使用OpenPifPaf调用摄像头检测姿态时视频无输出且报错

问题分析 #

你遇到的AttributeError: 'NoneType' object has no attribute '__array_interface__'和cv2断言错误，根源有两个：

• 摄像头帧读取可能失败，返回None的帧传入模型导致报错
• AnnotationPainter.annotations()返回的是matplotlib绘图对象，并非OpenCV可直接显示的numpy数组，所以cv2.imshow()无法处理

修复步骤 #

1. 校验摄像头帧读取结果 #

在处理帧之前，先判断是否成功读取到帧，避免None传入模型：

_, frame = video.read()
if frame is None:
    break  # 读取失败时退出循环

2. 转换matplotlib画布为OpenCV格式 #

OpenPifPaf的Canvas基于matplotlib，需要把画布内容转换成BGR格式的numpy数组（OpenCV默认格式）：

with openpifpaf.show.Canvas.image(frame) as ax:
    annotation_painter.annotations(ax, predictions)
    # 从matplotlib画布提取numpy数组
    fig = ax.figure
    fig.canvas.draw()
    # 转换为numpy数组并调整颜色通道
    img_array = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
    img_array = img_array.reshape(fig.canvas.get_width_height()[::-1] + (3,))
    img_array = cv2.cvtColor(img_array, cv2.COLOR_RGB2BGR)
    cv2.imshow('frame2', img_array)

3. 优化代码结构 #

添加摄像头初始化校验，避免循环内不必要的操作，确保资源正常释放。

修改后的完整代码 #

import openpifpaf
import cv2
import numpy as np

VIDEO_LINK = 0  # 0对应默认摄像头
video = cv2.VideoCapture(VIDEO_LINK)
# 检查摄像头是否成功打开
if not video.isOpened():
    print("无法打开摄像头")
    exit()

predictor = openpifpaf.Predictor(checkpoint='shufflenetv2k16')
annotation_painter = openpifpaf.show.AnnotationPainter()

while True:
    ret, frame = video.read()
    if not ret or frame is None:
        break  # 读取失败或视频结束时退出
    
    # 模型预测
    predictions, _, _ = predictor.numpy_image(frame)
    
    # 可视化处理
    with openpifpaf.show.Canvas.image(frame) as ax:
        annotation_painter.annotations(ax, predictions)
        # 转换画布为OpenCV可用格式
        fig = ax.figure
        fig.canvas.draw()
        img_rgb = np.frombuffer(fig.canvas.tostring_rgb(), dtype=np.uint8)
        img_rgb = img_rgb.reshape(fig.canvas.get_width_height()[::-1] + (3,))
        img_bgr = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2BGR)
        
        cv2.imshow('Pose Detection', img_bgr)
    
    # 按q键退出
    if cv2.waitKey(33) & 0xFF == ord('q'):
        break

video.release()
cv2.destroyAllWindows()

额外注意事项 #

• 确保摄像头权限正常，没有被其他程序占用
• 如果仍有报错，尝试更新OpenPifPaf和OpenCV到最新版本：

  pip install --upgrade openpifpaf opencv-python
  

内容的提问来源于stack exchange，提问作者JSVJ