10.1 基于深度学习的目标检测

首次完成时间：2024 年 11月 20 日

1. 使用OpenCV的dnn模块实现图像分类。

1）程序代码：

import numpy as np
import cv2

# 解析标签文件
row = open("model1/synset_words.txt").read().strip().split("\n")
class_label = [r[r.find(" "):].split(",")[0] for r in row]

# 载入caffe所需的配置文件
net = cv2.dnn.readNetFromCaffe("model1/bvlc_googlenet.prototxt",
                               "model1/bvlc_googlenet.caffemodel")

# 读取待分类图像
img = cv2.imread("photos/cat.jpg")  # 确保这里的路径是正确的

# 转换格式
blob = cv2.dnn.blobFromImage(img, 1, (224, 224), (104, 117, 123))

# 加载图像
net.setInput(blob)

# 预测
preds = net.forward()

# 排序，取概率最大的结果
idx = np.argsort(preds[0])[-1]

# 获取图片的原始尺寸
(h, w) = img.shape[:2]

# 等比例缩减图片大小
resized_img = cv2.resize(img, (w // 4, h // 4))

# 可视化处理，显示图像类别、置信度等信息
text = "label: {}-{:.2f}%".format(class_label[idx], preds[0][idx] * 100)
cv2.putText(resized_img, text, (5, 25), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 255, 0))

# 显示缩减后的图片
cv2.imshow("resized_result", resized_img)

# 保存缩减后的图片
cv2.imwrite("photos/resized_result.jpg", resized_img)

cv2.waitKey(0)
cv2.destroyAllWindows()

2）代码分析：

2. 使用OpenCV实现目标检测，实现发现不明车辆或行人进入检测区，即进行报警。

1）程序代码：

import numpy as np
import cv2

def prepareDataSet():
    # 准备数据集
    args = {}
    args["prototxt"] = "model2/MobileNetSSD_deploy.prototxt"
    args["model"] = "model2/MobileNetSSD_deploy.caffemodel"
    return args

def createNet():
    # 构建网络模型对象
    args = prepareDataSet()
    # load our serialized model from disk
    print("[INFO] loading model...")
    net = cv2.dnn.readNetFromCaffe(args["prototxt"], args["model"])
    return net

if __name__ == "__main__":
    # 定义类别名称序列
    CLASSES = ["background", "aeroplane", "bicycle", "bird", "boat",
               "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog",
               "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]

    # 定义边框颜色序列
    COLORS = np.random.uniform(0, 255, size=(len(CLASSES), 3))

    # 打开摄像头或视频文件
    camera = cv2.VideoCapture("videos/12686501_3840_2160_60fps.mp4")

    # 构建网络模型
    net = createNet()

    while True:
        ret, frame = camera.read()
        if ret:
            # 将帧的尺寸调整为1080p
            frame = cv2.resize(frame, (1920, 1080))
            (h, w) = frame.shape[:2]
            blob = cv2.dnn.blobFromImage(cv2.resize(
                frame, (300, 300)), 0.007843, (300, 300), 127.5)
            net.setInput(blob)
            detections = net.forward()

            # 遍历结果
            for i in np.arange(0, detections.shape[2]):
                # 获得置信度
                confidence = detections[0, 0, i, 2]
                # 根据置信度阈值过滤执行度
                if confidence > 0.2:
                    # 根据最大置信度获取类别下标
                    idx = int(detections[0, 0, i, 1])
                    # 获取位置信息
                    box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
                    (startX, startY, endX, endY) = box.astype("int")

                    # 显示类别信息和位置边框
                    label = "{}: {:.2f}%".format(CLASSES[idx], confidence * 100)
                    print("[INFO] {}".format(label))
                    cv2.rectangle(frame, (startX, startY), (endX, endY), COLORS[idx], 2)
                    y = startY - 15 if startY - 15 > 15 else startY + 15
                    cv2.putText(frame, label, (startX, y),
                                cv2.FONT_HERSHEY_SIMPLEX, 0.5, COLORS[idx], 2)

                    if CLASSES[idx] == "person":
                        print("raise the alarm")

            # 显示结果
            cv2.imshow("result", frame)

            # 按下空格退出 or esc
            if cv2.waitKey(1) == ord(' ') or cv2.waitKey(1) == 27:
                break
        else:
            break

    camera.release()
    cv2.destroyAllWindows()

2）代码分析：