python
from collections import deque
import numpy as np
import imutils
import time
import cv2
import os
from tqdm import tqdm

python
# 参数设置
video_path = "./video/pedestrian.mp4"  # 要检测的视频路径
filter_confidence = 0.5  # 用于筛除置信度过低的识别结果
threshold_prob = 0.3  # 用于NMS去除重复的锚框
model_path = "./yolo-obj"  # 模型文件的目录
# 载入数据集标签
labelsPath = os.path.sep.join([model_path, "coco.names"])
LABELS = open(labelsPath).read().strip().split("\n")
# 载入模型参数文件及配置文件
weightsPath = os.path.sep.join([model_path, "yolov4.weights"])
configPath = os.path.sep.join([model_path, "yolov4.cfg"])
        以下代码通过上面给出的路径从配置和参数文件中载入模型，载入模型使用OpenCV的readNetFromDarknet方法载入，我们可以利用它载入自行训练的模型权重以进行检测操作：
python
# 从配置和参数文件中载入模型
print("[INFO] 正在载入模型...")
net = cv2.dnn.readNetFromDarknet(configPath, weightsPath)
ln = net.getLayerNames()
ln = [ln[i[0] - 1] for i in net.getUnconnectedOutLayers()]
        为了后面标记检测的目标标记框以及目标移动路径，这里创建两个变量存储标记框的颜色及移动路径的点坐标，其代码如下：
python
# 初始化用于标记框的颜色
np.random.seed(42)
COLORS = np.random.randint(0, 255, size=(200, 3), dtype="uint8")
# 用于展示目标移动路径
pts = [deque(maxlen=30) for _ in range(9999)]
        准备就绪，开始从视频文件路径初始化视频对象，其代码如下：
python
# 初始化视频流
vs = cv2.VideoCapture(video_path)
(W, H) = (None, None)
frameIndex = 0  # 视频帧数
        尝试读取视频帧并获取视频总帧数total、每帧画面的尺寸（vw, vh），同时创建一个视频写入对象output_video用于后面保存检测标记的视频，该部分代码如下：
python
# 试运行，获取总的画面帧数
try:
    prop = cv2.cv.CV_CAP_PROP_FRAME_COUNT if imutils.is_cv2() \
        else cv2.CAP_PROP_FRAME_COUNT
    total = int(vs.get(prop))
    print("[INFO] 视频总帧数：{}".format(total))
# 若读取失败，报错退出
except:
    print("[INFO] could not determine # of frames in video")
    print("[INFO] no approx. completion time can be provided")
    total = -1
fourcc = cv2.VideoWriter_fourcc(*'XVID')
ret, frame = vs.read()
vw = frame.shape[1]
vh = frame.shape[0]
print("[INFO] 视频尺寸：{} * {}".format(vw, vh))
output_video = cv2.VideoWriter(video_path.replace(".mp4", "-det.avi"), fourcc, 20.0, (vw, vh))  # 处理后的视频对象
        接下来开始遍历视频帧进行检测，为了清楚地显示检测进度，我这里使用了tqdm，它可以在运行的命令行中显示当前的进度条。读取当前视频帧可以使用OpenCV中VideoCapture的read()，该方法返回当前画面和读取标记，可通过标记判断是否到达视频最后一帧：
python
# 遍历视频帧进行检测
for fr in tqdm(range(total)):
    # 从视频文件中逐帧读取画面
    (grabbed, frame) = vs.read()
    # 若grabbed为空，表示视频到达最后一帧，退出
    if not grabbed:
        break
    # 获取画面长宽
    if W is None or H is None:
        (H, W) = frame.shape[:2]

python
    # 将一帧画面读入网络
    blob = cv2.dnn.blobFromImage(frame, 1 / 255.0, (416, 416), swapRB=True, crop=False)
    net.setInput(blob)
    start = time.time()
    layerOutputs = net.forward(ln)
    end = time.time()
        以上代码中layerOutputs即为检测结果，这里写一个循环从中将检测框坐标、置信度值、识别到的类别序号分别存放在boxes、confidences、classIDs变量中。layerOutputs的结果是按照检测对象存放的，在循环中我们还需根据置信度值的阈值过滤掉一些置信度值不高的结果：
python
 boxes = []  # 用于检测框坐标
 confidences = []  # 用于存放置信度值
 classIDs = []  # 用于识别的类别序号
# 逐层遍历网络获取输出
for output in layerOutputs:
     # loop over each of the detections
     for detection in output:
         # extract the class ID and confidence (i.e., probability)
         # of the current object detection
         scores = detection[5:]
         classID = np.argmax(scores)
         confidence = scores[classID]
         # 过滤低置信度值的检测结果
         if confidence > filter_confidence:
             box = detection[0:4] * np.array([W, H, W, H])
             (centerX, centerY, width, height) = box.astype("int")
             # 转换标记框
             x = int(centerX - (width / 2))
             y = int(centerY - (height / 2))
             # 更新标记框、置信度值、类别列表
             boxes.append([x, y, int(width), int(height)])
             confidences.append(float(confidence))
             classIDs.append(classID)

python
# 使用NMS去除重复的标记框
    idxs = cv2.dnn.NMSBoxes(boxes, confidences, filter_confidence, threshold_prob)

python
 dets = []
    if len(idxs) > 0:
        # 遍历索引得到检测结果
        for i in idxs.flatten():
            (x, y) = (boxes[i][0], boxes[i][1])
            (w, h) = (boxes[i][2], boxes[i][3])
            dets.append([x, y, x + w, y + h, confidences[i], classIDs[i]])
    np.set_printoptions(formatter={'float': lambda x: "{0:0.3f}".format(x)})
    dets = np.asarray(dets)

python
    # 使用sort算法，开始进行追踪
    tracker = Sort()  # 实例化追踪器对象
    tracks = tracker.update(dets)
    boxes = []  # 存放追踪到的标记框
    indexIDs = []
    cls_IDs = []
    c = []
    for track in tracks:
        boxes.append([track[0], track[1], track[2], track[3]])
        indexIDs.append(int(track[4]))
        cls_IDs.append(int(track[5]))

python
    if len(boxes) > 0:
        i = int(0)
        for box in boxes:  # 遍历所有标记框
            (x, y) = (int(box[0]), int(box[1]))
            (w, h) = (int(box[2]), int(box[3]))
            # 在图像上标记目标框
            color = [int(c) for c in COLORS[indexIDs[i] % len(COLORS)]]
            cv2.rectangle(frame, (x, y), (w, h), color, 4)
            center = (int(((box[0]) + (box[2])) / 2), int(((box[1]) + (box[3])) / 2))
            pts[indexIDs[i]].append(center)
            thickness = 5
            # 显示某个对象标记框的中心
            cv2.circle(frame, center, 1, color, thickness)
            # 显示目标运动轨迹
            for j in range(1, len(pts[indexIDs[i]])):
                if pts[indexIDs[i]][j - 1] is None or pts[indexIDs[i]][j] is None:
                    continue
                thickness = int(np.sqrt(64 / float(j + 1)) * 2)
                cv2.line(frame, (pts[indexIDs[i]][j - 1]), (pts[indexIDs[i]][j]), color, thickness)
            # 标记跟踪到的目标和数目
            text = "{}-{}".format(LABELS[int(cls_IDs[i])], indexIDs[i])
            cv2.putText(frame, text, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 1, color, 3)
            i += 1
        以上操作成功将标记信息写入了frame画面中，要想实时显示在屏幕上只需调用OpenCV中的imshow方法开启一个窗口显示：
python
# 实时显示检测画面
    cv2.imshow('Stream', frame)
    output_video.write(frame)  # 保存标记后的视频
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
    # print("FPS:{}".format(int(0.6/(end-start))))
    frameIndex += 1
    if frameIndex >= total:  # 可设置检测的最大帧数提前退出
        print("[INFO] 运行结束...")
        output_video.release()
        vs.release()
        exit()