前言

上一篇介绍过使用onnxruntime实现模型推理部署，但在我的机器上视频效果仍不理想，本篇介绍使用openvino完成模型推理部署。
openvino是Intel开发的深度学习模型推理加速引擎，支持python和C++，使用起来比较方便。

一、环境

1、硬件

Intel® Core i5-7400 CPU @ 3.00GHZ
Intel® HD Graphics 630 内存4G 核显
内存 8G
win10 64位系统

2、软件

opencv4.6.0
yolov5 6.2版本
qt5.6.2
openvino_2022.1.0.643 runtime版本（关于安装自行谷歌。。。）

二、YOLO模型

我使用的是onnx模型，如果没有训练过自己的模型，可以使用官方的yolov5s.pt模型，将其转化为yolov5s.onnx模型，转化方法如下：

python export.py

在yolov5-master目录下，可以看到yolov5s.onnx模型已生成。

三、新建Qt项目

1、pro文件

在pro文件中，添加opencv相关配置，内容如下：

#-------------------------------------------------
#
# Project created by QtCreator 2022-10-31T09:37:31
#
#-------------------------------------------------QT       += core guigreaterThan(QT_MAJOR_VERSION, 4): QT += widgetsTARGET = yolov5-openvino-cpp
TEMPLATE = appCONFIG += C++11#CONFIG(debug, debug|release){#    DESTDIR = ../out
#}
#else {#    DESTDIR = ../out
#}SOURCES += main.cpp\mainwindow.cppHEADERS  += mainwindow.hFORMS    += mainwindow.uiINCLUDEPATH += C:/opencv4.6.0/build/includeC:/opencv4.6.0/build/include/opencv2
LIBS += -LC:/opencv4.6.0/build/x64/vc14/lib/ -lopencv_world460INCLUDEPATH += C:/onnxruntime-win-x64-1.11.1/include
LIBS += -LC:/onnxruntime-win-x64-1.11.1/lib/ -lonnxruntimeINCLUDEPATH += 'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/include''C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/include/ie''C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/include/ngraph''C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/include/openvino'LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino
LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino_c
LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino_ir_frontend
LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino_onnx_frontend
LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino_paddle_frontend
LIBS += -L'C:/Program Files (x86)/Intel/openvino_2022.1.0.643/runtime/lib/intel64/Release/' -lopenvino_tensorflow_fe

2、main.cpp

#include "mainwindow.h"
#include <QApplication>
#include <QDebug>
#include <fstream>
#include <iostream>
#include <sstream>
#include <opencv2\opencv.hpp>
#include <openvino\openvino.hpp>using namespace std;
using namespace ov;
using namespace cv;//数据集的标签 官方模型yolov5s.onnx为80种
vector<string> class_names = {"person", "bicycle", "car", "motorbike", "aeroplane", "bus", "train", "truck", "boat", "traffic light","fire hydrant","stop sign", "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow", "elephant", "bear", "zebra", "giraffe","backpack", "umbrella","handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite", "baseball bat", "baseball glove","skateboard", "surfboard","tennis racket", "bottle", "wine glass", "cup", "fork", "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange","broccoli", "carrot","hot dog", "pizza", "donut", "cake", "chair", "sofa", "pottedplant", "bed", "diningtable", "toilet", "tvmonitor", "laptop", "mouse","remote","keyboard", "cell phone", "microwave", "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase", "scissors", "teddy bear", "hair drier", "toothbrush"};//模型文件路径
string ir_filename = "G:/yolov5s.onnx";/*@brief 对网络的输入为图片数据的节点进行赋值，实现图片数据输入网络@param input_tensor 输入节点的tensor@param inpt_image 输入图片数据
*/
void fill_tensor_data_image(ov::Tensor& input_tensor, const cv::Mat& input_image)
{// 获取输入节点要求的输入图片数据的大小ov::Shape tensor_shape = input_tensor.get_shape();const size_t width = tensor_shape[3]; // 要求输入图片数据的宽度const size_t height = tensor_shape[2]; // 要求输入图片数据的高度const size_t channels = tensor_shape[1]; // 要求输入图片数据的维度// 读取节点数据内存指针float* input_tensor_data = input_tensor.data<float>();// 将图片数据填充到网络中// 原有图片数据为 H、W、C 格式，输入要求的为 C、H、W 格式for (size_t c = 0; c < channels; c++){for (size_t h = 0; h < height; h++){for (size_t w = 0; w < width; w++){input_tensor_data[c * width * height + h * width + w] = input_image.at<cv::Vec<float, 3>>(h, w)[c];}}}
}int main(int argc, char *argv[])
{ov::Core ie;ov::CompiledModel compiled_model = ie.compile_model(ir_filename, "AUTO");//AUTO  GPU CPUov::InferRequest infer_request = compiled_model.create_infer_request();// 预处理输入数据 - 格式化操作cv::VideoCapture cap("G:/1.mp4");////VideoCapture cap;if (!cap.isOpened()){std::cerr << "Error opening video file\n";return -1;}// 获取输入节点tensorTensor input_image_tensor = infer_request.get_tensor("images");int input_h = input_image_tensor.get_shape()[2]; //获得"images"节点的Heightint input_w = input_image_tensor.get_shape()[3]; //获得"images"节点的Width//int input_c = input_image_tensor.get_shape()[1]; //获得"images"节点的channelcout << "input_h:" << input_h << "; input_w:" << input_w << endl;cout << "input_image_tensor's element type:" << input_image_tensor.get_element_type() << endl;cout << "input_image_tensor's shape:" << input_image_tensor.get_shape() << endl;// 获取输出节点tensorTensor output_tensor = infer_request.get_tensor("output");int out_rows = output_tensor.get_shape()[1]; //获得"output"节点的out_rowsint out_cols = output_tensor.get_shape()[2]; //获得"output"节点的Widthcout << "out_cols:" << out_cols << "; out_rows:" << out_rows << endl;//Mat frame;//连续采集处理循环while (true){cap.read(frame);if (frame.empty()){std::cout << "End of stream\n";break;}//imshow("frame", frame);int64 start = cv::getTickCount();// 图像预处理 视频分辨率int w = frame.cols;int h = frame.rows;//qDebug()<<"w ="<<w<<"h ="<<h;int _max = std::max(h, w);cv::Mat image = cv::Mat::zeros(cv::Size(_max, _max), CV_8UC3);cv::Rect roi(0, 0, w, h);frame.copyTo(image(roi));cvtColor(image, image, COLOR_BGR2RGB); //交换RB通道//qDebug()<<"image cols ="<<image.cols<<"image rows"<<image.rows//       <<"input_w ="<<input_w<<"input_h ="<<input_h;float x_factor = image.cols / input_w;float y_factor = image.rows / input_h;cv::Mat blob_image;cv::resize(image, blob_image, cv::Size(input_w, input_h));//cv::cvtColor(blob_image, blob_image, cv::COLOR_BGR2RGB);blob_image.convertTo(blob_image, CV_32F);blob_image = blob_image / 255.0;// 将图片数据填充到tensor数据内存中fill_tensor_data_image(input_image_tensor, blob_image);// 执行推理计算infer_request.infer();// 获得推理结果const ov::Tensor& output_tensor = infer_request.get_tensor("output");// 解析推理结果，YOLOv5 output format: cx,cy,w,h,scorecv::Mat det_output(out_rows, out_cols, CV_32F, (float*)output_tensor.data());//vector<cv::Rect> boxes;vector<int> classIds;vector<float> confidences;for (int i = 0; i < det_output.rows; i++){float confidence = det_output.at<float>(i, 4);//qDebug()<<"confidence ="<<confidence;if (confidence < 0.4){continue;}//官方模型为85cv::Mat classes_scores = det_output.row(i).colRange(5, 85);cv::Point classIdPoint;double score;cv::minMaxLoc(classes_scores, 0, &score, 0, &classIdPoint);// 置信度 0～1之间//qDebug()<<"score ="<<score;if (score > 0.8)//0.65{float cx = det_output.at<float>(i, 0);float cy = det_output.at<float>(i, 1);float ow = det_output.at<float>(i, 2);float oh = det_output.at<float>(i, 3);int x = static_cast<int>((cx - 0.5 * ow) * x_factor);int y = static_cast<int>((cy - 0.5 * oh) * y_factor);int width = static_cast<int>(ow * x_factor);int height = static_cast<int>(oh * y_factor);//qDebug()<<"cx ="<<cx<<"cy ="<<cy//       <<"ow ="<<ow<<"oh ="<<oh//      <<"x ="<<x<<"y ="<<y//     <<"width ="<<width<<"height ="<<height;//cv::Rect box;box.x = x;box.y = y;box.width = width;box.height = height;boxes.push_back(box);classIds.push_back(classIdPoint.x);confidences.push_back(score);}}// NMSvector<int> indexes;cv::dnn::NMSBoxes(boxes, confidences, 0.25, 0.45, indexes);for (size_t i = 0; i < indexes.size(); i++){int index = indexes[i];int idx = classIds[index];float iscore = confidences[index];//cv::rectangle(frame, boxes[index], cv::Scalar(0, 0, 255), 2, 8);cv::rectangle(frame, cv::Point(boxes[index].tl().x, boxes[index].tl().y - 20),cv::Point(boxes[index].br().x, boxes[index].tl().y), cv::Scalar(0, 255, 255), -1);//String nameText = class_names[idx];nameText.append(to_string(iscore));cv::putText(frame, nameText/*class_names[idx]*/, cv::Point(boxes[index].tl().x, boxes[index].tl().y - 10), cv::FONT_HERSHEY_SIMPLEX, .5, cv::Scalar(0, 0, 0));//目标中心点Point2f point(boxes[index].x + boxes[index].width/2,boxes[index].y + boxes[index].height/2);circle(frame, point, 3, Scalar(0, 255, 0), -1);//String posText = "x:";posText.append(to_string(boxes[index].x + + boxes[index].width/2));posText.append(",");posText.append("y:");posText.append(to_string(boxes[index].y + + boxes[index].height/2));//qDebug()<<QString::fromStdString(posText);cv::putText(frame, posText, point, cv::FONT_HERSHEY_SIMPLEX, .8, cv::Scalar(255, 255, 255));}// 计算FPS render itfloat t = (cv::getTickCount() - start) / static_cast<float>(cv::getTickFrequency());//cout << "Infer time(ms): " << t * 1000 << "ms; Detections: " << indexes.size() << endl;putText(frame, cv::format("FPS: %.2f", 1.0 / t), cv::Point(20, 40), cv::FONT_HERSHEY_PLAIN, 2.0, cv::Scalar(255, 0, 0), 2, 8);cv::namedWindow("YOLOv5-6.2 + OpenVINO 2022.1 C++ Demo",0);cv::imshow("YOLOv5-6.2 + OpenVINO 2022.1 C++ Demo", frame);char c = cv::waitKey(1);if (c == 27){// ESCbreak;}}cv::waitKey(0);cv::destroyAllWindows();return 0;
}

四、效果

以本地视频文件1.mp4为例，效果如下：

可以看到FPS获得较大提升，视频流明显比之前效果要好。

五、后记

以上代码是基于qt5.6.2 编译的64位程序，32位编译不通过，貌似需要自己编译openvino 32bit版本。我尝试编译过openvino 32bit，但部分库编译失败，该qt demo运行失败。如果有博主编译成功该版本openvino 32bit，辛苦告知一下，万分感谢！！！

win10系统下使用openvino部署yolov5模型相关推荐

win10系统下使用opencv-dnn部署yolov5模型
文章目录前言一.环境 1.硬件 2.软件二.YOLO模型三.新建Qt项目 1.pro文件 2.main函数四.YOLO 类封装 1.yolo.h 2.yolo.cpp 3.classes.t ...
win10环境下基于OpenVINO部署yolov5模型
以在Intel的OpenVINO上部署Yolov5模型为例,加深对模型部署的理解. 1. 训练准备获取yolov5模型及数据集 git clone git://github.com/ultralyt ...
win10系统下使用onnxruntime部署yolov5模型
文章目录前言一.环境 1.硬件 2.软件二.YOLO模型三.新建Qt项目 1.pro文件 2.mainwindow.h 3.mainwindow.cpp 四.YOLO 类封装 1.yolov5 ...
Win10系统下面的TR1008解决方案
Win10系统下面的TR1008解决方案最近为了体验高大上的Win10系统,于是就把本本原来的Win7旗舰版升级到了Win10专业版.之后又在本本上部署了cognos,但是在打开Transfor ...
Vagrant在，win7/win10系统下搭建使用
vagrant是什么? vagrant是一个操作虚拟机的工具. 通过命令和配置文件来管理虚拟机,很快就能完成一套开发环境的部署,并可以打包传播,统一了开发环境,也解决了重复配置环境的麻 ...
egret与php相连,JavaScript_Mac OS X 系统下安装和部署Egret引擎开发环境，概述 Egret基于TypeScript开 - phpStudy...
Mac OS X 系统下安装和部署Egret引擎开发环境概述 Egret基于TypeScript开发的,而TypeScript编译工具tsc是基于Node.js开发的.所以在安装过程中,我们先需要对 ...
windows 10 php安装教程,win10系统下wnmp的安装教程介绍
本篇文章给大家带来的内容是关于win10系统下wnmp的安装教程介绍,有一定的参考价值,有需要的朋友可以参考一下,希望对你有所帮助. 刚刚开始学习PHP时,一直使用phpstudy,后面发现很多东西自 ...
tensorRT 部署 YOLOV5模型详解
tensorRT 部署 YOLOV5模型详解第一步: 下载tensorRT库 https://developer.nvidia.com/nvidia-tensorrt-8x-download 欢迎使 ...
基于Caffe格式部署YOLOV5模型
点击上方"3D视觉工坊",选择"星标" 干货第一时间送达 [导语]本文为大家介绍了一个caffe部署yolov5 模型的教程,并开源了全部代码.主要是教你如何搭 ...

win10系统下使用openvino部署yolov5模型

文章目录

前言