visioncpp/examples/optical_flow_LK.cpp at master · codeplaysoftware/visioncpp · GitHub

This repository was archived by the owner on Apr 29, 2024. It is now read-only.

Repository navigation

History

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

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

// This file is part of VisionCPP, a lightweight C++ template library

// for computer vision and image processing.

// Contact: visioncpp@codeplay.com

// Licensed under the Apache License, Version 2.0 (the "License");

// you may not use this file except in compliance with the License.

// You may obtain a copy of the License at

// http://www.apache.org/licenses/LICENSE-2.0

// Unless required by applicable law or agreed to in writing, software

// distributed under the License is distributed on an "AS IS" BASIS,

// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.

// See the License for the specific language governing permissions and

// limitations under the License.

/// \file optical_flow_LK.cc

///

/// -------Example--------

/// Lucas-Kanade Optical Flow

///

/// \brief This example implements the Lucas-Kanade Optical Flow

/// This optical flow needs to solve the following equation

/// | v | ( | Dx*Dy Dy^2 | ) | -Dy*Dt |

/// where w is the 3x3 window of ones

///

/// doing algebraic manipulation

/// n = sum(Dx^2) * sum (Dy^2) - (sum (Dx * Dy)^2)

/// u = (-sum (Dy^2) * sum (Dx*Dt) + sum (Dx*Dy) * sum (Dy*Dt)) / n

/// v = ( sum (Dx*Dt) * sum (Dx*Dy) - sum (Dx^2) * sum (Dy*Dt)) / n

// include OpenCV for camera display

#include <opencv2/opencv.hpp>

// include VisionCPP

#include <visioncpp.hpp>

// operator that transforms uv coordinates into polar coordinates

// it was created for displaying the optical flow in RGB

struct OP_UVtoPolar {

visioncpp::pixel::F32C3 operator()(visioncpp::pixel::F32C2 t) {

float intensity = cl::sycl::clamp(

cl::sycl::sqrt(t[0] * t[0] + t[1] * t[1]) / 2.0f, 0.0f, 1.0f);

float angle = cl::sycl::atan2(t[1], t[0]) / (2.0f * M_PI);

float chn = 1.0f;

return visioncpp::pixel::F32C3(angle, chn, intensity);

}

};

// main program

int main(int argc, char **argv) {

// open video or camera

cv::VideoCapture cap;

if (argc == 1) {

cap.open(0);

std::cout << "To use video" << std::endl;

std::cout << "example>: ./example path/to/video.avi" << std::endl;

} else if (argc > 1) {

cap.open(argv[1]);

}

// check if we succeeded

if (!cap.isOpened()) {

std::cout << "Opening Camera/Video Failed." << std::endl;

return -1;

}

// selecting device using sycl as backend

auto dev = visioncpp::make_device<visioncpp::backend::sycl,

visioncpp::device::cpu>();

// defining the image size constants

constexpr size_t COLS = 640;

constexpr size_t ROWS = 480;

constexpr size_t SM = 16;

// defining sum mask

constexpr size_t N = 15;

constexpr size_t NxN = N * N;

float sum_mask[NxN];

for (size_t i = 0; i < NxN; i++) {

sum_mask[i] = 1.0f;

}

// defining derivatives

float prewitt_x[9] = {-1.0f, 0.0f, 1.0f, -2.0f, 0.0f,

2.0f, -1.0f, 0.0f, 1.0f};

float prewitt_y[9] = {-1.0f, -2.0f, -1.0f, 0.0, 0.0f, 0.0f, 1.0f, 2.0f, 1.0f};

// initializing pointers which will store the final results

std::shared_ptr<uchar> rgbFlow(new uchar[COLS * ROWS * 3],

[](uchar *dataMem) { delete[] dataMem; });

// initializing matrices which will store current and previous frame

cv::Mat current;

cv::Mat previous;

// init matrix which displays in RGB the (u,v) matrix

cv::Mat rgbflow_mat(ROWS, COLS, CV_8UC3, rgbFlow.get());

// flag to control if it is the first frame

bool first = true;

for (;;) {

// don't capture the first frame.

// it needs the previous frame to compute optical flow

if (first) {

// read frame

cap.read(current);

// check if image was loaded

if (!current.data) {

break;

}

// resize image to the desirable size

cv::resize(current, current, cv::Size(COLS, ROWS), 0, 0, cv::INTER_CUBIC);

// set flag to false

first = false;

continue;

}

// Starting building the tree (use {} during the creation of the tree)

{

// copy previous frame

previous = current.clone();

// read frame

cap.read(current);

// check if image was loaded

if (!current.data) {

break;

}

// resize image to the desirable size

cv::resize(current, current, cv::Size(COLS, ROWS), 0, 0, cv::INTER_CUBIC);

// init terminal nodes

auto out =

visioncpp::terminal<visioncpp::pixel::U8C3, COLS, ROWS,

visioncpp::memory_type::Buffer2D>(rgbFlow.get());

auto in =

visioncpp::terminal<visioncpp::pixel::U8C3, COLS, ROWS,

visioncpp::memory_type::Buffer2D>(current.data);

auto prev =

visioncpp::terminal<visioncpp::pixel::U8C3, COLS, ROWS,

visioncpp::memory_type::Buffer2D>(previous.data);

// convert unsigned char to float

auto ifrgb = visioncpp::point_operation<visioncpp::OP_U8C3ToF32C3>(in);

// convert to grey scale previous and current frame

auto ifgrey = visioncpp::point_operation<visioncpp::OP_RGBToGREY>(ifrgb);

// convert unsigned char to float

auto pfrgb = visioncpp::point_operation<visioncpp::OP_U8C3ToF32C3>(prev);

// convert to grey scale previous and current frame

auto pfgrey = visioncpp::point_operation<visioncpp::OP_RGBToGREY>(pfrgb);

// apply derivatives in x, y and t (in current and previous images)

auto px_filter =

visioncpp::terminal<float, 3, 3, visioncpp::memory_type::Buffer2D,

visioncpp::scope::Constant>(prewitt_x);

auto py_filter =

visioncpp::terminal<float, 3, 3, visioncpp::memory_type::Buffer2D,

visioncpp::scope::Constant>(prewitt_y);

// break tree before convolution

auto iifgrey =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(ifgrey);

auto ppfgrey =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(pfgrey);

auto ipx = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

iifgrey, px_filter);

auto ipy = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

iifgrey, py_filter);

auto ppx = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppfgrey, px_filter);

auto ppy = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppfgrey, py_filter);

// break tree after convolution

auto iipx =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(ipx);

auto iipy =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(ipy);

auto pppx =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(ppx);

auto pppy =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(ppy);

// sum derivatives of current and previous frames

auto px = visioncpp::point_operation<visioncpp::OP_Add>(iipx, pppx);

auto py = visioncpp::point_operation<visioncpp::OP_Add>(iipy, pppy);

auto pt = visioncpp::point_operation<visioncpp::OP_Sub>(ifgrey, pfgrey);

auto px2 = visioncpp::point_operation<visioncpp::OP_PowerOf2>(px);

auto py2 = visioncpp::point_operation<visioncpp::OP_PowerOf2>(py);

auto pxy = visioncpp::point_operation<visioncpp::OP_Mul>(px, py);

auto pxt = visioncpp::point_operation<visioncpp::OP_Mul>(px, pt);

auto pyt = visioncpp::point_operation<visioncpp::OP_Mul>(py, pt);

// break tree before convolution

auto ppx2 =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(px2);

auto ppy2 =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(py2);

auto ppxy =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(pxy);

auto ppxt =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(pxt);

auto ppyt =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(pyt);

// Sum neighbours

auto sum_mask_node =

visioncpp::terminal<float, N, N, visioncpp::memory_type::Buffer2D,

visioncpp::scope::Constant>(sum_mask);

auto sumpx2 = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppx2, sum_mask_node);

auto sumpy2 = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppy2, sum_mask_node);

auto sumpxy = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppxy, sum_mask_node);

auto sumpxt = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppxt, sum_mask_node);

auto sumpyt = visioncpp::neighbour_operation<visioncpp::OP_Filter2D>(

ppyt, sum_mask_node);

// break tree after convolution

auto ksumpx2 =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(sumpx2);

auto ksumpy2 =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(sumpy2);

auto ksumpxy =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(sumpxy);

auto ksumpxt =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(sumpxt);

auto ksumpyt =

visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(sumpyt);

// now we can finally apply the formulas

// u = (-sum (Dy^2) * sum (Dx*Dt) + sum (Dx*Dy) * sum (Dy*Dt)) / n

// v = ( sum (Dx*Dt) * sum (Dx*Dy) - sum (Dx^2) * sum (Dy*Dt)) / n

// n = sum(Dx^2) * sum (Dy^2) - (sum (Dx * Dy)^2)

auto px2py2 =

visioncpp::point_operation<visioncpp::OP_Mul>(ksumpx2, ksumpy2);

auto pxy2 = visioncpp::point_operation<visioncpp::OP_PowerOf2>(ksumpxy);

auto px2py2_Sub_pxy2 =

visioncpp::point_operation<visioncpp::OP_Sub>(px2py2, pxy2);

auto norm = visioncpp::schedule<visioncpp::policy::Fuse, SM, SM, SM, SM>(

px2py2_Sub_pxy2);

// calculate V

// v = ( sum (Dx*Dt) * sum (Dx*Dy) - sum (Dx^2) * sum (Dy*Dt)) / n

auto pxtpxy =

visioncpp::point_operation<visioncpp::OP_Mul>(ksumpxt, ksumpxy);

auto px2pyt =

visioncpp::point_operation<visioncpp::OP_Mul>(ksumpx2, ksumpyt);

auto pxtpxy_Sub_px2pyt =

visioncpp::point_operation<visioncpp::OP_Sub>(pxtpxy, px2pyt);

auto v = visioncpp::point_operation<visioncpp::OP_Div>(pxtpxy_Sub_px2pyt,

norm);

// calculate U

// u = (-sum (Dy^2) * sum (Dx*Dt) + sum (Dx*Dy) * sum (Dy*Dt)) / n

auto pxypyt =

visioncpp::point_operation<visioncpp::OP_Mul>(ksumpxy, ksumpyt);

auto py2pxt =

visioncpp::point_operation<visioncpp::OP_Mul>(ksumpy2, ksumpxt);

auto pxypft_Sub_py2pxt =

visioncpp::point_operation<visioncpp::OP_Sub>(pxypyt, py2pxt);

auto u = visioncpp::point_operation<visioncpp::OP_Div>(pxypft_Sub_py2pxt,

norm);

// assign result in one matrix with 2 channels

// variable uv contains the optical flow computed for each pixel

auto uv = visioncpp::point_operation<visioncpp::OP_Merge2Chns>(u, v);

// The next operations were created to visualize the optical flow

// convert UV into polar coordinates

auto polar = visioncpp::point_operation<OP_UVtoPolar>(uv);

// convert into RGB

auto frgb = visioncpp::point_operation<visioncpp::OP_HSVToRGB>(polar);

// convert float to char

auto urgb = visioncpp::point_operation<visioncpp::OP_F32C3ToU8C3>(frgb);

// assign the urgb to the output node

auto k = visioncpp::assign(out, urgb);

// execute

visioncpp::execute<visioncpp::policy::Fuse, SM, SM, SM, SM>(k, dev);

}

// display optical flow

cv::imshow("Reference Image", current);

cv::imshow("Optical Flow Lucas-Kanade", rgbflow_mat);

// check button pressed to exit the program

if (cv::waitKey(1) >= 0) break;

}

// release video/camera

cap.release();

return 0;

}

visioncpp/examples/optical_flow_LK.cpp at master · codeplaysoftware/visioncpp · GitHub

Repository navigation

Expand file tree

Breadcrumbs

optical_flow_LK.cpp

Latest commit

History

Breadcrumbs

optical_flow_LK.cpp

File metadata and controls