linux C++ onnxruntime yolov8 视频检测Demo

天天代码码天天2024-07-27 16:13

linux C++ onnxruntime yolov8 视频检测Demo

目录

项目目录

效果

​编辑CMakeLists.txt

代码

下载

项目目录

效果

./yolov8_demo --help

./yolov8_demo -c=2 -p=true

./yolov8_demo -c=1 -s=true

CMakeLists.txt

复制代码 
# cmake needs this line
cmake_minimum_required(VERSION 3.0)

# Define project name
project(yolov8_demo)

# Release模式下的编译指令
# SET(CMAKE_BUILD_TYPE "Release")
# set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -s")
# set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -std=c++17 -pthread -Wall -Wl")

# Debug模式下的编译指令
SET(CMAKE_BUILD_TYPE "Debug")
set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_DEBUG}")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -std=c++17 -pthread") 

set(OpenCV_LIBS opencv_videoio opencv_imgcodecs opencv_imgproc opencv_core opencv_dnn opencv_highgui)

include_directories(
    /usr/local/include/opencv4
    ${PROJECT_SOURCE_DIR}/include
    ${PROJECT_SOURCE_DIR}/include/onnxruntime
)

link_directories(
    ${PROJECT_SOURCE_DIR}/lib/onnxruntime   # 第三方动态库文件
    /usr/local/lib/
)

#递归指定源码的路径
file(GLOB_RECURSE SRCS ${PROJECT_SOURCE_DIR}/src/*.cpp)

# Declare the executable target built from your sources
add_executable(yolov8_demo ${SRCS})

# Link your application with OpenCV libraries
target_link_libraries(yolov8_demo 
-lonnxruntime
${OpenCV_LIBS}
)

代码

main.cpp

#include <opencv2/core.hpp>

#include <opencv2/highgui.hpp>

#include <iostream>

#include <YoloV8.hpp>

#include <unistd.h>

#include <sys/syscall.h>

#include <thread>

int VideoDet(int index, bool showDet, bool useGPU, bool printPerStepInfo)

{

size_t threadId = static_cast<size_t>(syscall(SYS_gettid));

// std::cout << "index:" << index << " thread id: " << threadId << std::endl;

cv::VideoCapture capture("./test/test_car_person_1080P.mp4");

// 检查视频是否成功打开

if (!capture.isOpened())

{

std::cout << "无法读取视频文件" << std::endl;

return -1;

}

int frameCount = capture.get(cv::CAP_PROP_FRAME_COUNT); // 获取视频帧数

double fps = capture.get(cv::CAP_PROP_FPS); // 获取帧率

int delay = int(1000 / fps); // 根据帧率计算帧间间隔时间

// delay=1;

std::string model_path = "./models/yolov8n.onnx";

std::string lable_path = "./models/lable.txt";

int GPUCount = 2;

int device_id = 0;

if (index >= GPUCount)

{

device_id = index % GPUCount;

}

else

{

device_id = index;

}

// device_id=0;

YoloV8 yoloV8(model_path, lable_path, useGPU, device_id);

yoloV8.index = index;

yoloV8.threadId = threadId;

yoloV8.videoFps = fps;

yoloV8.frameCount = frameCount;

// std::cout << "device_id:" << yoloV8.device_id << std::endl;

// vector<DetectionResult> detectionResult;

// Mat frame=cv::imread("../test/dog.jpg");

// yoloV8.Detect(frame, detectionResult);

// std::cout << "detectionResult size:" << detectionResult.size() << std::endl;

string winname = "detectionResult-" + std::to_string(index);

while (true)

{

double start = (double)cv::getTickCount();

delay = int(1000 / fps);

Mat frame;

bool success = capture.read(frame); // 读取一帧数据

// 检查是否成功读取帧

if (!success)

{

std::cout << "index:" << index << ",读取完毕" << std::endl;

yoloV8.PrintAvgCostTime();

break;

}

vector<DetectionResult> detectionResult;

yoloV8.Detect(frame, detectionResult);

// std::cout <<"index:"<<index<< " thread id: " << threadId << " detectionResult size: " << detectionResult.size() << std::endl;

yoloV8.detectionResultSize = detectionResult.size();

if (printPerStepInfo)

{

yoloV8.PrintCostTime();

yoloV8.PrintAvgCostTime();

}

if (showDet)

{

yoloV8.Draw(frame, detectionResult);

imshow(winname, frame);

double costTime = ((double)getTickCount() - start) / getTickFrequency();

delay = delay - costTime;

if (delay <= 0)

{

delay = 1;

}

if (waitKey(delay) == 27) // 通过按下ESC键退出循环

{

break;

}

}

}

capture.release(); // 释放视频对象

if (showDet)

{

cv::destroyWindow(winname);

}

return 0;

}

int main(int argc, char *const argv[])

{

int threadCount = 1;

bool showDet = false;

bool useGPU = false;

bool printPerStepInfo = true;

const char *keys ="{h help || print this message}"

"{c threadCount | 1 | run thread number}"

"{s showDet | false | show detection result}"

"{g useGPU | false | use GPU}"

"{p printPerStepInfo | false | print per Step Info}";

cv::CommandLineParser parser(argc, argv, keys);

if(parser.has("help"))

{

parser.about("YoloV8 demo v1.0.0");

parser.printMessage();

return 0;

}

threadCount=parser.get<int>("threadCount");

showDet=parser.get<bool>("showDet");

useGPU=parser.get<bool>("useGPU");

printPerStepInfo=parser.get<bool>("printPerStepInfo");

std::cout << std::boolalpha;

std::cout << "threadCount:" << threadCount << ",showDet:" << showDet<< ",useGPU:" << useGPU << ",printPerStepInfo:" << printPerStepInfo << std::endl;

for (size_t i = 0; i < threadCount; i++)

{

std::thread thread(VideoDet, i, showDet, useGPU, printPerStepInfo);

thread.detach();

}

while (true)

{

sleep(100);

}

return 0;

}

复制代码 
#include <opencv2/core.hpp>
#include <opencv2/highgui.hpp>
#include <iostream>
#include <YoloV8.hpp>
#include <unistd.h>
#include <sys/syscall.h>
#include <thread>

int VideoDet(int index, bool showDet, bool useGPU, bool printPerStepInfo)
{
    size_t threadId = static_cast<size_t>(syscall(SYS_gettid));
    // std::cout << "index:" << index << " thread id: " << threadId << std::endl;

    cv::VideoCapture capture("./test/test_car_person_1080P.mp4");

    // 检查视频是否成功打开
    if (!capture.isOpened())
    {
        std::cout << "无法读取视频文件" << std::endl;
        return -1;
    }

    int frameCount = capture.get(cv::CAP_PROP_FRAME_COUNT); // 获取视频帧数
    double fps = capture.get(cv::CAP_PROP_FPS);             // 获取帧率
    int delay = int(1000 / fps);                            // 根据帧率计算帧间间隔时间
    // delay=1;

    std::string model_path = "./models/yolov8n.onnx";
    std::string lable_path = "./models/lable.txt";

    int GPUCount = 2;
    int device_id = 0;
    if (index >= GPUCount)
    {
        device_id = index % GPUCount;
    }
    else
    {
        device_id = index;
    }

    // device_id=0;

    YoloV8 yoloV8(model_path, lable_path, useGPU, device_id);

    yoloV8.index = index;
    yoloV8.threadId = threadId;
    yoloV8.videoFps = fps;
    yoloV8.frameCount = frameCount;

    // std::cout << "device_id:" << yoloV8.device_id << std::endl;

    // vector<DetectionResult> detectionResult;
    // Mat frame=cv::imread("../test/dog.jpg");
    // yoloV8.Detect(frame, detectionResult);
    // std::cout << "detectionResult size:" << detectionResult.size() << std::endl;

    string winname = "detectionResult-" + std::to_string(index);

    while (true)
    {

        double start = (double)cv::getTickCount();
        delay = int(1000 / fps);

        Mat frame;
        bool success = capture.read(frame); // 读取一帧数据

        // 检查是否成功读取帧
        if (!success)
        {
            std::cout << "index:" << index << ",读取完毕" << std::endl;
            yoloV8.PrintAvgCostTime();
            break;
        }

        vector<DetectionResult> detectionResult;

        yoloV8.Detect(frame, detectionResult);

        // std::cout <<"index:"<<index<< " thread id: " << threadId << " detectionResult size: " << detectionResult.size() << std::endl;
        yoloV8.detectionResultSize = detectionResult.size();

        if (printPerStepInfo)
        {
            yoloV8.PrintCostTime();
            yoloV8.PrintAvgCostTime();
        }

        if (showDet)
        {
            yoloV8.Draw(frame, detectionResult);

            imshow(winname, frame);

            double costTime = ((double)getTickCount() - start) / getTickFrequency();
            delay = delay - costTime;
            if (delay <= 0)
            {
                delay = 1;
            }

            if (waitKey(delay) == 27) // 通过按下ESC键退出循环
            {
                break;
            }
        }
    }

    capture.release(); // 释放视频对象

    if (showDet)
    {
        cv::destroyWindow(winname);
    }

    return 0;
}

int main(int argc, char *const argv[])
{
    int threadCount = 1;
    bool showDet = false;
    bool useGPU = false;
    bool printPerStepInfo = true;

    const char *keys ="{h help                || print this message}"
        "{c threadCount         | 1     | run thread number}"
        "{s showDet             | false | show detection result}"
        "{g useGPU              | false | use GPU}"
        "{p printPerStepInfo    | false | print per Step Info}";

    cv::CommandLineParser parser(argc, argv, keys);

    if(parser.has("help"))
    {
        parser.about("YoloV8 demo v1.0.0");
        parser.printMessage();
        return 0;
    }

    threadCount=parser.get<int>("threadCount");
    showDet=parser.get<bool>("showDet");
    useGPU=parser.get<bool>("useGPU");
    printPerStepInfo=parser.get<bool>("printPerStepInfo");

    std::cout << std::boolalpha;
    std::cout << "threadCount:" << threadCount << ",showDet:" << showDet<< ",useGPU:" << useGPU << ",printPerStepInfo:" << printPerStepInfo << std::endl;

    for (size_t i = 0; i < threadCount; i++)
    {
        std::thread thread(VideoDet, i, showDet, useGPU, printPerStepInfo);
        thread.detach();
    }

    while (true)
    {
        sleep(100);
    }

    return 0;
}

下载

源码下载

相关推荐
c769几秒前
【文献笔记】Automatic Chain of Thought Prompting in Large Language Models
人工智能·笔记·语言模型·论文笔记
Blossom.11829 分钟前
机器学习在智能供应链中的应用：需求预测与物流优化
人工智能·深度学习·神经网络·机器学习·计算机视觉·机器人·语音识别
Gyoku Mint36 分钟前
深度学习×第4卷：Pytorch实战——她第一次用张量去拟合你的轨迹
人工智能·pytorch·python·深度学习·神经网络·算法·聚类
zzywxc78738 分钟前
AI大模型的技术演进、流程重构、行业影响三个维度的系统性分析
人工智能·重构
点控云39 分钟前
智能私域运营中枢：从客户视角看 SCRM 的体验革新与价值重构
大数据·人工智能·科技·重构·外呼系统·呼叫中心
zhaoyi_he1 小时前
多模态大模型的技术应用与未来展望：重构AI交互范式的新引擎
人工智能·重构
葫三生2 小时前
如何评价《论三生原理》在科技界的地位？
人工智能·算法·机器学习·数学建模·量子计算
m0_751336393 小时前
突破性进展：超短等离子体脉冲实现单电子量子干涉，为飞行量子比特奠定基础
人工智能·深度学习·量子计算·材料科学·光子器件·光子学·无线电电子
美狐美颜sdk6 小时前
跨平台直播美颜SDK集成实录：Android/iOS如何适配贴纸功能
android·人工智能·ios·架构·音视频·美颜sdk·第三方美颜sdk
DeepSeek-大模型系统教程6 小时前
推荐 7 个本周 yyds 的 GitHub 项目。
人工智能·ai·语言模型·大模型·github·ai大模型·大模型学习
热门推荐
01手机电脑之间快速传输图片视频文件，不压缩画质、不限制大小的方法！02Coze实战第13讲：飞书多维表格读取+豆包生图模型，轻松批量生成短剧封面03Coze扣子平台完整体验和实践（附国内和国际版对比）04华为昇腾 910B 部署 DeepSeek-R1 蒸馏系列模型详细指南05免费可用！最强AI数字人对口型神器：让照片开口说话唱歌，支持多人对口型+全身动作，1分钟学会！（附保姆级教程）06扣子（coze）实战|我用扣子搭建了一个自动分析小红薯笔记内容的AI应用|详细步骤拆解07Coze平台 创建AI智能体的详细步骤指南08C#调用WechatOCR.exe实现本地OCR文字识别09DeepSeek各版本说明与优缺点分析10Coze（扣子）智能体工作流：自动批量生成书单号视频，1分钟100个，书单号博主都在用！