怎樣使用YOLOv8構(gòu)建目標(biāo)計(jì)數(shù)GUI呢？

標(biāo)檢測(cè)是在圖片或視頻中定位物體的過(guò)程。其中一個(gè)著名的目標(biāo)檢測(cè)框架是YOLO（You Only Look Once）。在某些情況下，我們不僅需要定位圖片或視頻中的物體，還需要了解每個(gè)物體的運(yùn)動(dòng)，就像在這篇文章中，我們想要計(jì)數(shù)通過(guò)某些位置的物體。這就是我們需要不僅僅是檢測(cè)的地方，我們需要稱為目標(biāo)跟蹤的東西的地方。

此外，大多數(shù)情況下，我們通常只是通過(guò)執(zhí)行腳本或命令行中的代碼來(lái)使用它，如果我們可以將我們的模型部署為GUI或應(yīng)用程序，以便用戶更輕松地使用它，那將更好。在這篇文章中，將分享如何使用框架YOLOv8和PySimpleGUI構(gòu)建一個(gè)目標(biāo)計(jì)數(shù)GUI。

什么是YOLOv8？

YOLOv8，即You Only Look Once version 8，是由UItralytics開(kāi)發(fā)的圖像處理框架，YOLOv8可以執(zhí)行目標(biāo)檢測(cè)和跟蹤、實(shí)例分割、圖像分類和姿態(tài)估計(jì)任務(wù)。

YOLOv8的用法

YOLOv8與先前版本的性能比較

使用YOLOv8進(jìn)行目標(biāo)跟蹤

有很多方法可以使用YOLOv8進(jìn)行目標(biāo)跟蹤。我喜歡Python腳本方法，因?yàn)槲铱梢杂懈嗟目刂茩?quán)，使用此方法需要執(zhí)行以下幾個(gè)步驟，您還可以從Ultralytics的文檔中查看此鏈接。

1. 創(chuàng)建虛擬環(huán)境

python -m venv yologui

2. 激活虛擬環(huán)境

yologuiScriptsactivate

3. 安裝YOLOv8的依賴項(xiàng)

pip install -r https://raw.githubusercontent.com/ultralytics/ultralytics/main/requirements.txt

4. 安裝YOLOv8

pip install ultralytics

5. 創(chuàng)建一個(gè)yolo腳本

# https://docs.ultralytics.com/modes/track/#python-examples
import cv2
from ultralytics import YOLO


# Load the YOLOv8 model
model = YOLO('yolov8n.pt')


# Open the camera
cap = cv2.VideoCapture(0)


# Loop through the video frames
while cap.isOpened():
    # Read a frame from the video
    success, frame = cap.read()


    if success:
        # Run YOLOv8 tracking on the frame, persisting tracks between frames
        results = model.track(frame, persist=True)
        # Visualize the results on the frame
        annotated_frame = results[0].plot()
        # Display the annotated frame
        cv2.imshow("YOLOv8 Tracking", annotated_frame)
        # Break the loop if 'q' is pressed
        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
    else:
        # Break the loop if the end of the video is reached
        break


# Release the video capture object and close the display window
cap.release()
cv2.destroyAllWindows()

執(zhí)行腳本，您應(yīng)該可以得到由YOLOv8進(jìn)行的目標(biāo)跟蹤：

python yolo.py

這是結(jié)果，當(dāng)滿意時(shí)按“Q”退出。

目標(biāo)追蹤結(jié)果

您可以看到我們的腳本可以記住左側(cè)圖片中的泰迪熊與右側(cè)圖片中的泰迪熊是相同的。實(shí)際上，他是一只狗....

添加目標(biāo)跟蹤的計(jì)數(shù)邏輯

目標(biāo)跟蹤為我們提供了比目標(biāo)檢測(cè)更多的選項(xiàng)，其中之一是我們可以使用它來(lái)計(jì)算進(jìn)入或離開(kāi)框架中某個(gè)區(qū)域或線的獨(dú)特目標(biāo)的數(shù)量。

跟蹤物體A和B的運(yùn)動(dòng)

要計(jì)算通過(guò)我們的線左右經(jīng)過(guò)的物體數(shù)量，我們需要從我們的目標(biāo)跟蹤腳本中提取目標(biāo)的id和位置。Ultralytics的模型已經(jīng)為我們提供了該屬性，通過(guò)使用以下命令來(lái)處理跟蹤結(jié)果。

# Using Model
results= model.track(frame, persist=True)


# Extrack result
result = results[0].cpu().boxes
detect_id = result.id.tolist() if result.id != None else []
detect_xyxy = result.xyxy.tolist() if result.xyxy != None else []
frame_counting_buffer = dict(zip(detect_id, detect_xyxy))

因此，我們可以獲取每個(gè)目標(biāo)的位置，接下來(lái)是進(jìn)行一些計(jì)算，以了解目標(biāo)是從右到左還是從左到右經(jīng)過(guò)。在這種情況下，我只關(guān)注目標(biāo)的中心和X軸。以下是在X軸上進(jìn)行目標(biāo)計(jì)數(shù)的完整腳本。

# https://docs.ultralytics.com/modes/track/#python-examples
import cv2
from ultralytics import YOLO


# Load the YOLOv8 model
model = YOLO('yolov8n.pt')


# Counting config
line_position = 50
text_size = 30
text_x_position = 50
text_y_position = 0


# Open the camera
cap = cv2.VideoCapture(0)


# Get Camera Parameter
width  = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)


# Counting prep
x_line= line_position*width/100
pt1 = (int(x_line), 0)
pt2 = (int(x_line), int(height))
counting_buffer = {}
counting_result = {'left-to-right' : 0, 'right-to-left' : 0}


# Loop through the video frames
while cap.isOpened():
    # Read a frame from the video
    success, frame = cap.read()
    if success:
        # Run YOLOv8 tracking on the frame, persisting tracks between frames
        results = model.track(frame, persist=True, verbose=False)


        # Visualize the results on the frame
        annotated_frame = results[0].plot()


        # Get Data for counting
        result = results[0].cpu().boxes
        detect_id = result.id.tolist() if result.id != None else []
        detect_xyxy = result.xyxy.tolist() if result.xyxy != None else []
        frame_counting_buffer = dict(zip(detect_id, detect_xyxy))
        # Process
        for i in frame_counting_buffer :
            # Prep count buffer
            counting_buffer[i] = counting_buffer.get(i,[])
            if len(counting_buffer[i]) >= 2 : counting_buffer[i] = counting_buffer[i][-1:]
            # Append avg x axis to buffer
            avg_x = (frame_counting_buffer[i][0] + frame_counting_buffer[i][2])/2
            counting_buffer[i].append(avg_x)
            # Count logic
            if len(counting_buffer[i]) >= 2 : 
                if (counting_buffer[i][0] > x_line) & (counting_buffer[i][1] < x_line) :
                    counting_result['right-to-left'] += 1
                elif (counting_buffer[i][0] < x_line) & (counting_buffer[i][1] > x_line) :
                    counting_result['left-to-right'] += 1


        # Create Line
        cv2.line(annotated_frame, pt1= pt1, pt2= pt2 , color= (0,0,255), thickness= 2) 
        # Put Counting to picture
        text_position = text_y_position
        for i in counting_result : 
            text_position += text_size
            info_text = f"{i} : {counting_result[i]}"
            annotated_frame = cv2.putText(annotated_frame
                                          , info_text
                                          , (int(width*text_x_position/100), text_position)
                                          , cv2.FONT_HERSHEY_SIMPLEX
                                          , 1, (0,0,255), 1, cv2.LINE_AA)


        # Display the annotated frame
        cv2.imshow("YOLOv8 Tracking", annotated_frame)


        # Break the loop if 'q' is pressed
        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
    else:
        # Break the loop if the end of the video is reached
        break


# Release the video capture object and close the display window
cap.release()
cv2.destroyAllWindows()

然后，我們將得到一個(gè)如圖所示的目標(biāo)計(jì)數(shù)腳本，當(dāng)滿意時(shí)按Q。

什么是PySimpleGUI？

PySimpleGUI是一個(gè)用于制作GUI的Python庫(kù)，具有跨平臺(tái)性且非常簡(jiǎn)單易用。

從PySimpleGUI的頁(yè)面獲取圖片

GUI的框架

創(chuàng)建GUI時(shí)，有一些比編碼更重要的東西，即設(shè)計(jì)階段。我們?cè)诰幋a之前應(yīng)該有一些草圖設(shè)計(jì)。在這種情況下，我畫(huà)了一個(gè)目標(biāo)計(jì)數(shù)GUI的初步設(shè)計(jì)，就像這張圖片一樣。

GUI初步設(shè)計(jì)

設(shè)計(jì)階段后，接下來(lái)就是編碼：

1. 安裝PySimpleGUI。

pip install pysimplegui

2. 創(chuàng)建一個(gè)GUI腳本

# https://www.pysimplegui.org/en/latest/#jump-start
import PySimpleGUI as sg


# Create Layouy of the GUI
layout = [  
            [sg.Text('GUI Object Counting with Yolo V8')],
            [sg.Text('Enter Model Name', size= (15)), sg.InputText(key='model_name')],
            [sg.Text('X-line for Counting', size= (15)), sg.InputText(key= 'line_position')],
            [sg.Button('Run'), sg.Button('Stop'), sg.Button('Close')],
            [sg.Text('PICTURE')],
            [sg.Text('Left-to-Right', size= (15), key='out1'), sg.Text("0", key='out1-v')],
            [sg.Text('Right-to-Left', size= (15), key='out2'), sg.Text("0", key='out2-v')]
            ]


# Create the Window
window = sg.Window('GUIYoloV8-Counting', layout)
# Event Loop to process "events"
while True:
    event, values = window.read()
    # When press Run
    if event == 'Run' : 
        print(values)
    # When close window or press Close
    if event in (sg.WIN_CLOSED, 'Close'): break


# Close window
window.close()

正如您所見(jiàn)，PySimpleGUI的腳本非常簡(jiǎn)單。之后，我們所要做的就是運(yùn)行此GUI腳本。

python gui.py

之后，您應(yīng)該會(huì)得到一個(gè)像圖中這樣的簡(jiǎn)單GUI，嘗試玩耍并在滿意時(shí)關(guān)閉。

合并

在我們有了目標(biāo)計(jì)數(shù)和GUI框架之后，是時(shí)候?qū)⑺鼈兒喜⒌揭黄鹆?。我們可以將兩個(gè)腳本合并成一個(gè)名為main.py的新腳本，如下所示：

# https://www.pysimplegui.org/en/latest/#jump-start
import PySimpleGUI as sg
import cv2
# https://docs.ultralytics.com/modes/track/#python-examples
from ultralytics import YOLO


# Create Layouy of the GUI
layout = [  
            [sg.Text('GUI Object Counting with Yolo V8')],
            [sg.Text('Enter Model Name', size= (15)), sg.InputText(key='model_name')],
            [sg.Text('X-line for Counting', size= (15)), sg.InputText(key= 'line_position')],
            [sg.Button('Run'), sg.Button('Stop'), sg.Button('Close')],
            [sg.Image(filename='', key='image')],
            [sg.Text('Right-to-Left', size= (15), key='out1'), sg.Text("0", key='right-to-left')],
            [sg.Text('Left-to-Right', size= (15), key='out2'), sg.Text("0", key='left-to-right')]
            ]


# Create the Window
window = sg.Window('GUIYoloV8-Counting', layout)
run_model, verbose = False, False
# Event Loop to process "events"
while True:
    event, values = window.read(timeout=0)
    # When press Run
    if event == 'Run' : 
        # Load the YOLOv8 model
        model = YOLO(values['model_name'])
        # Counting config
        line_position = 50
        text_size = 30
        text_x_position = 50
        text_y_position = 0
        # Open the camera
        cap = cv2.VideoCapture(0)
        # Get Camera Parameter
        width  = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
        height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
        # Counting prep
        x_line= line_position*width/100
        pt1 = (int(x_line), 0)
        pt2 = (int(x_line), int(height))
        counting_buffer = {}
        counting_result = {'left-to-right' : 0, 'right-to-left' : 0}
        # Run Signal
        run_model = True
    # When close window or press Close
    elif event in ('Stop', sg.WIN_CLOSED, 'Close'):
        if run_model : 
            run_model = False # Stop running
            cap.release() # Release video
            if event != sg.WIN_CLOSED : window['image'].update(filename='') # Destroy picture
        # When close window or press Close
        if event in (sg.WIN_CLOSED, 'Close'): break
    # Run Model
    if run_model : 
        # Read a frame from the video
        success, frame = cap.read()
        if success:
            # Run YOLOv8 tracking on the frame, persisting tracks between frames
            results = model.track(frame
                                  , persist=True
                                  , verbose=False
                                  )


            # Visualize the results on the frame
            annotated_frame = results[0].plot()


            # Get Data for counting
            result = results[0].cpu().boxes
            detect_id = result.id.tolist() if result.id != None else []
            detect_xyxy = result.xyxy.tolist() if result.xyxy != None else []
            frame_counting_buffer = dict(zip(detect_id, detect_xyxy))
            # Process
            for i in frame_counting_buffer :
                # Prep count buffer
                counting_buffer[i] = counting_buffer.get(i,[])
                if len(counting_buffer[i]) >= 2 : counting_buffer[i] = counting_buffer[i][-1:]
                # Append avg x axis to buffer
                avg_x = (frame_counting_buffer[i][0] + frame_counting_buffer[i][2])/2
                counting_buffer[i].append(avg_x)
                # Count logic
                if len(counting_buffer[i]) >= 2 : 
                    if (counting_buffer[i][0] > x_line) & (counting_buffer[i][1] < x_line) :
                        counting_result['right-to-left'] += 1
                    elif (counting_buffer[i][0] < x_line) & (counting_buffer[i][1] > x_line) :
                        counting_result['left-to-right'] += 1
            # Create Line
            cv2.line(annotated_frame, pt1= pt1, pt2= pt2 , color= (0,0,255), thickness= 2) 
            # Put Counting to picture
            text_position = text_y_position
            for i in counting_result : 
                text_position += text_size
                info_text = f"{i} : {counting_result[i]}"
                annotated_frame = cv2.putText(  annotated_frame
                                            , info_text
                                            , (int(width*text_x_position/100)
                                            , text_position)
                                            , cv2.FONT_HERSHEY_SIMPLEX
                                            , 1
                                            , (0,0,255)
                                            , 1
                                            , cv2.LINE_AA)
            # Show Image
            imgbytes = cv2.imencode('.png', annotated_frame)[1].tobytes()
            window['image'].update(data=imgbytes)
            window['right-to-left'].update(str(counting_result['right-to-left']))
            window['left-to-right'].update(str(counting_result['left-to-right']))
        else: 
            # Break the loop if not read
            cap.release()
            run_model = False
        
# Close window
window.close()

然后，我們可以運(yùn)行主腳本。

python main.py

然后，我們就可以得到以下圖片所示的目標(biāo)計(jì)數(shù)GUI。這一次，我將使用一個(gè)定制模型來(lái)識(shí)別我從Ultralytics HUB創(chuàng)建的Pompompurin。

初始階段

按下“RUN”后，我的新模型知道它現(xiàn)在是一只狗... 接下來(lái)，為了讓我們的用戶更輕松地使用，我們可以創(chuàng)建一個(gè)包含以下代碼的“cmd”文件。

python main.py
pause

審核編輯：劉清