使用OpenCV保存FLIR相机图像时，Matplotlib弹出保存图表弹窗问题

环境信息 #

• OpenCV 4.7
• Matplotlib 3.7.1
• Spinnaker-Python 3.0.0.118
• Python 3.10
• Windows 10 x64

核心问题 #

同时用OpenCV显示热成像图、Matplotlib同步绘制直方图时，按s键调用cv2.imwrite()会触发Matplotlib的保存弹窗，最终保存的是直方图而非目标热成像图；移除Matplotlib后功能恢复正常。

原因分析 #

Matplotlib默认将s键绑定为“保存当前Figure”的快捷键，且当Matplotlib窗口处于焦点时，按键事件会被Matplotlib的事件循环捕获，而非OpenCV的cv2.waitKey()。这就导致按s时触发的是Matplotlib的保存逻辑，而非你编写的OpenCV保存代码。

解决方案 #

方案1：禁用Matplotlib的s快捷键 #

在创建Matplotlib Figure后，断开默认的按键处理逻辑，避免s被Matplotlib拦截：

fig = plt.figure(1)
fig.canvas.mpl_connect('close_event', handle_close)
# 禁用默认按键处理器，阻止's'触发保存弹窗
fig.canvas.mpl_disconnect(fig.canvas.manager.key_press_handler_id)

方案2：让OpenCV窗口始终置顶 #

强制OpenCV窗口保持在最上层，确保按键事件优先被OpenCV捕获：

cv2.imshow("FLIR", image_data)
# 设置窗口置顶
cv2.setWindowProperty("FLIR", cv2.WND_PROP_TOPMOST, 1)

方案3：优化Matplotlib绘图逻辑（推荐） #

避免每次循环重绘整个Figure，改为更新已有直方图数据，既提升性能又减少事件冲突：

# 初始化阶段创建直方图对象
plt.ion()
fig, ax = plt.subplots(num=1)
fig.canvas.mpl_connect('close_event', handle_close)
fig.canvas.mpl_disconnect(fig.canvas.manager.key_press_handler_id)

# 提前获取第一帧数据初始化直方图
image_result = cam.GetNextImage(1000)
if not image_result.IsIncomplete():
    image_data = image_result.GetNDArray()
    image_data = cv2.rotate(image_data, cv2.ROTATE_90_COUNTERCLOCKWISE)
    image_data = cv2.flip(image_data, 1)
    count, bins = np.histogram(image_data.flatten())
    line, = ax.stairs(count, bins)
    image_result.Release()

# 循环内仅更新数据
count, bins = np.histogram(image_data.flatten())
line.set_data(bins[:-1], count)
ax.relim()
ax.autoscale_view()
plt.draw()
plt.pause(0.001)

完整修改示例 #

import PySpin
import cv2
import numpy as np
import matplotlib.pyplot as plt
import keyboard

def handle_close(event):
    global continue_recording
    continue_recording = False

continue_recording = True

# 假设已完成相机初始化（如枚举、选择、初始化相机）
# cam = ... 

cam.BeginAcquisition()
print('Acquiring images...')
print('Press enter to close the program..')

# 初始化Matplotlib并禁用默认's'快捷键
plt.ion()
fig, ax = plt.subplots(num=1)
fig.canvas.mpl_connect('close_event', handle_close)
fig.canvas.mpl_disconnect(fig.canvas.manager.key_press_handler_id)

acq_count = 0 

# 初始化直方图
image_result = cam.GetNextImage(1000)
if not image_result.IsIncomplete():
    image_data = image_result.GetNDArray()
    image_data = cv2.rotate(image_data, cv2.ROTATE_90_COUNTERCLOCKWISE)
    image_data = cv2.flip(image_data, 1)
    count, bins = np.histogram(image_data.flatten())
    line, = ax.stairs(count, bins)
    image_result.Release()

while(continue_recording):
    try:
        image_result = cam.GetNextImage(1000)

        if image_result.IsIncomplete():
            print('Image incomplete with image status %d ...' % image_result.GetImageStatus())

        else:                    
            image_data = image_result.GetNDArray()
            image_data = cv2.rotate(image_data, cv2.ROTATE_90_COUNTERCLOCKWISE)
            image_data = cv2.flip(image_data, 1)
                
            count, bins = np.histogram(image_data.flatten())
                        
            cv2.imshow("FLIR", image_data)
            cv2.setWindowProperty("FLIR", cv2.WND_PROP_TOPMOST, 1)
            
            # 更新直方图
            line.set_data(bins[:-1], count)
            ax.relim()
            ax.autoscale_view()
            plt.draw()
            plt.pause(0.001)
            
            key = cv2.waitKey(100) & 0xFF
            if key == ord('s'):
                f_name = f"test_{acq_count}.png"
                cv2.imwrite(f_name, image_data)
                acq_count += 1
                print(f"Saved image: {f_name}")
                        
            if keyboard.is_pressed('ENTER'):
                print('Program is closing...')
                cv2.destroyAllWindows()
                plt.close('all')             
                input('Done! Press Enter to exit...')
                continue_recording=False                        

            image_result.Release()

    except PySpin.SpinnakerException as ex:
        print('Error: %s' % ex)
        continue_recording = False

cam.EndAcquisition()
# cam.DeInit() # 按需添加相机销毁逻辑

内容的提问来源于stack exchange，提问作者DrBwts