---

前言

基于YOLOv5目标检测算法，设计GUI软件，通过ONNX推理框架部署到树莓派上运行。
工程已上传gitee，点击下载即可。

---

一、树莓派系统环境配置

1.树莓派4B

内容	参数
运行内存	4GB
TF卡	64GB
操作系统	Linux (Raspberry Pi OS with Desktop)
系统位数	32位

树莓派Linux操作系统可以从树莓派官方或镜像源进行下载，系统烧录过程（略）。

2.树莓派系统环境依赖安装

opencv安装可参考流浪猫博主分享的《超简单教你在树莓派上安装opencv(一)》
numpy、onnxruntime、pyqt5安装使用pip进行安装即可。

pip install numpy / onnxruntime / pyqt5

二、设计GUI界面

1.打开Qtdesigner设计GUI界面

简单Demo：
界面元素：
一个QLabel：用于视频显示
两个pushButton（按钮）：打开和关闭摄像头
设计完成后，点击保存，生成ui文件

2.ui文件格式转换

利用PyQt5的PyUIC工具将其ui文件转换成py文件，如下图：

demo1.py：

# -*- coding: utf-8 -*-

# Form implementation generated from reading ui file 'demo1.ui'
#
# Created by: PyQt5 UI code generator 5.15.4
#
# WARNING: Any manual changes made to this file will be lost when pyuic5 is
# run again.  Do not edit this file unless you know what you are doing.


from PyQt5 import QtCore, QtGui, QtWidgets


class Ui_MainWindow(object):
    def setupUi(self, MainWindow):
        MainWindow.setObjectName("MainWindow")
        MainWindow.resize(794, 600)
        self.centralwidget = QtWidgets.QWidget(MainWindow)
        self.centralwidget.setObjectName("centralwidget")
        self.horizontalLayout = QtWidgets.QHBoxLayout(self.centralwidget)
        self.horizontalLayout.setObjectName("horizontalLayout")
        self.verticalLayout = QtWidgets.QVBoxLayout()
        self.verticalLayout.setSizeConstraint(QtWidgets.QLayout.SetNoConstraint)
        self.verticalLayout.setContentsMargins(10, 0, 10, -1)
        self.verticalLayout.setSpacing(7)
        self.verticalLayout.setObjectName("verticalLayout")
        self.pushButton = QtWidgets.QPushButton(self.centralwidget)
        self.pushButton.setObjectName("pushButton")
        self.verticalLayout.addWidget(self.pushButton)
        self.pushButton_2 = QtWidgets.QPushButton(self.centralwidget)
        self.pushButton_2.setObjectName("pushButton_2")
        self.verticalLayout.addWidget(self.pushButton_2)
        self.horizontalLayout.addLayout(self.verticalLayout)
        self.label = QtWidgets.QLabel(self.centralwidget)
        self.label.setText("")
        self.label.setObjectName("label")
        self.horizontalLayout.addWidget(self.label)
        self.horizontalLayout.setStretch(0, 2)
        self.horizontalLayout.setStretch(1, 5)
        MainWindow.setCentralWidget(self.centralwidget)
        self.menubar = QtWidgets.QMenuBar(MainWindow)
        self.menubar.setGeometry(QtCore.QRect(0, 0, 794, 26))
        self.menubar.setObjectName("menubar")
        MainWindow.setMenuBar(self.menubar)
        self.toolBar = QtWidgets.QToolBar(MainWindow)
        self.toolBar.setObjectName("toolBar")
        MainWindow.addToolBar(QtCore.Qt.TopToolBarArea, self.toolBar)

        self.retranslateUi(MainWindow)
        QtCore.QMetaObject.connectSlotsByName(MainWindow)

    def retranslateUi(self, MainWindow):
        _translate = QtCore.QCoreApplication.translate
        MainWindow.setWindowTitle(_translate("MainWindow", "MainWindow"))
        self.pushButton.setText(_translate("MainWindow", "打开摄像头"))
        self.pushButton_2.setText(_translate("MainWindow", "关闭摄像头"))
        self.toolBar.setWindowTitle(_translate("MainWindow", "toolBar"))

3.编写运行代码

demo1_main.py：

# here put the import lib
import math
import sys

from PyQt5 import QtGui, QtWidgets

from general import plot_one_box, infer_img
from demo1 import Ui_MainWindow

import numpy as np
import cv2
# import time
# from random import uniform
from PyQt5.Qt import *
import onnxruntime as ort


class Open_Camera(QtWidgets.QMainWindow, Ui_MainWindow):
    def __init__(self):
        super(Open_Camera, self).__init__()
        self.setupUi(self)  # 创建窗体对象
        self.setWindowTitle('yolov5目标检测demo')
        self.init()

        self.openfile_name_model = 'best.onnx'  # 模型名称

        # self.so = ort.SessionOptions()  # 树莓派上保留以下两段代码,注释下面那行代码
        # self.net = ort.InferenceSession(self.openfile_name_model, self.so)

        # ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']
        # CUDAExecutionProvider: GPU推理；
        self.net = ort.InferenceSession(self.openfile_name_model,
                                        providers=['CUDAExecutionProvider'])  # 在树莓派上这里不需指定推理设备

        # 标签字典
        self.dic_labels = {0: 'person'}

        # 模型参数
        self.model_h = 320
        self.model_w = 320

    def init(self):
        # self.label.setScaledContents(True)  # 图片自适应
        # 定时器让其定时读取显示图片
        self.camera_timer = QTimer()
        self.camera_timer.timeout.connect(self.show_image)
        # 打开摄像头
        self.pushButton.clicked.connect(self.open_camera)
        # 关闭摄像头
        self.pushButton_2.clicked.connect(self.close_camera)

    '''开启摄像头'''
    def open_camera(self):
        self.cap = cv2.VideoCapture(0)  # 初始化摄像头
        self.camera_timer.start(40)  # 每40毫秒读取一次，即刷新率为25帧
        self.show_image()

    '''关闭摄像头'''
    def close_camera(self):
        self.cap.release()  # 释放摄像头
        self.label.clear()  # 清除label组件上的图片
        self.camera_timer.stop()  # 停止读取

    '''显示图片'''
    def show_image(self):
        flag, self.image = self.cap.read()  # 从视频流中读取图片

        if flag:
            # image_show = cv2.resize(self.image, (1280, 720))  # 把读到的帧的大小重新设置为 600*360
            image_show = self.image
            width, height = image_show.shape[:2]  # 行:宽，列:高
            # image_show = cv2.cvtColor(image_show, cv2.COLOR_BGR2RGB)  # opencv读的通道是BGR,要转成RGB
            # image_show = cv2.flip(image_show, 1)  # 水平翻转，因为摄像头拍的是镜像的。 用相机不用翻转
            # start 图片检测
            det_boxes, scores, ids = infer_img(image_show, self.net, self.model_h, self.model_w,
                                               thred_nms=0.5, thred_cond=0.75)
            # image_show = self.image
            for box, score, id in zip(det_boxes, scores, ids):
                label = '%s:%.2f' % (self.dic_labels[id], score)
                plot_one_box(box.astype(np.int16), image_show, color=(255, 0, 0), label=label, line_thickness=None)

        image_show = cv2.cvtColor(image_show, cv2.COLOR_BGR2RGB)  # opencv读的通道是BGR,要转成RGB
        # 把读取到的视频数据变成QImage形式(图片数据、高、宽、RGB颜色空间，三个通道各有2**8=256种颜色)
        self.showImage = QtGui.QImage(image_show.data, height, width, QImage.Format_RGB888)
        self.label.setPixmap(QPixmap.fromImage(self.showImage))  # 往显示视频的Label里显示QImage
        self.label.setScaledContents(True)  # 图片自适应


if __name__ == '__main__':
    app = QtWidgets.QApplication(sys.argv)
    ui = Open_Camera()
    ui.show()
    sys.exit(app.exec_())

general.py：

import numpy as np
import cv2
import random


# 标注目标
def plot_one_box(x, img, color=None, label=None, line_thickness=None):
    """
    description: Plots one bounding box on image img,
                 this function comes from YoLov5 project.
    param:
        x:      a box likes [x1,y1,x2,y2]
        img:    a opencv image object
        color:  color to draw rectangle, such as (0,255,0)
        label:  str
        line_thickness: int
    return:
        no return
    """
    tl = (
            line_thickness or round(0.002 * (img.shape[0] + img.shape[1]) / 2) + 1
    )  # line/font thickness
    color = color or [random.randint(0, 255) for _ in range(3)]
    c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
    cv2.rectangle(img, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
    if label:
        tf = max(tl - 1, 1)  # font thickness
        t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
        c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
        cv2.rectangle(img, c1, c2, color, -1, cv2.LINE_AA)  # filled
        cv2.putText(
            img,
            label,
            (c1[0], c1[1] - 2),
            0,
            tl / 3,
            [225, 255, 255],
            thickness=tf,
            lineType=cv2.LINE_AA,
        )


# 极大值抑制
def post_process_opencv(outputs, model_h, model_w, img_h, img_w, thred_nms, thred_cond):
    conf = outputs[:, 4].tolist()
    c_x = outputs[:, 0] / model_w * img_w
    c_y = outputs[:, 1] / model_h * img_h
    w = outputs[:, 2] / model_w * img_w
    h = outputs[:, 3] / model_h * img_h
    p_cls = outputs[:, 5:]
    if len(p_cls.shape) == 1:
        p_cls = np.expand_dims(p_cls, 1)
    cls_id = np.argmax(p_cls, axis=1)

    p_x1 = np.expand_dims(c_x - w / 2, -1)
    p_y1 = np.expand_dims(c_y - h / 2, -1)
    p_x2 = np.expand_dims(c_x + w / 2, -1)
    p_y2 = np.expand_dims(c_y + h / 2, -1)
    areas = np.concatenate((p_x1, p_y1, p_x2, p_y2), axis=-1)

    areas = areas.tolist()
    ids = cv2.dnn.NMSBoxes(areas, conf, thred_cond, thred_nms)
    if len(ids) > 0:
        return np.array(areas)[ids], np.array(conf)[ids], cls_id[ids]
    else:
        return [], [], []


# 推理
def infer_img(img0, net, model_h, model_w, thred_nms=0.4, thred_cond=0.5):
    # 图像预处理
    img = cv2.resize(img0, [model_w, model_h], interpolation=cv2.INTER_AREA)  # 缩放
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  # 格式转换
    img = img.astype(np.float32) / 255.0  # 归一化
    blob = np.expand_dims(np.transpose(img, (2, 0, 1)), axis=0)  # 维度转换

    # 模型推理
    outs = net.run(None, {net.get_inputs()[0].name: blob})[0].squeeze(axis=0)

    # 输出坐标矫正
    # outs = cal_outputs(outs, nl, na, model_w, model_h, anchor_grid, stride)

    # 检测框计算
    img_h, img_w, _ = np.shape(img0)
    boxes, confs, ids = post_process_opencv(outs, model_h, model_w, img_h, img_w, thred_nms, thred_cond)

    return boxes, confs, ids

4.准备权重文件

下载yolov5n.pt并将其转换成yolov5.onnx，放在工程同目录下。

三、树莓派运行效果

1.将文件传到树莓派

2.修改代码demo1_main.py代码

3.运行代码

在命令行窗口切换到工程文件路径下，运行代码

python demo1_main.py

运行效果：

yolov5n模型在树莓派上部署，FPS大约为5-6FPS。

四、UI软件打包

如果你想将qt文件进行打包输出exe，可参考pyqt打包输出exe。
选中demo1_main.py进行输出，应该最后要将general.py改成general.pyc放在输出的exe同路径下。

---

总结

将YOLOv5通过ONNX推理引擎部署在树莓派上，同时利用PyQt技术设计了简单的GUI界面。（仅个人记录分享）