#include "PL_SensetimeFaceTrack.h" #include "MaterialBuffer.h" #include "logger.h" #include #include struct PL_SensetimeFaceTrack_Internal { //uint8_t buffer[1920*1080*4]; //size_t buffSize; //size_t buffSizeMax; MB_Frame lastFrame; SensetimeFaceTrackConfig config; bool payError; cv_handle_t handle_track; PL_SensetimeFaceTrack_Internal() : //buffSize(0), buffSizeMax(sizeof(buffer)), lastFrame(), config(), payError(true), handle_track(nullptr) { } ~PL_SensetimeFaceTrack_Internal() { } void reset() { //buffSize = 0; payError = true; MB_Frame _lastFrame; lastFrame = _lastFrame; SensetimeFaceTrackConfig _config; config = _config; handle_track = nullptr; } }; PipeLineElem* create_PL_SensetimeFaceTrack() { return new PL_SensetimeFaceTrack; } PL_SensetimeFaceTrack::PL_SensetimeFaceTrack() : internal(new PL_SensetimeFaceTrack_Internal) { } PL_SensetimeFaceTrack::~PL_SensetimeFaceTrack() { delete (PL_SensetimeFaceTrack_Internal*)internal; internal= nullptr; } bool PL_SensetimeFaceTrack::init(void* args) { PL_SensetimeFaceTrack_Internal* in = (PL_SensetimeFaceTrack_Internal*)internal; in->reset(); SensetimeFaceTrackConfig* config = (SensetimeFaceTrackConfig*)args; in->config = *config; if (in->config.point_size == 21) in->config.point_size_config = CV_DETECT_ENABLE_ALIGN_21; else if (in->config.point_size == 106) in->config.point_size_config = CV_DETECT_ENABLE_ALIGN_106; else { LOG_ERROR << "alignment point size must be 21 or 106"; return false; } // init handle cv_result_t cv_result = cv_face_create_tracker(&(in->handle_track), nullptr, in->config.point_size_config | CV_FACE_TRACKING_TWO_THREAD); if (cv_result != CV_OK) { LOG_ERROR << "cv_face_create_tracker failed, error code" << cv_result; return false; } int val = 0; cv_result = cv_face_track_set_detect_face_cnt_limit(in->handle_track, in->config.detect_face_cnt_limit, &val); if (cv_result != CV_OK) { LOG_ERROR << "cv_face_track_set_detect_face_cnt_limit failed, error : " << cv_result; return false; } else LOG_ERROR << "detect face count limit : " << val; return true; } void PL_SensetimeFaceTrack::finit() { PL_SensetimeFaceTrack_Internal* in = (PL_SensetimeFaceTrack_Internal*)internal; // destroy track handle cv_face_destroy_tracker(in->handle_track); in->handle_track = nullptr; } int doFaceTrack(PL_SensetimeFaceTrack_Internal* in, uint8_t* buffer, size_t width, size_t height, size_t stride, cv_pixel_format cvPixFmt) { //resize(bgr_frame, bgr_frame, Size(frame_width, frame_height), 0, 0, INTER_LINEAR); int face_count = 0; cv_result_t cv_result = CV_OK; cv_face_t* p_face = nullptr; // realtime track cv_result = cv_face_track(in->handle_track, buffer, cvPixFmt, width, height, stride, CV_FACE_UP, &p_face, &face_count); if (cv_result != CV_OK) { LOG_ERROR << "cv_face_track failed, error : " << cv_result; cv_face_release_tracker_result(p_face, face_count); return -1; } // draw the video //cv::Mat yuvMat(cv::Size(width,height), CV_8UC3, buffer); cv::Mat yMat(cv::Size(width,height), CV_8UC1, buffer); for (int i = 0; i < face_count; i++) { LOGP(DEBUG, "face: %d-----[%d, %d, %d, %d]-----id: %d", i, p_face[i].rect.left, p_face[i].rect.top, p_face[i].rect.right, p_face[i].rect.bottom, p_face[i].ID); LOGP(DEBUG, "face pose: [yaw: %.2f, pitch: %.2f, roll: %.2f, eye distance: %.2f]", p_face[i].yaw, p_face[i].pitch, p_face[i].roll, p_face[i].eye_dist); if (in->config.draw_face_rect) { cv::Scalar scalar_color = CV_RGB(p_face[i].ID * 53 % 256, p_face[i].ID * 93 % 256, p_face[i].ID * 143 % 256); //cv::rectangle(yMat, cv::Point2f(0, 0), cv::Point2f(50, 50), scalar_color, 2); //cv::rectangle(yMat, cv::Point2f(500, 500), cv::Point2f(550, 550), scalar_color, 2); cv::rectangle(yMat, cv::Point2f(static_cast(p_face[i].rect.left), static_cast(p_face[i].rect.top)), cv::Point2f(static_cast(p_face[i].rect.right), static_cast(p_face[i].rect.bottom)), scalar_color, 2); } for (int j = 0; j < p_face[i].points_count; j++) { if (in->config.draw_face_feature_point) { cv::circle(yMat, cv::Point2f(p_face[i].points_array[j].x, p_face[i].points_array[j].y), 1, cv::Scalar(255, 255, 255)); } } } //if (face_count > 0) //{ // static size_t f=0; // char fname[50]; // sprintf(fname, "face-%u.yuv420", ++f); // FILE * pFile = fopen (fname,"wb"); // fwrite (yuvMat.data , sizeof(char), 1920*1080*1.5, pFile); // printf("write face file %s\n", fname); // fclose(pFile); //} // release the memory of face cv_face_release_tracker_result(p_face, face_count); return face_count; } bool PL_SensetimeFaceTrack::pay(const PipeMaterial& pm) { PL_SensetimeFaceTrack_Internal* in = (PL_SensetimeFaceTrack_Internal*)internal; if (pm.type != PipeMaterial::PMT_FRAME) { LOG_ERROR << "PL_H264Encoder::pay only support PMT_FRAME"; return false; } if (pm.buffer == nullptr) return false; MB_Frame* frame = (MB_Frame*)pm.buffer; if (frame->type != MB_Frame::MBFT_YUV420) { LOG_ERROR << "PL_H264Encoder::pay only support MBFT_YUV420"; return false; } int face_count = doFaceTrack( in, (uint8_t*)frame->buffer, frame->width, frame->height, frame->width, CV_PIX_FMT_YUV420P); if (face_count < 0) { in->payError = true; return false; } else in->payError = false; //in->buffer readly in->lastFrame.type = MB_Frame::MBFT_YUV420; in->lastFrame.buffer = frame->buffer;//#todo should copy in->lastFrame.buffSize = frame->buffSize; in->lastFrame.width = frame->width; in->lastFrame.height = frame->height; in->lastFrame.pts = frame->pts; return true; } bool PL_SensetimeFaceTrack::gain(PipeMaterial& pm) { PL_SensetimeFaceTrack_Internal* in = (PL_SensetimeFaceTrack_Internal*)internal; if (!in->payError) { pm.type = PipeMaterial::PMT_FRAME; pm.buffer = &(in->lastFrame); pm.buffSize = 0; pm.former = this; } pm.former = this; return !in->payError; }