如何使用Python和OpenCV进行实时目标检测实例详解

import cv2
import numpy as np
import time
from concurrent.futures import ThreadPoolExecutor

conf_threshold = 0.5
nms_threshold = 0.4
Width = 416
Height = 416

net = cv2.dnn.readNet('../needFiles/yolov3.weights', '../needFiles/yolov3.cfg')
with open('../needFiles/coco.names', 'r') as f:
    classes = f.read().strip().split('\n')

color_list = np.random.uniform(0, 255, size=(len(classes), 3))

def fetch_frame(cap):
    ...
def process_frame(frame):
    ...

executor = ThreadPoolExecutor(max_workers=2)
frame_future = executor.submit(fetch_frame, cap)

while True:
    ...
    ret, frame, height, width, channels = frame_future.result()
    frame_future = executor.submit(fetch_frame, cap)
    processed_frame = executor.submit(process_frame, frame).result()
    ...

if cv2.waitKey(1) & 0xFF == ord('q'):
    break
...
cap.release()
cv2.destroyAllWindows()

# 导入必要的库
import cv2
import numpy as np
import time
from concurrent.futures import ThreadPoolExecutor

# 设置置信度阈值和非极大值抑制（NMS）阈值
conf_threshold = 0.5
nms_threshold = 0.4
Width = 416
Height = 416

# 加载预训练的 YOLOv3 模型
net = cv2.dnn.readNet('../needFiles/yolov3.weights', '../needFiles/yolov3.cfg')

# 加载可识别的类名
with open('../needFiles/coco.names', 'r') as f:
    classes = f.read().strip().split('\n')

# 为不同的类别创建一个颜色列表
color_list = np.random.uniform(0, 255, size=(len(classes), 3))

# 打开摄像头进行视频帧的捕获
cap = cv2.VideoCapture(0)

# 初始化一个ThreadPoolExecutor用于多线程处理
executor = ThreadPoolExecutor(max_workers=2)


# 定义fetch_frame函数，从视频流中获取视频帧
def fetch_frame(cap):
    ret, frame = cap.read()  # 读取一帧图像
    height, width, channels = frame.shape  # 获取图像的尺寸和通道信息
    return ret, frame, height, width, channels


# 定义process_frame函数，处理每帧图像并进行目标检测
def process_frame(frame):
    # 将帧转换为模型的输入格式
    blob = cv2.dnn.blobFromImage(frame, 1 / 255.0, (416, 416), swapRB=True, crop=False)
    net.setInput(blob)
    output_layers = net.getUnconnectedOutLayersNames()  # 获取输出层的名字
    layer_outputs = net.forward(output_layers)  # 进行前向传播，获取检测结果

    boxes = []  # 用于存储检测到的边界框
    confidences = []  # 用于存储边界框的置信度
    class_ids = []  # 用于存储边界框的类别ID

    # 循环每个输出层的检测结果
    for output in layer_outputs:
        for detection in output:
            scores = detection[5:]  # 获取类别的得分
            class_id = np.argmax(scores)  # 获取得分最高的类别ID
            confidence = scores[class_id]  # 获取得分最高的置信度

            # 过滤低置信度的检测结果
            if confidence > conf_threshold:
                center_x = int(detection[0] * width)
                center_y = int(detection[1] * height)
                w = int(detection[2] * width)
                h = int(detection[3] * height)

                # 计算边界框的位置和尺寸
                x = int(center_x - w / 2)
                y = int(center_y - h / 2)

                # 将边界框的位置、尺寸、置信度和类别ID添加到列表中
                boxes.append([x, y, w, h])
                confidences.append(float(confidence))
                class_ids.append(class_id)

    # 使用非极大值抑制去除重叠的边界框
    indices = cv2.dnn.NMSBoxes(boxes, confidences, conf_threshold, nms_threshold)

    # 在原图上绘制边界框和类别标签
    for i in indices.flatten():
        box = boxes[i]
        x = box[0]
        y = box[1]
        w = box[2]
        h = box[3]
        label = str(classes[class_ids[i]])
        color = color_list[class_ids[i]]
        cv2.rectangle(frame, (x, y), (x + w, y + h), color, 2)
        cv2.putText(frame, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)

    return frame


# 在进入循环前，先读取一帧以开始异步处理
frame_future = executor.submit(fetch_frame, cap)

# 主循环
while True:
    start = time.time()  # 记录开始处理的时间点
    # 获取当前帧和相应信息
    ret, frame, height, width, channels = frame_future.result()

    # 异步读取下一帧
    frame_future = executor.submit(fetch_frame, cap)

    # 如果当前帧读取成功，则继续处理
    if ret:
        # 使用线程池异步处理当前帧
        processed_frame = executor.submit(process_frame, frame).result()

        # 计算FPS
        end = time.time()
        fps = 1 / (end - start)
        # 在处理好的帧上显示FPS
        cv2.putText(processed_frame, "FPS: " + str(round(fps, 2)), (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 255, 0),
                    2)
        # 显示处理好的帧
        cv2.imshow('frame', processed_frame)

        # 如果我们按下 ‘q' 键，退出循环
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    else:
        break  # 如果帧没有被成功读取，退出循环

# 释放视频捕获对象和销毁所有OpenCV窗口
cap.release()
cv2.destroyAllWindows()

   # 加载预训练的 YOLOv3 模型
   net = cv2.dnn.readNetFromDarknet('../needFiles/yolov3.cfg', '../needFiles/yolov3.weights')
   
   # 启用 CUDA
   net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
   net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)