detect_manager.cpp 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270
  1. //
  2. // Created by zx on 2023/11/24.
  3. //
  4. #include "detect_manager.h"
  5. Error_manager DetectManager::Init(const google::protobuf::RepeatedPtrField<tof3dVzenseEtc> &tof_devices) {
  6. for (int device_index = 0; device_index < DeviceAzimuth::DEVICE_AZIMUTH_MAX; device_index++) {
  7. updateTrans((DeviceAzimuth)device_index, tof_devices[device_index].trans());
  8. }
  9. m_thread = new std::thread(&DetectManager::run, this);
  10. m_condit.notify_all(true);
  11. return {};
  12. }
  13. Error_manager DetectManager::Release() {
  14. stop();
  15. return {};
  16. }
  17. Error_manager DetectManager::updateTrans(const DeviceAzimuth &azimuth, const CoordinateTransformation3D &trans) {
  18. Eigen::Matrix4f rotaition = Eigen::Matrix4f::Identity();
  19. Eigen::Matrix3f rotation_matrix3 = Eigen::Matrix3f::Identity();
  20. rotation_matrix3 =
  21. Eigen::AngleAxisf(trans.yaw() * M_PI / 180.f, Eigen::Vector3f::UnitZ()) *
  22. Eigen::AngleAxisf(trans.pitch() * M_PI / 180.f, Eigen::Vector3f::UnitY()) *
  23. Eigen::AngleAxisf(trans.roll() * M_PI / 180.f, Eigen::Vector3f::UnitX());
  24. Eigen::Vector3f move;
  25. move << trans.x(), trans.y(), trans.z();
  26. rotaition.block<3, 3>(0, 0) = rotation_matrix3;
  27. rotaition.col(3) << move;
  28. LOG(INFO) << "device " << azimuth << " update rotation info: \n" << rotaition;
  29. rotation_matrix4[azimuth] = rotaition;
  30. return {};
  31. }
  32. #include "tool/time.hpp"
  33. void DetectManager::run() {
  34. LOG(INFO) << "detect thread running in " << std::this_thread::get_id();
  35. DetectResult m_detect_result_record;
  36. std::vector<pcl::PointCloud<pcl::PointXYZ>::Ptr> vct_wheel_cloud;
  37. for (int i = 0; i < DeviceAzimuth::DEVICE_AZIMUTH_MAX; i++) {
  38. vct_wheel_cloud.emplace_back(new pcl::PointCloud<pcl::PointXYZ>);
  39. }
  40. auto t_start_time = std::chrono::steady_clock::now();
  41. std::chrono::duration<double> cost = std::chrono::steady_clock::now() - t_start_time;
  42. RecordImages record_images;
  43. while (m_condit.is_alive()) {
  44. m_condit.wait();
  45. // 参数初始化
  46. cost = std::chrono::steady_clock::now() - t_start_time;
  47. std::this_thread::sleep_for(std::chrono::milliseconds(50 - std::min<int>(49, cost.count() * 1000)));
  48. t_start_time = std::chrono::steady_clock::now();
  49. if (m_condit.is_alive()) {
  50. std::vector<int> have_cloud;
  51. DetectResult detect_result;
  52. JetStream::LabelYolo seg_results;
  53. for (int i = 0; i < DeviceAzimuth::DEVICE_AZIMUTH_MAX; i++) {
  54. vct_wheel_cloud[i]->clear();
  55. }
  56. if (!LabelYoloDataBuffer::iter()->lock_exe(&LabelYoloDataBuffer::get, seg_results, 10)) {
  57. continue;
  58. }
  59. int lable = TimeTool::timeDropoutLL(TimeTool::TemporalPrecision::TemporalPrecisionMs, TimeTool::TemporalPrecision::TemporalPrecisionHour);
  60. LOG(INFO) << "after " << (lable - seg_results.label()) << "ms";
  61. // 判断识别结果数量以及是否存图
  62. int seg_num = 0;
  63. for (int device_index = 0; device_index < DeviceAzimuth::DEVICE_AZIMUTH_MAX; device_index++) {
  64. if (seg_results.boxes(device_index).confidence() > 0.7) {
  65. seg_num++;
  66. }
  67. }
  68. if (seg_num == 1 && record_images.name_1.empty()) {
  69. if (MatDataBuffer::iter()->lock_exe(&MatDataBuffer::getIr, seg_results.label(), record_images.img_1, 10)) {
  70. record_images.name_1 = TimeTool::timeDropoutStr();
  71. } else {
  72. record_images.name_1.clear();
  73. }
  74. }
  75. if (seg_num == 2 && record_images.name_2.empty()) {
  76. if (MatDataBuffer::iter()->lock_exe(&MatDataBuffer::getIr, seg_results.label(), record_images.img_2, 10)) {
  77. record_images.name_2 = TimeTool::timeDropoutStr();
  78. } else {
  79. record_images.name_2.clear();
  80. }
  81. }
  82. if (seg_num == 3 && record_images.name_3.empty()) {
  83. if (MatDataBuffer::iter()->lock_exe(&MatDataBuffer::getIr, seg_results.label(), record_images.img_3, 10)) {
  84. record_images.name_3 = TimeTool::timeDropoutStr();
  85. } else {
  86. record_images.name_3.clear();
  87. }
  88. }
  89. if (seg_num == 4 && record_images.name_4.empty()) {
  90. if (MatDataBuffer::iter()->lock_exe(&MatDataBuffer::getIr, seg_results.label(), record_images.img_4, 10)) {
  91. record_images.name_4 = TimeTool::timeDropoutStr();
  92. } else {
  93. record_images.name_4.clear();
  94. }
  95. }
  96. // 取高置信度识别结果
  97. for (int device_index = 0; device_index < DeviceAzimuth::DEVICE_AZIMUTH_MAX; device_index++) {
  98. if (seg_results.boxes(device_index).confidence() < 0.9) {
  99. DLOG(INFO) << device_index << " empty";
  100. continue;
  101. }
  102. cv::Mat pointMat;
  103. //LOG(INFO) << "b get data" << " empty";
  104. if (MatDataBuffer::iter()->lock_exe(&MatDataBuffer::get, device_index, seg_results.label(), pointMat, 10)) {
  105. cv::imshow(std::to_string(device_index), pointMat);
  106. cv::waitKey(1);
  107. have_cloud.push_back(device_index);
  108. for (int row = 0; row < 480; row++) {
  109. for (int col = seg_results.boxes(device_index).lines(row).begin(); col < seg_results.boxes(device_index).lines(row).end(); col++) {
  110. vct_wheel_cloud[device_index]->emplace_back(pointMat.at<cv::Vec3f>(row, col)[0] * 0.001,
  111. pointMat.at<cv::Vec3f>(row, col)[1] * 0.001,
  112. pointMat.at<cv::Vec3f>(row, col)[2] * 0.001);
  113. }
  114. }
  115. pcl::transformPointCloud(*vct_wheel_cloud[device_index], *vct_wheel_cloud[device_index], rotation_matrix4[device_index]());
  116. // Point3D_tool::WriteTxtCloud(ETC_PATH PROJECT_NAME "/data/cloud_" + std::to_string(device_index) + ".txt", vct_wheel_cloud[device_index]);
  117. }
  118. }
  119. // 点云处理
  120. if (!have_cloud.empty()) {
  121. WheelCloudOptimize(vct_wheel_cloud, detect_result);
  122. cost = std::chrono::steady_clock::now() - t_start_time;
  123. DLOG(INFO) << "cost time is " << cost.count() * 1000 << " ms";
  124. bool ceres_detect_ret = false;
  125. auto wheel_detect = detect_wheel_ceres::iter();
  126. // 采用上次测量结果的数据
  127. if (!m_detect_result_time_lock_data.timeout(0.5)) {
  128. detect_result = m_detect_result_time_lock_data();
  129. detect_result.car.wheel_width = MIN(detect_result.car.wheel_width + 0.4, 2.5);
  130. } else {
  131. detect_result.car.wheel_width = 3;
  132. detect_result.car.wheel_base = 2.7;
  133. }
  134. // 根据车轮数采用不同的优化方式
  135. if (m_detect_result_record.car.wheel_width < 1.0 || m_detect_result_record.car.wheel_base < 2.0 ||
  136. have_cloud.size() > 2) {
  137. ceres_detect_ret = wheel_detect->detect_2Step(vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_LF], vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_RF],
  138. vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_LR], vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_RR],
  139. detect_result.car.wheels_center_x, detect_result.car.wheels_center_y,
  140. detect_result.car.theta, detect_result.car.wheel_base,
  141. detect_result.car.wheel_width,detect_result.car.front_wheels_theta, detect_result.car.loss);
  142. if ((m_detect_result_record.car.wheel_width < 1.0 && m_detect_result_record.car.wheel_base < 1.0 && have_cloud.size() == 2) || have_cloud.size() < 2) {
  143. detect_result.car.wheel_width = m_detect_result_record.car.wheel_width;
  144. detect_result.car.wheel_base = m_detect_result_record.car.wheel_base;
  145. } else if (have_cloud.size() == 2) {
  146. for (int device_index = 0; device_index < DeviceAzimuth::DEVICE_AZIMUTH_MAX; device_index++) {
  147. if (!vct_wheel_cloud[device_index]->empty()) {
  148. have_cloud.push_back(device_index);
  149. }
  150. }
  151. if (have_cloud[0] % 2 == have_cloud[1] %2 ) {
  152. detect_result.car.wheel_width = m_detect_result_record.car.wheel_width;
  153. } else {
  154. detect_result.car.wheel_base = m_detect_result_record.car.wheel_base;
  155. }
  156. }
  157. } else {
  158. ceres_detect_ret = wheel_detect->detect_dynP(vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_LF], vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_RF],
  159. vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_LR], vct_wheel_cloud[DeviceAzimuth::DEVICE_AZIMUTH_RR],
  160. detect_result.car.wheels_center_x, detect_result.car.wheels_center_y,
  161. detect_result.car.theta, m_detect_result_record.car.wheel_base,
  162. m_detect_result_record.car.wheel_width,detect_result.car.front_wheels_theta, detect_result.car.loss);
  163. detect_result.car.wheel_base = m_detect_result_record.car.wheel_base;
  164. detect_result.car.wheel_width = m_detect_result_record.car.wheel_width;
  165. }
  166. // 如果车辆运动,不断刷新运动时间,如果由运动转为静止则刷新运动状态到静止
  167. float move_distance = detect_result.car.wheels_center_y - m_detect_result_record.car.wheels_center_y;
  168. if (fabs(move_distance) > 0.1) {
  169. move_statu = true;
  170. } else if (move_statu()) {
  171. move_statu = false;
  172. }
  173. if (ceres_detect_ret) {
  174. // 记录下测量结果
  175. m_detect_result_record = detect_result;
  176. detect_statu = Measure_OK;
  177. m_detect_result_time_lock_data = m_detect_result_record;
  178. } else {
  179. detect_statu = Measure_Failture;
  180. }
  181. }
  182. cost = std::chrono::steady_clock::now() - t_start_time;
  183. DLOG(INFO) << "cost time is " << cost.count() * 1000 << " ms";
  184. } else {
  185. cv::destroyAllWindows();
  186. }
  187. }
  188. LOG(INFO) << "detect thread " << std::this_thread::get_id() << " closed.";
  189. }
  190. void DetectManager::stop() {
  191. //杀死线程,
  192. m_condit.kill_all();
  193. //在线程join结束之后, 就可以可以回收线程内存资源
  194. m_thread->join();
  195. delete m_thread;
  196. m_thread = nullptr;
  197. }
  198. Error_manager DetectManager::WheelCloudOptimize(std::vector<pcl::PointCloud<pcl::PointXYZ>::Ptr> &vct_wheel_cloud, DetectResult &result) {
  199. //下采样
  200. pcl::VoxelGrid<pcl::PointXYZ> vox; //创建滤波对象
  201. vox.setLeafSize(0.03f, 0.03f, 0.03f); //设置滤波时创建的体素体积为1cm的立方体
  202. pcl::PointCloud<pcl::PointXYZ>::Ptr filtered_cloud(new pcl::PointCloud<pcl::PointXYZ>);
  203. for (int device_index = 0; device_index < DeviceAzimuth::DEVICE_AZIMUTH_MAX; device_index++) {
  204. if (vct_wheel_cloud[device_index]->empty()) { continue;}
  205. vox.setInputCloud(vct_wheel_cloud[device_index]); //设置需要过滤的点云给滤波对象
  206. vox.filter(*vct_wheel_cloud[device_index]); //执行滤波处理,存储输出
  207. }
  208. return {};
  209. }
  210. MeasureStatu DetectManager::getMeasureResult(DetectManager::DetectResult &result) {
  211. if (detect_statu() != MeasureStatu::Measure_OK) {
  212. return detect_statu();
  213. }
  214. if (m_detect_result_time_lock_data.timeout()) {
  215. return MeasureStatu::Measure_Failture;
  216. }
  217. result = m_detect_result_time_lock_data();
  218. return MeasureStatu::Measure_OK;
  219. }
  220. Error_manager DetectManager::carMoveStatus() {
  221. if (move_statu.timeout(3) && !move_statu()) {
  222. return {FAILED, NORMAL};
  223. }
  224. return {};
  225. }
  226. Error_manager DetectManager::Stop() {
  227. m_condit.kill_all();
  228. m_thread->join();
  229. delete m_thread;
  230. m_thread = nullptr;
  231. return {};
  232. }
  233. Error_manager DetectManager::Continue() {
  234. m_condit.reset();
  235. m_thread = new std::thread(&DetectManager::run, this);
  236. m_condit.notify_all(true);
  237. return {};
  238. }