早些時(shí)間，小編想把PID控制器優(yōu)化部分通過(guò)Py Qt來(lái)實(shí)現(xiàn)用戶界面化，不過(guò)看著窗口一堆參數(shù)，有點(diǎn)發(fā)怵：這玩意誰(shuí)用啊？

參考《PID控制器參數(shù)自動(dòng)優(yōu)化示例和比較》

后來(lái)就擱置了。 在通過(guò)Python以及YOLO來(lái)檢測(cè)圖中或者視頻中的各種物體以及人物的時(shí)候，就會(huì)考慮：全部用代碼來(lái)輸出的？怎么通過(guò)一個(gè)簡(jiǎn)單直觀的窗口界面來(lái)測(cè)試相應(yīng)的功能？我覺(jué)得要再試一下，哪怕是最簡(jiǎn)單的方式呈現(xiàn)。這便是這篇文章的目的所在。 我們通過(guò)YOLO現(xiàn)成的各種檢測(cè)模塊，實(shí)現(xiàn)：

圖中物體的識(shí)別檢測(cè)

檢出物體的區(qū)域分割

人體姿態(tài)的檢測(cè)

還有其他的檢測(cè)功能，感覺(jué)功能有重疊，就選了上面的三個(gè)功能簡(jiǎn)單地集成到一個(gè)windows的因?yàn)榉诸惖墓δ苤皇擒浖缑嬷小?以下圖片均有AI生成，但是人體姿態(tài)檢測(cè)的結(jié)果可能看不清楚輸出。

按照我們之前的慣例，先提供結(jié)果，后提供代碼。

[1]物體檢測(cè)

[2]物體分割

[3]姿態(tài)檢測(cè)

我們可以自己嘗試運(yùn)行一下。

# This program is part of a project developed by Amphenol Sensors.
#Copyright(C)2024,Leo Lu
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see .


import sys
from PyQt5.QtWidgets import (QApplication, QLabel, QPushButton, QVBoxLayout, QWidget, 
                             QFileDialog, QComboBox, QHBoxLayout)
from PyQt5.QtGui import QPixmap, QImage
from PyQt5.QtCore import Qt
import cv2
import numpy as np
import random


from ultralytics import YOLO
from ultralytics.models.yolo.pose.predict import PosePredictor


class YOLOInterface(QWidget):
    def __init__(self):
        super().__init__()


        self.initUI()
        self.model_detec = YOLO('./yolo_models/yolo11s.pt')
        self.model_seg = YOLO('./yolo_models/yolo11s-seg.pt')
        self.model_bd = './yolo_models/yolo11s-pose.pt'
        self.connections = ((2, 4), (1, 3), (10, 8), (8, 6), (6, 5), (5, 7), (7, 9), 
                            (6, 12), (12, 14), (14, 16), (5, 11), (11, 13), (13, 15))


    def initUI(self):
        self.setWindowTitle('YOLO 圖像檢測(cè)')
        self.setGeometry(100, 100, 1200, 600)
        self.setFixedSize(1200, 600)
        # 主布局
        main_layout = QVBoxLayout()


        # 圖像顯示布局
        image_layout = QHBoxLayout()


        self.image_label = QLabel(self)
        self.image_label.setFixedSize(600, 400)
        self.image_label.setAlignment(Qt.AlignCenter)
        image_layout.addWidget(self.image_label)


        self.result_label = QLabel(self)
        self.result_label.setFixedSize(600, 400)
        self.result_label.setAlignment(Qt.AlignCenter)
        image_layout.addWidget(self.result_label)


        main_layout.addLayout(image_layout)


        # 控制布局
        control_layout = QHBoxLayout()


        self.detect_button = QPushButton('選擇圖片', self)
        self.detect_button.clicked.connect(self.load_image)
        control_layout.addWidget(self.detect_button)


        self.yolo_combo = QComboBox(self)
        self.yolo_combo.addItems(['物體檢測(cè)', '物體分割', '人體姿態(tài)識(shí)別'])  # 假設(shè)在此處添加不同的YOLO任務(wù)
        control_layout.addWidget(self.yolo_combo)


        self.run_button = QPushButton('開(kāi)始檢測(cè)', self)
        self.run_button.clicked.connect(self.run_yolo)
        control_layout.addWidget(self.run_button)


        self.quit_button = QPushButton('退出', self)
        self.quit_button.clicked.connect(self.close_application)
        control_layout.addWidget(self.quit_button)


        main_layout.addLayout(control_layout)
        self.setLayout(main_layout)


    def load_image(self):
        options = QFileDialog.Options()
        file_name, _ = QFileDialog.getOpenFileName(self, "選擇圖片文件", "", "Images (*.png *.jpg *.bmp)", options=options)
        if file_name:
            self.current_image = file_name
            pixmap = QPixmap(file_name)
            scaled_pixmap = pixmap.scaled(self.image_label.size(), Qt.KeepAspectRatio)
            self.image_label.setPixmap(scaled_pixmap)


    def plot_keypoints(self, image, keypoints, line_color=(60, 179, 113), point_color=(255, 0, 0),
                   offset=(0, 0), show_idx=False):
        if keypoints is None:
            return image
        
        for data in keypoints.xy:
            if len(data) == 0:
                continue
            # Draw connections between keypoints
            for start_index, end_index in self.connections:
                start_point, end_point = data[start_index], data[end_index]
                if all(start_point[:2] > 0) and all(end_point[:2] > 0):  # Ignore invalid points
                    cv2.line(image, 
                             tuple(map(lambda v, o: int(v + o), start_point[:2], offset)), 
                             tuple(map(lambda v, o: int(v + o), end_point[:2], offset)), 
                             line_color, 2)
            # Draw keypoints
            for index, (x, y) in enumerate(data[:, :2]):
                if x > 0 or y > 0:  # Ignore invalid points
                    cv2.circle(image, 
                               (int(x + offset[0]), int(y + offset[1])), 
                               5, point_color, -1)
                    if show_idx:
                        cv2.putText(image, 
                                    str(index), 
                                    (int(x + offset[0]), int(y + offset[1])), 
                                    cv2.FONT_HERSHEY_SIMPLEX, 0.5, point_color, 1, cv2.LINE_AA)
    
        return image


    def run_yolo(self):
        if hasattr(self, 'current_image'):
            img = cv2.imread(self.current_image)
            
            # YOLO推理示例:
            task = self.yolo_combo.currentText()
            if task =='物體檢測(cè)':
                #model = self.model_detec #YOLO('yolo11s.pt')
                results = self.model_detec(img)
                
                for result in results:
                    boxes = result.boxes                # Pseudo-code; adjust based on actual library


                    for box in boxes:
                        x1, y1, x2, y2 = box.xyxy[0]    # Extracting the bounding box coordinates
                        class_name = self.model_detec.names[int(box.cls[0])]  # Using class_id to get class_name from model
                        confidence = box.conf.item()
                        cv2.rectangle(img, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
                        txt_y_pos = int(y1) - 10
                        if txt_y_pos <= 10:
                            txt_y_pos = int(y2) - 10
                        
                        class_name = class_name + " "+ "{:.2g}".format(confidence)
                        cv2.putText(img, class_name, (int(x1), txt_y_pos), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)


            elif task =='物體分割':
                #model = YOLO('yolo11s-seg.pt')
                results = self.model_seg(img)
                
                # Prepare an overlay image with the same dimensions
                overlay = img.copy()
                
                for result in results:
                    boxes = result.boxes  # Pseudo-code; adjust based on actual library
                    masks = result.masks
                    names = result.names
                    for box, mask in zip(boxes, masks):
                        for cls, contour in zip(box.cls, mask.xy):
                            class_id = cls.item()  # Get scalar from tensor
                            class_name = names[class_id]
                            #print(class_name)
                            #print(cv2.contourArea(contour))  # Calculate contour area
                            
                            # Generate a random color
                            color = [random.randint(0, 255) for _ in range(3)]
                            
                            # Fill the contour on the overlay with the random color
                            cv2.drawContours(overlay, [contour.astype(np.int32)], -1, color, thickness=cv2.FILLED)
                # Define the alpha (transparency factor)
                alpha = 0.4  # Value between 0 (transparent) and 1 (opaque)
                
                # Blend the overlay with the original image
                """
                Parameters
                overlay (src1):


                This is the first input array (image).
                In your context, this represents the overlay image, which typically contains modifications like semi-transparent masks drawn over contours.
                
                alpha:
                This is the weight of the first array (image).
                It controls the opacity of the overlay. A value closer to 1 makes the overlay more prominent, while a value closer to 0 makes it less prominent.
                
                img (src2):
                This is the second input array (image).
                It represents the original image onto which the overlay is being applied.
                
                1 - alpha (beta):
                This is the weight of the second array (image).
                Complementary to alpha, it controls the visibility of the original image. A value closer to 1 makes the original image more visible, while closer to 0 makes it less visible.
                
                0 (gamma):
                A scalar added to each sum.
                Typically set to 0 when blending for direct overlay purposes without additional brightness adjustment.
                
                img (dst):
                The destination array where the output is stored.
                It uses the same variable as the original image, implying that the blended result will overwrite this variable.
                """
                cv2.addWeighted(overlay, alpha, img, 1 - alpha, 0, img)


            elif task == '人體姿態(tài)識(shí)別':
                #model = YOLO('yolo11s-pose.pt')
                #results = model(img)
                
                overrides_Body_pose = {
                    "task": "pose",
                    "mode": "predict",
                    "model": self.model_bd,        #'yolo11s-pose.pt'
                    "verbose": False,
                    "classes": [0],
                    "iou": 0.5,
                    "conf": 0.3
                    }


                predictor_ren_pose = PosePredictor(overrides=overrides_Body_pose)
                pose_ren = predictor_ren_pose(img)[0]
                
                img = self.plot_keypoints(img, pose_ren.keypoints)
                
            # Convert the image to a format suitable for PyQt
            height, width, channel = img.shape
            bytes_per_line = 3 * width
            q_image = QImage(img.data, width, height, bytes_per_line, QImage.Format_RGB888).rgbSwapped()
            pixmap = QPixmap.fromImage(q_image)
            self.display_results(pixmap)


    def display_results(self, pixmap):
        scaled_pixmap = pixmap.scaled(self.result_label.size(), Qt.KeepAspectRatio)
        self.result_label.setPixmap(scaled_pixmap)


    def close_application(self):
        self.close()


if __name__ == '__main__':
    app = QApplication(sys.argv)
    ex = YOLOInterface()
    ex.show()
    sys.exit(app.exec_())

代碼中涉及到的python軟件包，需要額外安裝，模型文件用的是YOLO現(xiàn)成的。小編把這個(gè)軟件用打包工具生成的exe文件，因?yàn)槿晾ǖ能浖啵?jiǎn)單的這些功能，打包后的文件壓縮后的大小竟高達(dá)近900MB（分割壓縮工具要收費(fèi)咯），而很多平臺(tái)不支持這么大的文件上傳，所以這里也無(wú)法提供打包后的文件下載鏈接。

視線收回，我們心曠神怡一下。感謝AI的提供。

是否在懷疑這個(gè)世界的真實(shí)性？用傳感器來(lái)檢測(cè)一下吧，這個(gè)世界的冷熱、正負(fù)壓，振動(dòng)，周圍的空氣......可以負(fù)責(zé)任地告訴各位，我們還是碳基生物。