blockIdx.x * TILES_PER_BLOCK_GEOM + threadIdx.y int task_num = blockIdx.x * blockDim.y + threadIdx.y; // blockIdx.x * TILES_PER_BLOCK_GEOM + threadIdx.y
if (task_num >= num_tiles){ if (task_num >= num_tiles){
return; return;
...@@ -306,6 +305,7 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -306,6 +305,7 @@ extern "C" __global__ void get_tiles_offsets(
__shared__ float rByRDist [RBYRDIST_LEN]; __shared__ float rByRDist [RBYRDIST_LEN];
__shared__ struct corr_vector extrinsic_corr; __shared__ struct corr_vector extrinsic_corr;
__shared__ trot_deriv rot_deriv; __shared__ trot_deriv rot_deriv;
__shared__ float pY_offsets[TILES_PER_BLOCK_GEOM][NUM_CAMS];
float pXY[2]; // result to be copied to task float pXY[2]; // result to be copied to task
// copy data common to all threads // copy data common to all threads
{ {
...@@ -362,8 +362,7 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -362,8 +362,7 @@ extern "C" __global__ void get_tiles_offsets(
(extrinsic_corr.imu_move[0] != 0.0) || (extrinsic_corr.imu_move[0] != 0.0) ||
(extrinsic_corr.imu_move[1] != 0.0) || (extrinsic_corr.imu_move[1] != 0.0) ||
(extrinsic_corr.imu_move[2] != 0.0); (extrinsic_corr.imu_move[2] != 0.0);
// Temporary
imu_exists = 0;
#ifdef DEBUG21 #ifdef DEBUG21
if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){ if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){
printf("\nTile = %d, camera= %d\n", task_num, ncam); printf("\nTile = %d, camera= %d\n", task_num, ncam);
...@@ -373,6 +372,9 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -373,6 +372,9 @@ extern "C" __global__ void get_tiles_offsets(
} }
__syncthreads();// __syncwarp(); __syncthreads();// __syncwarp();
#endif // DEBUG21 #endif // DEBUG21
// String dbg_s = corr_vector.toString(); // String dbg_s = corr_vector.toString();
/* Starting with required tile center X, Y and nominal distortion, for each sensor port: /* Starting with required tile center X, Y and nominal distortion, for each sensor port:
* 1) unapply common distortion (maybe for different - master camera) * 1) unapply common distortion (maybe for different - master camera)
...@@ -401,15 +403,10 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -401,15 +403,10 @@ extern "C" __global__ void get_tiles_offsets(
float pXcd = px - 0.5 * geometry_correction.pixelCorrectionWidth; float pXcd = px - 0.5 * geometry_correction.pixelCorrectionWidth;
float pYcd = py - 0.5 * geometry_correction.pixelCorrectionHeight; float pYcd = py - 0.5 * geometry_correction.pixelCorrectionHeight;
// float rXY [NUM_CAMS][2];
float rXY [2]; float rXY [2];
// for (int i = 0; i < NUM_CAMS;i++){
// rXY[ncam][0] = geometry_correction.rXY[ncam][0];
// rXY[ncam][1] = geometry_correction.rXY[ncam][1];
rXY[0] = geometry_correction.rXY[ncam][0]; rXY[0] = geometry_correction.rXY[ncam][0];
rXY[1] = geometry_correction.rXY[ncam][1]; rXY[1] = geometry_correction.rXY[ncam][1];
// }
float rD = sqrtf(pXcd*pXcd + pYcd*pYcd)*0.001*geometry_correction.pixelSize; // distorted radius in a virtual center camera float rD = sqrtf(pXcd*pXcd + pYcd*pYcd)*0.001*geometry_correction.pixelSize; // distorted radius in a virtual center camera
float rND2R=getRByRDist(rD/geometry_correction.distortionRadius, rByRDist); float rND2R=getRByRDist(rD/geometry_correction.distortionRadius, rByRDist);
...@@ -489,9 +486,17 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -489,9 +486,17 @@ extern "C" __global__ void get_tiles_offsets(
float pYid = pYci * rD2rND; float pYid = pYci * rD2rND;
pXY[0] = pXid + geometry_correction.pXY0[ncam][0]; pXY[0] = pXid + geometry_correction.pXY0[ncam][0];
pXY[1] = pYid + geometry_correction.pXY0[ncam][1]; pXY[1] = pYid + geometry_correction.pXY0[ncam][1];
// new for ERS
pY_offsets[threadIdx.y][ncam] = pXY[1] - geometry_correction.woi_tops[ncam];
__syncthreads();
// Each thread re-calculate same sum
float lines_avg = 0;
for (int i = 0; i < NUM_CAMS; i ++){
lines_avg += pY_offsets[threadIdx.y][i];
}
lines_avg *= (1.0/NUM_CAMS);
// used when calculating derivatives, TODO: combine calculations ! // used when calculating derivatives, TODO: combine calculations !
float pY_offset = pY_offsets[threadIdx.y][ncam] - lines_avg;
#ifdef DEBUG21 #ifdef DEBUG21
if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){ if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){
printf("pXci0 = %f, pYci0 = %f\n", pXci0, pYci0); printf("pXci0 = %f, pYci0 = %f\n", pXci0, pYci0);
...@@ -501,6 +506,7 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -501,6 +506,7 @@ extern "C" __global__ void get_tiles_offsets(
printf("rD2rND = %f\n", rD2rND); printf("rD2rND = %f\n", rD2rND);
printf("pXid = %f, pYid = %f\n", pXid, pYid); printf("pXid = %f, pYid = %f\n", pXid, pYid);
printf("pXY[0] = %f, pXY[1] = %f\n", pXY[0], pXY[1]); // OK printf("pXY[0] = %f, pXY[1] = %f\n", pXY[0], pXY[1]); // OK
printf("lines_avg = %f, pY_offset = %f\n", lines_avg, pY_offset);
} }
__syncthreads();// __syncwarp(); __syncthreads();// __syncwarp();
#endif // DEBUG21 #endif // DEBUG21
...@@ -514,14 +520,10 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -514,14 +520,10 @@ extern "C" __global__ void get_tiles_offsets(
#pragma unroll #pragma unroll
for (int j = 0; j< 3; j++){ for (int j = 0; j< 3; j++){
// drvi_daz[j] = rot_deriv.d_daz[ncam][j][0] * rvi[0] + rot_deriv.d_daz[ncam][j][1] * rvi[1] + rot_deriv.d_daz[ncam][j][2] * rvi[2];
// drvi_dtl[j] = rot_deriv.d_tilt[ncam][j][0] * rvi[0] + rot_deriv.d_tilt[ncam][j][1] * rvi[1] + rot_deriv.d_tilt[ncam][j][2] * rvi[2];
// drvi_drl[j] = rot_deriv.d_roll[ncam][j][0] * rvi[0] + rot_deriv.d_roll[ncam][j][1] * rvi[1] + rot_deriv.d_roll[ncam][j][2] * rvi[2];
drvi_daz[j] = rot_deriv.d_daz[ncam][j][0] * pXci0 + rot_deriv.d_daz[ncam][j][1] * pYci0 + rot_deriv.d_daz[ncam][j][2] * fl_pix; drvi_daz[j] = rot_deriv.d_daz[ncam][j][0] * pXci0 + rot_deriv.d_daz[ncam][j][1] * pYci0 + rot_deriv.d_daz[ncam][j][2] * fl_pix;
drvi_dtl[j] = rot_deriv.d_tilt[ncam][j][0] * pXci0 + rot_deriv.d_tilt[ncam][j][1] * pYci0 + rot_deriv.d_tilt[ncam][j][2] * fl_pix; drvi_dtl[j] = rot_deriv.d_tilt[ncam][j][0] * pXci0 + rot_deriv.d_tilt[ncam][j][1] * pYci0 + rot_deriv.d_tilt[ncam][j][2] * fl_pix;
drvi_drl[j] = rot_deriv.d_roll[ncam][j][0] * pXci0 + rot_deriv.d_roll[ncam][j][1] * pYci0 + rot_deriv.d_roll[ncam][j][2] * fl_pix; drvi_drl[j] = rot_deriv.d_roll[ncam][j][0] * pXci0 + rot_deriv.d_roll[ncam][j][1] * pYci0 + rot_deriv.d_roll[ncam][j][2] * fl_pix;
} }
// double [][] avi = {{pXci0}, {pYci0},{fl_pix}};
float dpXci_dazimuth = drvi_daz[0] * norm_z - pXci * drvi_daz[2] / rvi[2]; float dpXci_dazimuth = drvi_daz[0] * norm_z - pXci * drvi_daz[2] / rvi[2];
float dpYci_dazimuth = drvi_daz[1] * norm_z - pYci * drvi_daz[2] / rvi[2]; float dpYci_dazimuth = drvi_daz[1] * norm_z - pYci * drvi_daz[2] / rvi[2];
...@@ -573,25 +575,6 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -573,25 +575,6 @@ extern "C" __global__ void get_tiles_offsets(
// unity vector in the direction of radius // unity vector in the direction of radius
float c_dist = pXci/rNDi; float c_dist = pXci/rNDi;
float s_dist = pYci/rNDi; float s_dist = pYci/rNDi;
/*
double [][] arot2= {
{c_dist, s_dist},
{-s_dist, c_dist}};
Matrix rot2 = new Matrix(arot2); // convert from non-distorted X,Y to parallel and perpendicular (CCW) to the radius
double [][] ascale_distort = {
{rD2rND + ri* drD2rND_dri, 0 },
{0, rD2rND}};
Matrix scale_distort = new Matrix(ascale_distort); // scale component parallel to radius as distortion derivative, perpendicular - as distortion
Matrix dd2 = rot2.transpose().times(scale_distort).times(rot2).times(dd1);
disp_dist[i][0] = dd2.get(0, 0);
disp_dist[i][1] = dd2.get(0, 1);
disp_dist[i][2] = dd2.get(1, 0); // d_py/d_disp
disp_dist[i][3] = dd2.get(1, 1);
*/
//#undef NVRTC_BUG //#undef NVRTC_BUG
float drD2rND_dri = 0.0; float drD2rND_dri = 0.0;
{ {
...@@ -612,11 +595,6 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -612,11 +595,6 @@ extern "C" __global__ void get_tiles_offsets(
} }
float scale_distort00 = rD2rND + ri* drD2rND_dri; float scale_distort00 = rD2rND + ri* drD2rND_dri;
float scale_distort11 = rD2rND; float scale_distort11 = rD2rND;
// float rot2Xdd1[2][2];
// rot2Xdd1[0][0] = c_dist * dd1[0][0] + s_dist * dd1[1][0];
// rot2Xdd1[0][1] = c_dist * dd1[0][1] + s_dist * dd1[1][1];
// rot2Xdd1[1][0] = -s_dist * dd1[0][0] + c_dist * dd1[1][0];
// rot2Xdd1[1][1] = -s_dist * dd1[0][1] + c_dist * dd1[1][1];
float scale_distortXrot2Xdd1[2][2]; float scale_distortXrot2Xdd1[2][2];
scale_distortXrot2Xdd1[0][0] = ( c_dist * dd1[0][0] + s_dist * dd1[1][0]) * scale_distort00; scale_distortXrot2Xdd1[0][0] = ( c_dist * dd1[0][0] + s_dist * dd1[1][0]) * scale_distort00;
scale_distortXrot2Xdd1[0][1] = ( c_dist * dd1[0][1] + s_dist * dd1[1][1]) * scale_distort00; scale_distortXrot2Xdd1[0][1] = ( c_dist * dd1[0][1] + s_dist * dd1[1][1]) * scale_distort00;
...@@ -651,6 +629,7 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -651,6 +629,7 @@ extern "C" __global__ void get_tiles_offsets(
// float imu_move[3]; // dx/dt, dy/dt, dz/dt 16..19 geometry_correction.imu_move // float imu_move[3]; // dx/dt, dy/dt, dz/dt 16..19 geometry_correction.imu_move
// ERS linear does not yet use per-port rotations, probably not needed // ERS linear does not yet use per-port rotations, probably not needed
if (imu_exists){ if (imu_exists){
/*
float delta_t = disp_dist[2] * disparity * geometry_correction.line_time; // positive for top cameras, negative - for bottom //disp_dist[2]=dd2.get(1, 0) float delta_t = disp_dist[2] * disparity * geometry_correction.line_time; // positive for top cameras, negative - for bottom //disp_dist[2]=dd2.get(1, 0)
float ers_Xci = delta_t * ( float ers_Xci = delta_t * (
dpXci_dtilt * extrinsic_corr.imu_rot[0] + dpXci_dtilt * extrinsic_corr.imu_rot[0] +
...@@ -660,9 +639,22 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -660,9 +639,22 @@ extern "C" __global__ void get_tiles_offsets(
dpYci_dtilt * extrinsic_corr.imu_rot[0] + dpYci_dtilt * extrinsic_corr.imu_rot[0] +
dpYci_dazimuth * extrinsic_corr.imu_rot[1] + dpYci_dazimuth * extrinsic_corr.imu_rot[1] +
dpYci_droll * extrinsic_corr.imu_rot[2]); dpYci_droll * extrinsic_corr.imu_rot[2]);
#ifdef DEBUG210 */
float ers_x =
dpXci_dtilt * extrinsic_corr.imu_rot[0] +
dpXci_dazimuth * extrinsic_corr.imu_rot[1] +
dpXci_droll * extrinsic_corr.imu_rot[2];
float ers_y =
dpYci_dtilt * extrinsic_corr.imu_rot[0] +
dpYci_dazimuth * extrinsic_corr.imu_rot[1] +
dpYci_droll * extrinsic_corr.imu_rot[2];
#ifdef DEBUG21
if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){ if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){
printf("delta_t = %f, ers_Xci = %f, ers_Yci = %f\n", delta_t, ers_Xci, ers_Yci); // printf("delta_t = %f, ers_Xci = %f, ers_Yci = %f\n", delta_t, ers_Xci, ers_Yci);
printf("ers_x = %f, ers_y = %f\n", ers_x, ers_y);
} }
__syncthreads();// __syncwarp(); __syncthreads();// __syncwarp();
#endif // DEBUG21 #endif // DEBUG21
...@@ -674,22 +666,30 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -674,22 +666,30 @@ extern "C" __global__ void get_tiles_offsets(
dpXci_pYci_imu_lin[1][1] = wdisparity / k; // dpy/ dworld_Y dpXci_pYci_imu_lin[1][1] = wdisparity / k; // dpy/ dworld_Y
dpXci_pYci_imu_lin[0][2] = (xyz[0] / k) * dwdisp_dz; // dpx/ dworld_Z dpXci_pYci_imu_lin[0][2] = (xyz[0] / k) * dwdisp_dz; // dpx/ dworld_Z
dpXci_pYci_imu_lin[1][2] = (xyz[1] / k) * dwdisp_dz; // dpy/ dworld_Z dpXci_pYci_imu_lin[1][2] = (xyz[1] / k) * dwdisp_dz; // dpy/ dworld_Z
/*
ers_Xci += delta_t* ( ers_Xci += delta_t* (
dpXci_pYci_imu_lin[0][0] * extrinsic_corr.imu_move[0] + dpXci_pYci_imu_lin[0][0] * extrinsic_corr.imu_move[0] +
dpXci_pYci_imu_lin[0][2] * extrinsic_corr.imu_move[2]); dpXci_pYci_imu_lin[0][2] * extrinsic_corr.imu_move[2]);
ers_Yci += delta_t* ( ers_Yci += delta_t* (
dpXci_pYci_imu_lin[1][1] * extrinsic_corr.imu_move[1] + dpXci_pYci_imu_lin[1][1] * extrinsic_corr.imu_move[1] +
dpXci_pYci_imu_lin[1][2] * extrinsic_corr.imu_move[2]); dpXci_pYci_imu_lin[1][2] * extrinsic_corr.imu_move[2]);
pXY[0] += ers_Xci * rD2rND; // added correction to pixel X */
pXY[1] += ers_Yci * rD2rND; // added correction to pixel Y ers_x += dpXci_pYci_imu_lin[0][0] * extrinsic_corr.imu_move[0] +
dpXci_pYci_imu_lin[0][2] * extrinsic_corr.imu_move[2];
ers_y += dpXci_pYci_imu_lin[1][1] * extrinsic_corr.imu_move[1] +
dpXci_pYci_imu_lin[1][2] * extrinsic_corr.imu_move[2];
float delta_t = (pY_offset/ (1.0 - geometry_correction.line_time * ers_y)) * geometry_correction.line_time; // positive for top cameras, negative - for bottom //disp_dist[2]=dd2.get(1, 0)
#ifdef DEBUG210 pXY[0] += delta_t * ers_x * rD2rND; // added correction to pixel X
pXY[1] += delta_t * ers_y * rD2rND; // added correction to pixel Y
#ifdef DEBUG21
if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){ if ((ncam == DBG_CAM) && (task_num == DBG_TILE)){
printf("k = %f, wdisparity = %f, dwdisp_dz = %f\n", k, wdisparity, dwdisp_dz); printf("k = %f, wdisparity = %f, dwdisp_dz = %f\n", k, wdisparity, dwdisp_dz);
printf("dpXci_pYci_imu_lin[0][0] = %f, dpXci_pYci_imu_lin[0][2] = %f\n", dpXci_pYci_imu_lin[0][0],dpXci_pYci_imu_lin[0][2]); printf("dpXci_pYci_imu_lin[0][0] = %f, dpXci_pYci_imu_lin[0][2] = %f\n", dpXci_pYci_imu_lin[0][0],dpXci_pYci_imu_lin[0][2]);
printf("dpXci_pYci_imu_lin[1][1] = %f, dpXci_pYci_imu_lin[1][2] = %f\n", dpXci_pYci_imu_lin[1][1],dpXci_pYci_imu_lin[1][2]); printf("dpXci_pYci_imu_lin[1][1] = %f, dpXci_pYci_imu_lin[1][2] = %f\n", dpXci_pYci_imu_lin[1][1],dpXci_pYci_imu_lin[1][2]);
printf("delta_t = %f, ers_Xci = %f, ers_Yci = %f\n", delta_t, ers_Xci, ers_Yci); printf("delta_t = %f, ers_x = %f, ers_y = %f\n", delta_t, ers_x, ers_y);
printf("pXY[0] = %f, pXY[1] = %f\n", pXY[0], pXY[1]); // OK printf("pXY[0] = %f, pXY[1] = %f\n", pXY[0], pXY[1]); // OK
} }
__syncthreads();// __syncwarp(); __syncthreads();// __syncwarp();
...@@ -703,6 +703,7 @@ extern "C" __global__ void get_tiles_offsets( ...@@ -703,6 +703,7 @@ extern "C" __global__ void get_tiles_offsets(
} }
extern "C" __global__ void calcReverseDistortionTable( extern "C" __global__ void calcReverseDistortionTable(
struct gc * geometry_correction, struct gc * geometry_correction,
float * rByRDist) float * rByRDist)
...@@ -841,6 +842,7 @@ __device__ void printGeometryCorrection(struct gc * g){ ...@@ -841,6 +842,7 @@ __device__ void printGeometryCorrection(struct gc * g){
printf("%22s: %f\n","cameraRadius", g->cameraRadius); printf("%22s: %f\n","cameraRadius", g->cameraRadius);
printf("%22s: %f\n","disparityRadius", g->disparityRadius); printf("%22s: %f\n","disparityRadius", g->disparityRadius);
printf("%22s: %f, %f, %f, %f \n","woi_tops", g->woi_tops[0], g->woi_tops[1], g->woi_tops[2], g->woi_tops[3]);
#endif //ifndef JCUDA #endif //ifndef JCUDA
} }
......
...@@ -138,6 +138,7 @@ struct gc { ...@@ -138,6 +138,7 @@ struct gc {
// only used for the multi-quad systems // only used for the multi-quad systems
float cameraRadius; // =0; // average distance from the "mass center" of the sensors to the sensors float cameraRadius; // =0; // average distance from the "mass center" of the sensors to the sensors
float disparityRadius; // =150.0; // distance between cameras to normalize disparity units to. sqrt(2)*disparityRadius for quad float disparityRadius; // =150.0; // distance between cameras to normalize disparity units to. sqrt(2)*disparityRadius for quad
float woi_tops [NUM_CAMS]; // used to calculate scanline timing
}; };
#define RAD_COEFF_LEN 7 #define RAD_COEFF_LEN 7
extern "C" __global__ void get_tiles_offsets( extern "C" __global__ void get_tiles_offsets(
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment