# 上料识别，获取上料图片，保存到data/images文件夹下 import os import cv2 import time import numpy as np import onnxruntime import win32com.client def list_webcams(): # 创建一个WMI客户端实例 wmi = win32com.client.GetObject("winmgmts:") webcams = wmi.InstancesOf("Win32_PnPEntity") # 创建摄像头和索引字典 webcam_dict = {} index = 0 for device in webcams: name = getattr(device, 'Name', None) pnp_class = getattr(device, 'PNPClass', None) if name is not None and cam1 in name: # 将设备名字和索引添加到字典中 webcam_dict[cam1] = index index += 1 elif name is not None and cam2 in name: # 将设备名字和索引添加到字典中 webcam_dict[cam2] = index index += 1 return webcam_dict # 调用另一个长焦镜头，拍摄清晰的局部药材图片 def get_image(): camera2_index = webcams.get(cam2) print("第二个摄像头索引：" + str(camera2_index)) # 打开摄像头 capture = cv2.VideoCapture(camera2_index, cv2.CAP_DSHOW) # 设置分辨率 capture.set(cv2.CAP_PROP_FRAME_WIDTH, 1920) # 宽度 capture.set(cv2.CAP_PROP_FRAME_HEIGHT, 1080) # 高度 # 检查摄像头是否成功打开 if not capture.isOpened(): print("无法打开摄像头2") exit() width2 = capture.get(cv2.CAP_PROP_FRAME_WIDTH) height2 = capture.get(cv2.CAP_PROP_FRAME_HEIGHT) print("摄像头2分辨率:", width2, "x", height2) # 循环读取摄像头画面 # Shadows name 'width' from outer scope count = 0 while True: ret2, frame2 = capture.read() if not ret2: print("无法读取摄像头画面") break count += 1 # 1920*1080的图像，中心裁剪640*480的区域 # a2 = int(height2 / 2 - target_height / 2) # b2 = int(height2 / 2 + target_height / 2) # c2 = int(width2 / 2 - target_width / 2) # d2 = int(width2 / 2 + target_width / 2) # cropped_frame2 = frame2[a2:b2, c2:d2] if count == 2: # 调整图像尺寸 resized_frame2 = cv2.resize(frame2, (target_width, target_height)) # 显示画面 # cv2.imshow('Output2', resized_frame2) # 生成保存文件名，以当前时间命名 save_name2 = time.strftime("%Y%m%d%H%M%S", time.localtime()) + ".jpg" # 保存调整尺寸后的图片 cv2.imwrite(save_path + "2/" + save_name2, resized_frame2) # cv2.imshow("Camera", resized_frame2) print("保存图片:", save_name2) break # 结束线程 capture.release() if __name__ == '__main__': cam1 = "USB Camera" cam2 = "PC Camera" webcams = list_webcams() print(webcams) target_width = 1024 target_height = 768 save_path = "data/images/" # 加载ONNX模型 session = onnxruntime.InferenceSession("model/loading.onnx") # 摄像头索引号，通常为0表示第一个摄像头 camera_index = webcams.get(cam1) print("第一个摄像头索引：" + str(camera_index)) modelmeta = session.get_modelmeta() metadata_map = modelmeta.custom_metadata_map classes = eval(metadata_map['names']) # 打开摄像头 cap = cv2.VideoCapture(camera_index, cv2.CAP_DSHOW) # 设置分辨率 # cap.set(cv2.CAP_PROP_FRAME_WIDTH, 3840) # 宽度 # cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 2160) # 高度 # 检查摄像头是否成功打开 if not cap.isOpened(): print("无法打开摄像头") exit() width = cap.get(cv2.CAP_PROP_FRAME_WIDTH) height = cap.get(cv2.CAP_PROP_FRAME_HEIGHT) print("摄像头分辨率:", width, "x", height) # 目标图像尺寸 # 计时器 frame_count = 0 start_time = time.time() stime = time.time() if not os.path.exists(save_path): os.makedirs(save_path) class_old = "1" count = 0 status = "没有上料" # 循环读取摄像头画面 while True: # 睡眠100毫秒 time.sleep(0.1) ret, frame = cap.read() if not ret: print("无法读取摄像头画面") break resized_frame = frame if height > target_height and width > target_width: # 1920*1080的图像，中心裁剪640*480的区域 a = int(height / 2 - target_height / 2) b = int(height / 2 + target_height / 2) c = int(width / 2 - target_width / 2) d = int(width / 2 + target_width / 2) print(a, b, c, d) cropped_frame = frame[a:b, c:d] # 调整图像尺寸 resized_frame = cv2.resize(cropped_frame, (target_width, target_height)) # 获取当前时间 current_time = time.time() # 如果距离上一次保存已经过去1秒，则保存当前画面 # if current_time - start_time >= 3.0: # # 生成保存文件名，以当前时间命名 # save_name = time.strftime("%Y%m%d%H%M%S", time.localtime()) + ".jpg" # # 保存调整尺寸后的图片 # cv2.imwrite(save_path + save_name, frame) # print("保存图片:", save_name) # # 重置计时器 # start_time = time.time() # 预处理 blob = cv2.dnn.blobFromImage(resized_frame, 1 / 255.0, (640, 640), swapRB=True, crop=False) # 模型推理 outputs = session.run(None, {session.get_inputs()[0].name: blob}) # # print(outputs) # 应用softmax函数 probabilities = outputs[0] # 找到最大概率的类别 predicted_class = np.argmax(probabilities, axis=1)[0] max_probability = np.max(probabilities, axis=1)[0] class_ = classes[predicted_class] # 计算类型重复的次数，类别更换之后重新计数 if class_ != class_old: count = 0 else: count += 1 class_old = class_ print(f"{class_}:{count}: {max_probability}") # 判断是否上料并且上料次数大于10次 if class_ == "shangliao" and count > 10: status = "正在上料" # 每隔3秒取一帧图像 # 如果距离上一次保存已经过去1秒，则保存当前画面 if current_time - stime >= 10.0: save_name = time.strftime("%Y%m%d%H%M%S", time.localtime()) + ".jpg" # 保存调整尺寸后的图片 cv2.imwrite(save_path + "1/" + save_name, resized_frame) # 重置计时器 stime = time.time() get_image() else: status = "没有上料" print(status) # # 找到概率较高的前十个类别 # top_ten_classes = np.argsort(probabilities, axis=1)[0][-1:] # # # 输出前十个类别 # print("Top 5 Classes:") # for i in top_ten_classes: # print(f"{classes[i]}: {probabilities[0][i]}") # 计算帧速率 frame_count += 1 end_time = time.time() elapsed_time = end_time - start_time fps = frame_count / elapsed_time # print(f"FPS: {fps:.2f}") # 将FPS绘制在图像上 cv2.putText(resized_frame, f"FPS: {fps:.2f}", (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255), 2, cv2.LINE_AA) # 显示画面 cv2.imshow("Camera", resized_frame) # 检测按键，如果按下q键则退出循环 if cv2.waitKey(1) & 0xFF == ord('q'): break # 关闭摄像头 cap.release() # 关闭所有窗口 cv2.destroyAllWindows()