【重要说明】
该系统以opencvsharp作图像处理,onnxruntime做推理引擎,使用CPU进行推理,适合有显卡或者没有显卡windows x64系统均可,不支持macOS和Linux系统,不支持x86的windows操作系统。由于采用CPU推理,要比GPU慢。为了适合大部分操作系统我们暂时只写了CPU推理源码,GPU推理源码后期根据需要可能会调整,目前只考虑CPU推理,主要是为了照顾现在大部分使用该源码是学生,很多人并没有显卡的电脑情况。
【算法介绍】
基于YOLOv8的人脸表情检测系统是一个结合了先进目标检测算法(YOLOv8)与深度学习技术的项目,旨在实时或离线地识别并分类人脸表情(如快乐、悲伤、愤怒、惊讶、恐惧、厌恶、中立等)。以下是一个简短的介绍,概述了该系统Python源码的核心要点:
该系统直接利用YOLOv8模型进行人脸表情识别。YOLOv8以其高效的速度和准确性著称,非常适合实时应用。
Python源码实现通常包括以下几个关键部分:
- 环境配置:安装必要的库,如onnxruntime(用于YOLOv8模型推理)、OpenCV(用于图像处理)等。
- 模型加载:加载预训练的YOLOv8人脸表情识别模型。
- 图像/视频处理:读取输入图像或视频,进行预处理(如缩放、归一化)。
- 人脸检测与表情识别:使用YOLOv8直接输出人脸表情检测结果。
- 结果展示:在图像或视频帧上标注检测到的人脸和表情,并显示结果。
这样的系统可以广泛应用于人机交互、情感分析、安全监控等领域。
【效果演示】
【测试环境】
windows10 x64系统
VS2019
netframework4.7.2
opencvsharp4.8.0
onnxruntime1.16.3
【模型可以检测出类别】
anger
content
disgust
fear
happy
neutral
sad
surprise
【相关参考训练数据集】
https://blog.csdn.net/FL1623863129/article/details/128648951
【训练信息】
|-----------------|-------|
| 参数 | 值 |
| 训练集图片数 | 6586 |
| 验证集图片数 | 1873 |
| 训练map | 76.6% |
| 训练精度(Precision) | 64.1% |
| 训练召回率(Recall) | 75.4% |
【部分实现源码】
using OpenCvSharp;
using System;
using System.Collections.Generic;
using System.ComponentModel;
using System.Data;
using System.Diagnostics;
using System.Drawing;
using System.IO;
using System.Linq;
using System.Text;
using System.Threading;
using System.Threading.Tasks;
using System.Windows.Forms;
namespace FIRC
{
public partial class Form1 : Form
{
public bool videoStart = false;//视频停止标志
string weightsPath = Application.StartupPath + "\\weights";//模型目录
string labelTxt= Application.StartupPath + "\\weights\\class_names.txt";//类别文件
Yolov8Manager detetor = new Yolov8Manager();//推理引擎
public Form1()
{
InitializeComponent();
CheckForIllegalCrossThreadCalls = false;//线程更新控件不报错
}
private void LoadWeightsFromDir()
{
var di = new DirectoryInfo(weightsPath);
foreach(var fi in di.GetFiles("*.onnx"))
{
comboBox1.Items.Add(fi.Name);
}
if(comboBox1.Items.Count>0)
{
comboBox1.SelectedIndex = 0;
}
else
{
tssl_show.Text = "未找到模型,请关闭程序,放入模型到weights文件夹!";
tsb_pic.Enabled = false;
tsb_video.Enabled = false;
tsb_camera.Enabled = false;
}
}
private void Form1_Load(object sender, EventArgs e)
{
LoadWeightsFromDir();//从目录加载模型
}
public string GetResultString(Result result)
{
Dictionary<string, int> resultDict = new Dictionary<string, int>();
for (int i = 0; i < result.length; i++)
{
if(resultDict.ContainsKey( result.classes[i]) )
{
resultDict[result.classes[i]]++;
}
else
{
resultDict[result.classes[i]]=1;
}
}
var resultStr = "";
foreach(var item in resultDict)
{
resultStr += string.Format("{0}:{1}\n",item.Key,item.Value);
}
return resultStr;
}
private void tsb_pic_Click(object sender, EventArgs e)
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.Filter = "*.*|*.bmp;*.jpg;*.jpeg;*.tiff;*.tiff;*.png";
if (ofd.ShowDialog() != DialogResult.OK) return;
tssl_show.Text = "正在检测中...";
Task.Run(() => {
var sw = new Stopwatch();
sw.Start();
Mat image = Cv2.ImRead(ofd.FileName);
detetor.Confidence =Convert.ToSingle(numericUpDown1.Value);
detetor.IOU = Convert.ToSingle(numericUpDown2.Value);
var results=detetor.Inference(image);
var resultImage = detetor.DrawImage(OpenCvSharp.Extensions.BitmapConverter.ToBitmap(image), results);
sw.Stop();
pb_show.Image = resultImage;
tb_res.Text = GetResultString(results);
tssl_show.Text = "检测已完成!总计耗时"+sw.Elapsed.TotalSeconds+"秒";
});
}
public void VideoProcess(string videoPath)
{
Task.Run(() => {
detetor.Confidence = Convert.ToSingle(numericUpDown1.Value);
detetor.IOU = Convert.ToSingle(numericUpDown2.Value);
VideoCapture capture = new VideoCapture(videoPath);
if (!capture.IsOpened())
{
tssl_show.Text="视频打开失败!";
return;
}
Mat frame = new Mat();
var sw = new Stopwatch();
int fps = 0;
while (videoStart)
{
capture.Read(frame);
if (frame.Empty())
{
Console.WriteLine("data is empty!");
break;
}
sw.Start();
var results = detetor.Inference(frame);
var resultImg = detetor.DrawImage(frame,results);
sw.Stop();
fps = Convert.ToInt32(1 / sw.Elapsed.TotalSeconds);
sw.Reset();
Cv2.PutText(resultImg, "FPS=" + fps, new OpenCvSharp.Point(30, 30), HersheyFonts.HersheyComplex, 1.0, new Scalar(255, 0, 0), 3);
//显示结果
pb_show.Image = OpenCvSharp.Extensions.BitmapConverter.ToBitmap(resultImg);
tb_res.Text = GetResultString(results);
Thread.Sleep(5);
}
capture.Release();
pb_show.Image = null;
tssl_show.Text = "视频已停止!";
tsb_video.Text = "选择视频";
});
}
public void CameraProcess(int cameraIndex=0)
{
Task.Run(() => {
detetor.Confidence = Convert.ToSingle(numericUpDown1.Value);
detetor.IOU = Convert.ToSingle(numericUpDown2.Value);
VideoCapture capture = new VideoCapture(cameraIndex);
if (!capture.IsOpened())
{
tssl_show.Text = "摄像头打开失败!";
return;
}
Mat frame = new Mat();
var sw = new Stopwatch();
int fps = 0;
while (videoStart)
{
capture.Read(frame);
if (frame.Empty())
{
Console.WriteLine("data is empty!");
break;
}
sw.Start();
var results = detetor.Inference(frame);
var resultImg = detetor.DrawImage(frame, results);
sw.Stop();
fps = Convert.ToInt32(1 / sw.Elapsed.TotalSeconds);
sw.Reset();
Cv2.PutText(resultImg, "FPS=" + fps, new OpenCvSharp.Point(30, 30), HersheyFonts.HersheyComplex, 1.0, new Scalar(255, 0, 0), 3);
//显示结果
pb_show.Image = OpenCvSharp.Extensions.BitmapConverter.ToBitmap(resultImg);
tb_res.Text = GetResultString(results);
Thread.Sleep(5);
}
capture.Release();
pb_show.Image = null;
tssl_show.Text = "摄像头已停止!";
tsb_camera.Text = "打开摄像头";
});
}
private void tsb_video_Click(object sender, EventArgs e)
{
if(tsb_video.Text=="选择视频")
{
OpenFileDialog ofd = new OpenFileDialog();
ofd.Filter = "视频文件(*.*)|*.mp4;*.avi";
if (ofd.ShowDialog() != DialogResult.OK) return;
videoStart = true;
VideoProcess(ofd.FileName);
tsb_video.Text = "停止";
tssl_show.Text = "视频正在检测中...";
}
else
{
videoStart = false;
}
}
private void tsb_camera_Click(object sender, EventArgs e)
{
if (tsb_camera.Text == "打开摄像头")
{
videoStart = true;
CameraProcess(0);
tsb_camera.Text = "停止";
tssl_show.Text = "摄像头正在检测中...";
}
else
{
videoStart = false;
}
}
private void tsb_exit_Click(object sender, EventArgs e)
{
videoStart = false;
this.Close();
}
private void trackBar1_Scroll(object sender, EventArgs e)
{
numericUpDown1.Value = Convert.ToDecimal(trackBar1.Value / 100.0f);
}
private void trackBar2_Scroll(object sender, EventArgs e)
{
numericUpDown2.Value = Convert.ToDecimal(trackBar2.Value / 100.0f);
}
private void numericUpDown1_ValueChanged(object sender, EventArgs e)
{
trackBar1.Value = (int)(Convert.ToSingle(numericUpDown1.Value) * 100);
}
private void numericUpDown2_ValueChanged(object sender, EventArgs e)
{
trackBar2.Value = (int)(Convert.ToSingle(numericUpDown2.Value) * 100);
}
private void comboBox1_SelectedIndexChanged(object sender, EventArgs e)
{
tssl_show.Text="加载模型:"+comboBox1.Text;
detetor.LoadWeights(weightsPath+"\\"+comboBox1.Text,labelTxt);
tssl_show.Text = "模型加载已完成!";
}
}
}
【使用步骤】
使用步骤:
(1)首先根据官方框架https://github.com/ultralytics/ultralytics安装教程安装好yolov8环境,并根据官方export命令将自己pt模型转成onnx模型
(2)使用vs2019打开sln项目,选择x64 release并且修改一些必要的参数,比如输入shape等,点击运行即可查看最后效果
特别注意如果运行报错了,请参考我的博文进行重新引用我源码的DLL:[C#]opencvsharp报错System.Memory,Version=4.0.1.2,Culture=neutral,PublicKeyToken=cc7b13fcd2ddd51"版本高于所引_未能加载文件或程序集"system.memory, version=4.0.1.2, culture-CSDN博客
【提供文件】
C#源码
yolov8n.onnx模型(不提供pytorch模型)
训练的map,P,R曲线图(在weights\results.png)
测试图片(在test_img文件夹下面)
【源码下载地址】
关注下方名片并回复【firc7】即可获取下载方式