1 #include <iostream>
2 #include <fstream>
3 #include <string>
4 #include <sstream>
5 #include <iomanip>
6 #include <stdexcept>
7 #include <opencv2/core/utility.hpp>
8 #include "opencv2/cudaobjdetect.hpp"
9 #include "opencv2/highgui.hpp"
10 #include "opencv2/objdetect.hpp"
11 #include "opencv2/imgproc.hpp"
12
13 using namespace std;
14 using namespace cv;
15
16 bool help_showed = false;
17
18 class Args
19 {
20 public:
21 Args();
22 static Args read(int argc, char** argv);
23
24 string src;
25 bool src_is_folder;
26 bool src_is_video;
27 bool src_is_camera;
28 int camera_id;
29
30 bool svm_load;
31 string svm;
32
33 bool write_video;
34 string dst_video;
35 double dst_video_fps;
36
37 bool make_gray;
38
39 bool resize_src;
40 int width, height;
41
42 double scale;
43 int nlevels;
44 int gr_threshold;
45
46 double hit_threshold;
47 bool hit_threshold_auto;
48
49 int win_width;
50 int win_stride_width, win_stride_height;
51 int block_width;
52 int block_stride_width, block_stride_height;
53 int cell_width;
54 int nbins;
55
56 bool gamma_corr;
57 };
58
59
60 class App
61 {
62 public:
63 App(const Args& s);
64 void run();
65
66 void handleKey(char key);
67
68 void hogWorkBegin();
69 void hogWorkEnd();
70 string hogWorkFps() const;
71
72 void workBegin();
73 void workEnd();
74 string workFps() const;
75
76 string message() const;
77
78 private:
79 App operator=(App&);
80
81 Args args;
82 bool running;
83
84 bool use_gpu;
85 bool make_gray;
86 double scale;
87 int gr_threshold;
88 int nlevels;
89 double hit_threshold;
90 bool gamma_corr;
91
92 int64 hog_work_begin;
93 double hog_work_fps;
94
95 int64 work_begin;
96 double work_fps;
97 };
98
printHelp()99 static void printHelp()
100 {
101 cout << "Histogram of Oriented Gradients descriptor and detector sample.\n"
102 << "\nUsage: hog\n"
103 << " (<image>|--video <vide>|--camera <camera_id>) # frames source\n"
104 << " or"
105 << " (--folder <folder_path>) # load images from folder\n"
106 << " [--svm <file> # load svm file"
107 << " [--make_gray <true/false>] # convert image to gray one or not\n"
108 << " [--resize_src <true/false>] # do resize of the source image or not\n"
109 << " [--width <int>] # resized image width\n"
110 << " [--height <int>] # resized image height\n"
111 << " [--hit_threshold <double>] # classifying plane distance threshold (0.0 usually)\n"
112 << " [--scale <double>] # HOG window scale factor\n"
113 << " [--nlevels <int>] # max number of HOG window scales\n"
114 << " [--win_width <int>] # width of the window\n"
115 << " [--win_stride_width <int>] # distance by OX axis between neighbour wins\n"
116 << " [--win_stride_height <int>] # distance by OY axis between neighbour wins\n"
117 << " [--block_width <int>] # width of the block\n"
118 << " [--block_stride_width <int>] # distance by 0X axis between neighbour blocks\n"
119 << " [--block_stride_height <int>] # distance by 0Y axis between neighbour blocks\n"
120 << " [--cell_width <int>] # width of the cell\n"
121 << " [--nbins <int>] # number of bins\n"
122 << " [--gr_threshold <int>] # merging similar rects constant\n"
123 << " [--gamma_correct <int>] # do gamma correction or not\n"
124 << " [--write_video <bool>] # write video or not\n"
125 << " [--dst_video <path>] # output video path\n"
126 << " [--dst_video_fps <double>] # output video fps\n";
127 help_showed = true;
128 }
129
main(int argc,char ** argv)130 int main(int argc, char** argv)
131 {
132 try
133 {
134 Args args;
135 if (argc < 2)
136 {
137 printHelp();
138 args.camera_id = 0;
139 args.src_is_camera = true;
140 }
141 else
142 {
143 args = Args::read(argc, argv);
144 if (help_showed)
145 return -1;
146 }
147 App app(args);
148 app.run();
149 }
150 catch (const Exception& e) { return cout << "error: " << e.what() << endl, 1; }
151 catch (const exception& e) { return cout << "error: " << e.what() << endl, 1; }
152 catch(...) { return cout << "unknown exception" << endl, 1; }
153 return 0;
154 }
155
156
Args()157 Args::Args()
158 {
159 src_is_video = false;
160 src_is_camera = false;
161 src_is_folder = false;
162 svm_load = false;
163 camera_id = 0;
164
165 write_video = false;
166 dst_video_fps = 24.;
167
168 make_gray = false;
169
170 resize_src = false;
171 width = 640;
172 height = 480;
173
174 scale = 1.05;
175 nlevels = 13;
176 gr_threshold = 8;
177 hit_threshold = 1.4;
178 hit_threshold_auto = true;
179
180 win_width = 48;
181 win_stride_width = 8;
182 win_stride_height = 8;
183 block_width = 16;
184 block_stride_width = 8;
185 block_stride_height = 8;
186 cell_width = 8;
187 nbins = 9;
188
189 gamma_corr = true;
190 }
191
192
read(int argc,char ** argv)193 Args Args::read(int argc, char** argv)
194 {
195 Args args;
196 for (int i = 1; i < argc; i++)
197 {
198 if (string(argv[i]) == "--make_gray") args.make_gray = (string(argv[++i]) == "true");
199 else if (string(argv[i]) == "--resize_src") args.resize_src = (string(argv[++i]) == "true");
200 else if (string(argv[i]) == "--width") args.width = atoi(argv[++i]);
201 else if (string(argv[i]) == "--height") args.height = atoi(argv[++i]);
202 else if (string(argv[i]) == "--hit_threshold")
203 {
204 args.hit_threshold = atof(argv[++i]);
205 args.hit_threshold_auto = false;
206 }
207 else if (string(argv[i]) == "--scale") args.scale = atof(argv[++i]);
208 else if (string(argv[i]) == "--nlevels") args.nlevels = atoi(argv[++i]);
209 else if (string(argv[i]) == "--win_width") args.win_width = atoi(argv[++i]);
210 else if (string(argv[i]) == "--win_stride_width") args.win_stride_width = atoi(argv[++i]);
211 else if (string(argv[i]) == "--win_stride_height") args.win_stride_height = atoi(argv[++i]);
212 else if (string(argv[i]) == "--block_width") args.block_width = atoi(argv[++i]);
213 else if (string(argv[i]) == "--block_stride_width") args.block_stride_width = atoi(argv[++i]);
214 else if (string(argv[i]) == "--block_stride_height") args.block_stride_height = atoi(argv[++i]);
215 else if (string(argv[i]) == "--cell_width") args.cell_width = atoi(argv[++i]);
216 else if (string(argv[i]) == "--nbins") args.nbins = atoi(argv[++i]);
217 else if (string(argv[i]) == "--gr_threshold") args.gr_threshold = atoi(argv[++i]);
218 else if (string(argv[i]) == "--gamma_correct") args.gamma_corr = (string(argv[++i]) == "true");
219 else if (string(argv[i]) == "--write_video") args.write_video = (string(argv[++i]) == "true");
220 else if (string(argv[i]) == "--dst_video") args.dst_video = argv[++i];
221 else if (string(argv[i]) == "--dst_video_fps") args.dst_video_fps = atof(argv[++i]);
222 else if (string(argv[i]) == "--help") printHelp();
223 else if (string(argv[i]) == "--video") { args.src = argv[++i]; args.src_is_video = true; }
224 else if (string(argv[i]) == "--camera") { args.camera_id = atoi(argv[++i]); args.src_is_camera = true; }
225 else if (string(argv[i]) == "--folder") { args.src = argv[++i]; args.src_is_folder = true;}
226 else if (string(argv[i]) == "--svm") { args.svm = argv[++i]; args.svm_load = true;}
227 else if (args.src.empty()) args.src = argv[i];
228 else throw runtime_error((string("unknown key: ") + argv[i]));
229 }
230 return args;
231 }
232
233
App(const Args & s)234 App::App(const Args& s)
235 {
236 cv::cuda::printShortCudaDeviceInfo(cv::cuda::getDevice());
237
238 args = s;
239 cout << "\nControls:\n"
240 << "\tESC - exit\n"
241 << "\tm - change mode GPU <-> CPU\n"
242 << "\tg - convert image to gray or not\n"
243 << "\t1/q - increase/decrease HOG scale\n"
244 << "\t2/w - increase/decrease levels count\n"
245 << "\t3/e - increase/decrease HOG group threshold\n"
246 << "\t4/r - increase/decrease hit threshold\n"
247 << endl;
248
249 use_gpu = true;
250 make_gray = args.make_gray;
251 scale = args.scale;
252 gr_threshold = args.gr_threshold;
253 nlevels = args.nlevels;
254
255 if (args.hit_threshold_auto)
256 args.hit_threshold = args.win_width == 48 ? 1.4 : 0.;
257 hit_threshold = args.hit_threshold;
258
259 gamma_corr = args.gamma_corr;
260
261 cout << "Scale: " << scale << endl;
262 if (args.resize_src)
263 cout << "Resized source: (" << args.width << ", " << args.height << ")\n";
264 cout << "Group threshold: " << gr_threshold << endl;
265 cout << "Levels number: " << nlevels << endl;
266 cout << "Win size: (" << args.win_width << ", " << args.win_width*2 << ")\n";
267 cout << "Win stride: (" << args.win_stride_width << ", " << args.win_stride_height << ")\n";
268 cout << "Block size: (" << args.block_width << ", " << args.block_width << ")\n";
269 cout << "Block stride: (" << args.block_stride_width << ", " << args.block_stride_height << ")\n";
270 cout << "Cell size: (" << args.cell_width << ", " << args.cell_width << ")\n";
271 cout << "Bins number: " << args.nbins << endl;
272 cout << "Hit threshold: " << hit_threshold << endl;
273 cout << "Gamma correction: " << gamma_corr << endl;
274 cout << endl;
275 }
276
277
run()278 void App::run()
279 {
280 running = true;
281 cv::VideoWriter video_writer;
282
283 Size win_stride(args.win_stride_width, args.win_stride_height);
284 Size win_size(args.win_width, args.win_width * 2);
285 Size block_size(args.block_width, args.block_width);
286 Size block_stride(args.block_stride_width, args.block_stride_height);
287 Size cell_size(args.cell_width, args.cell_width);
288
289 cv::Ptr<cv::cuda::HOG> gpu_hog = cv::cuda::HOG::create(win_size, block_size, block_stride, cell_size, args.nbins);
290 cv::HOGDescriptor cpu_hog(win_size, block_size, block_stride, cell_size, args.nbins);
291
292 if(args.svm_load) {
293 std::vector<float> svm_model;
294 const std::string model_file_name = args.svm;
295 FileStorage ifs(model_file_name, FileStorage::READ);
296 if (ifs.isOpened()) {
297 ifs["svm_detector"] >> svm_model;
298 } else {
299 const std::string what =
300 "could not load model for hog classifier from file: "
301 + model_file_name;
302 throw std::runtime_error(what);
303 }
304
305 // check if the variables are initialized
306 if (svm_model.empty()) {
307 const std::string what =
308 "HoG classifier: svm model could not be loaded from file"
309 + model_file_name;
310 throw std::runtime_error(what);
311 }
312
313 gpu_hog->setSVMDetector(svm_model);
314 cpu_hog.setSVMDetector(svm_model);
315 } else {
316 // Create HOG descriptors and detectors here
317 Mat detector = gpu_hog->getDefaultPeopleDetector();
318
319 gpu_hog->setSVMDetector(detector);
320 cpu_hog.setSVMDetector(detector);
321 }
322
323 cout << "gpusvmDescriptorSize : " << gpu_hog->getDescriptorSize()
324 << endl;
325 cout << "cpusvmDescriptorSize : " << cpu_hog.getDescriptorSize()
326 << endl;
327
328 while (running)
329 {
330 VideoCapture vc;
331 Mat frame;
332 vector<String> filenames;
333
334 unsigned int count = 1;
335
336 if (args.src_is_video)
337 {
338 vc.open(args.src.c_str());
339 if (!vc.isOpened())
340 throw runtime_error(string("can't open video file: " + args.src));
341 vc >> frame;
342 }
343 else if (args.src_is_folder) {
344 String folder = args.src;
345 cout << folder << endl;
346 glob(folder, filenames);
347 frame = imread(filenames[count]); // 0 --> .gitignore
348 if (!frame.data)
349 cerr << "Problem loading image from folder!!!" << endl;
350 }
351 else if (args.src_is_camera)
352 {
353 vc.open(args.camera_id);
354 if (!vc.isOpened())
355 {
356 stringstream msg;
357 msg << "can't open camera: " << args.camera_id;
358 throw runtime_error(msg.str());
359 }
360 vc >> frame;
361 }
362 else
363 {
364 frame = imread(args.src);
365 if (frame.empty())
366 throw runtime_error(string("can't open image file: " + args.src));
367 }
368
369 Mat img_aux, img, img_to_show;
370 cuda::GpuMat gpu_img;
371
372 // Iterate over all frames
373 while (running && !frame.empty())
374 {
375 workBegin();
376
377 // Change format of the image
378 if (make_gray) cvtColor(frame, img_aux, COLOR_BGR2GRAY);
379 else if (use_gpu) cvtColor(frame, img_aux, COLOR_BGR2BGRA);
380 else frame.copyTo(img_aux);
381
382 // Resize image
383 if (args.resize_src) resize(img_aux, img, Size(args.width, args.height));
384 else img = img_aux;
385 img_to_show = img;
386
387 vector<Rect> found;
388
389 // Perform HOG classification
390 hogWorkBegin();
391 if (use_gpu)
392 {
393 gpu_img.upload(img);
394 gpu_hog->setNumLevels(nlevels);
395 gpu_hog->setHitThreshold(hit_threshold);
396 gpu_hog->setWinStride(win_stride);
397 gpu_hog->setScaleFactor(scale);
398 gpu_hog->setGroupThreshold(gr_threshold);
399 gpu_hog->detectMultiScale(gpu_img, found);
400 }
401 else
402 {
403 cpu_hog.nlevels = nlevels;
404 cpu_hog.detectMultiScale(img, found, hit_threshold, win_stride,
405 Size(0, 0), scale, gr_threshold);
406 }
407 hogWorkEnd();
408
409 // Draw positive classified windows
410 for (size_t i = 0; i < found.size(); i++)
411 {
412 Rect r = found[i];
413 rectangle(img_to_show, r.tl(), r.br(), Scalar(0, 255, 0), 3);
414 }
415
416 if (use_gpu)
417 putText(img_to_show, "Mode: GPU", Point(5, 25), FONT_HERSHEY_SIMPLEX, 1., Scalar(255, 100, 0), 2);
418 else
419 putText(img_to_show, "Mode: CPU", Point(5, 25), FONT_HERSHEY_SIMPLEX, 1., Scalar(255, 100, 0), 2);
420 putText(img_to_show, "FPS HOG: " + hogWorkFps(), Point(5, 65), FONT_HERSHEY_SIMPLEX, 1., Scalar(255, 100, 0), 2);
421 putText(img_to_show, "FPS total: " + workFps(), Point(5, 105), FONT_HERSHEY_SIMPLEX, 1., Scalar(255, 100, 0), 2);
422 imshow("opencv_gpu_hog", img_to_show);
423
424 if (args.src_is_video || args.src_is_camera) vc >> frame;
425 if (args.src_is_folder) {
426 count++;
427 if (count < filenames.size()) {
428 frame = imread(filenames[count]);
429 } else {
430 Mat empty;
431 frame = empty;
432 }
433 }
434
435 workEnd();
436
437 if (args.write_video)
438 {
439 if (!video_writer.isOpened())
440 {
441 video_writer.open(args.dst_video, VideoWriter::fourcc('x','v','i','d'), args.dst_video_fps,
442 img_to_show.size(), true);
443 if (!video_writer.isOpened())
444 throw std::runtime_error("can't create video writer");
445 }
446
447 if (make_gray) cvtColor(img_to_show, img, COLOR_GRAY2BGR);
448 else cvtColor(img_to_show, img, COLOR_BGRA2BGR);
449
450 video_writer << img;
451 }
452
453 handleKey((char)waitKey(3));
454 }
455 }
456 }
457
458
handleKey(char key)459 void App::handleKey(char key)
460 {
461 switch (key)
462 {
463 case 27:
464 running = false;
465 break;
466 case 'm':
467 case 'M':
468 use_gpu = !use_gpu;
469 cout << "Switched to " << (use_gpu ? "CUDA" : "CPU") << " mode\n";
470 break;
471 case 'g':
472 case 'G':
473 make_gray = !make_gray;
474 cout << "Convert image to gray: " << (make_gray ? "YES" : "NO") << endl;
475 break;
476 case '1':
477 scale *= 1.05;
478 cout << "Scale: " << scale << endl;
479 break;
480 case 'q':
481 case 'Q':
482 scale /= 1.05;
483 cout << "Scale: " << scale << endl;
484 break;
485 case '2':
486 nlevels++;
487 cout << "Levels number: " << nlevels << endl;
488 break;
489 case 'w':
490 case 'W':
491 nlevels = max(nlevels - 1, 1);
492 cout << "Levels number: " << nlevels << endl;
493 break;
494 case '3':
495 gr_threshold++;
496 cout << "Group threshold: " << gr_threshold << endl;
497 break;
498 case 'e':
499 case 'E':
500 gr_threshold = max(0, gr_threshold - 1);
501 cout << "Group threshold: " << gr_threshold << endl;
502 break;
503 case '4':
504 hit_threshold+=0.25;
505 cout << "Hit threshold: " << hit_threshold << endl;
506 break;
507 case 'r':
508 case 'R':
509 hit_threshold = max(0.0, hit_threshold - 0.25);
510 cout << "Hit threshold: " << hit_threshold << endl;
511 break;
512 case 'c':
513 case 'C':
514 gamma_corr = !gamma_corr;
515 cout << "Gamma correction: " << gamma_corr << endl;
516 break;
517 }
518 }
519
520
hogWorkBegin()521 inline void App::hogWorkBegin() { hog_work_begin = getTickCount(); }
522
hogWorkEnd()523 inline void App::hogWorkEnd()
524 {
525 int64 delta = getTickCount() - hog_work_begin;
526 double freq = getTickFrequency();
527 hog_work_fps = freq / delta;
528 }
529
hogWorkFps() const530 inline string App::hogWorkFps() const
531 {
532 stringstream ss;
533 ss << hog_work_fps;
534 return ss.str();
535 }
536
537
workBegin()538 inline void App::workBegin() { work_begin = getTickCount(); }
539
workEnd()540 inline void App::workEnd()
541 {
542 int64 delta = getTickCount() - work_begin;
543 double freq = getTickFrequency();
544 work_fps = freq / delta;
545 }
546
workFps() const547 inline string App::workFps() const
548 {
549 stringstream ss;
550 ss << work_fps;
551 return ss.str();
552 }
553