如何用YOLOv8在屏幕截图上直接显示检测边界框？

你的代码已经成功完成了屏幕捕获和目标检测（输出显示检测到3个物体），但缺少检测结果可视化的步骤。以下是两种实现边界框显示的方案：

方案1：使用YOLO内置可视化方法（最简单） #

直接利用Ultralytics YOLO提供的plot()方法生成带边界框的图像，再用OpenCV显示：

from ultralytics import YOLO
import pyautogui
import cv2
import numpy as np

# Load the model
model_path = 'C:/Users/toto/Desktop/DETECT/best.pt'
model = YOLO(model_path)

# Set the confidence threshold
conf_threshold = 0.4

# Capture the entire screen
screen_image = pyautogui.screenshot()
screen_image = cv2.cvtColor(np.array(screen_image), cv2.COLOR_RGB2BGR)

# Perform object detection on the captured screen image
results = model(screen_image, conf=conf_threshold)  # 传入置信度过滤结果

# 生成带边界框、类别和置信度的可视化图像
annotated_image = results[0].plot()

# 显示检测结果窗口
cv2.imshow("Screen Detection Result", annotated_image)
cv2.waitKey(0)  # 按任意键关闭窗口
cv2.destroyAllWindows()

方案2：手动绘制边界框（自定义样式） #

如果需要自定义边界框颜色、文本样式等，可手动提取检测结果并绘制：

from ultralytics import YOLO
import pyautogui
import cv2
import numpy as np

# Load the model
model_path = 'C:/Users/toto/Desktop/DETECT/best.pt'
model = YOLO(model_path)

# Set the confidence threshold
conf_threshold = 0.4

# Capture the entire screen
screen_image = pyautogui.screenshot()
screen_image = cv2.cvtColor(np.array(screen_image), cv2.COLOR_RGB2BGR)

# Perform object detection on the captured screen image
results = model(screen_image, conf=conf_threshold)

# 遍历检测结果并手动绘制
for result in results:
    boxes = result.boxes
    for box in boxes:
        # 提取边界框坐标（xyxy格式：左上角x1,y1，右下角x2,y2）
        x1, y1, x2, y2 = map(int, box.xyxy[0])
        # 提取类别名称和置信度
        cls_idx = int(box.cls[0])
        confidence = box.conf[0]
        class_name = model.names[cls_idx]
        
        # 绘制绿色边界框（线宽2）
        cv2.rectangle(screen_image, (x1, y1), (x2, y2), (0, 255, 0), 2)
        # 绘制类别+置信度文本
        text = f"{class_name}: {confidence:.2f}"
        cv2.putText(screen_image, text, (x1, y1-10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)

# 显示检测结果窗口
cv2.imshow("Screen Detection Result", screen_image)
cv2.waitKey(0)
cv2.destroyAllWindows()

关键说明 #

• 调用model()时传入conf=conf_threshold，确保只保留置信度达标的检测结果；
• cv2.waitKey(0)会让窗口保持打开，直到按下任意键关闭；若需自动关闭，可替换为cv2.waitKey(3000)（3秒后关闭）；
• 手动绘制方案中，可修改边界框颜色（如(0,0,255)为红色）、线宽、文本字体等参数，实现个性化可视化。

内容的提问来源于stack exchange，提问作者shiny