如何在OpenCV中实现多目标自动检测追踪，替代手动选择bbox

我正在用OpenCV开发一个算法，用来追踪视频特定区域内的多个移动目标。现在的算法需要在第一帧手动选择bbox（边界框）才能开始追踪，我想改成自动检测所有目标，但不知道具体怎么实现。

当前手动选择bbox的代码片段：

# Select the bounding box in the first frame
bbox = cv2.selectROI(frame, False)
ret = tracker.init(frame, bbox)

完整原始代码：

tracker = cv2.TrackerKCF_create()

# import the necessary packages
# noinspection PyUnresolvedReferences
from imutils.video import VideoStream
import argparse
# noinspection PyUnresolvedReferences
import imutils
import time
import cv2

print("GeeksForGeeks")
print("Your OpenCV version is: " + cv2.__version__)

tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'MOSSE', 'CSRT']
tracker_type = tracker_types[5]

if tracker_type == 'BOOSTING':
    tracker = cv2.legacy.TrackerBoosting_create()
if tracker_type == 'MIL':
    tracker = cv2.TrackerMIL_create()
if tracker_type == 'KCF':
    tracker = cv2.TrackerKCF_create()
if tracker_type == 'TLD':
    tracker = cv2.legacy.TrackerTLD_create()
if tracker_type == 'MEDIANFLOW':
    tracker = cv2.legacy.TrackerMedianFlow_create()
#if tracker_type == 'GOTURN':
   # tracker = cv2.TrackerGOTURN_create()
if tracker_type == 'MOSSE':
    tracker = cv2.legacy.TrackerMOSSE_create()
if tracker_type == "CSRT":
   tracker = cv2.TrackerCSRT_create()


# Get the video file and read it
vid_capture = cv2.VideoCapture('/Users/leonienelz/Downloads/capture_output60_countr02_1676545157.mp4')
ret, frame = vid_capture.read()


frame_height, frame_width = frame.shape[:2]
# Resize the video for a more convinient view
frame = cv2.resize(frame, [frame_width // 2, frame_height // 2])
# Initialize video writer to save the results
output = cv2.VideoWriter(f'{tracker_type}.avi',
                         cv2.VideoWriter_fourcc(*'XVID'), 60.0,
                         (frame_width // 2, frame_height // 2), True)
if not ret:
    print('cannot read the video')
# Select the bounding box in the first frame
bbox = cv2.selectROI(frame, False)
ret = tracker.init(frame, bbox)
# Start tracking
while True:
    ret, frame = vid_capture.read()
    frame = cv2.resize(frame, [frame_width // 2, frame_height // 2])
    if not ret:
        print('something went wrong')
        break
    timer = cv2.getTickCount()
    ret, bbox = tracker.update(frame)
    fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
    if ret:
        p1 = (int(bbox[0]), int(bbox[1]))
        p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
        cv2.rectangle(frame, p1, p2, (255, 0, 0), 2, 1)
    else:
        cv2.putText(frame, "Tracking failure detected", (100, 80),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    cv2.putText(frame, tracker_type + " Tracker", (100, 20),
                cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
    cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50),
                cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
    cv2.imshow("Tracking", frame)
    output.write(frame)
    k = cv2.waitKey(1) & 0xff
    if k == 27: break

vid_capture.release()
output.release()
#cv2.destroyAllWindows()

核心思路 #

要实现自动检测+多目标追踪，分两步走：

1. 目标自动检测：在第一帧用目标检测模型识别所有需要追踪的目标边界框
2. 多目标追踪管理：为每个检测到的目标初始化独立追踪器，后续帧统一更新所有追踪器的位置

下面以HOG+SVM行人检测（适合行人追踪场景）和OpenCV多追踪器为例，修改代码实现需求：

修改后的完整代码 #

# 导入必要包
from imutils.video import VideoStream
import argparse
import imutils
import time
import cv2

print("GeeksForGeeks")
print("Your OpenCV version is: " + cv2.__version__)

tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'MOSSE', 'CSRT']
tracker_type = tracker_types[5]  # 选用MOSSE，速度快适合多目标场景

# 初始化多追踪器，用于管理多个目标的追踪器
multi_tracker = cv2.legacy.MultiTracker_create()

# 初始化HOG行人检测器（如果追踪其他目标，替换为对应检测器即可）
hog = cv2.HOGDescriptor()
hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())

# 读取视频文件
vid_capture = cv2.VideoCapture('/Users/leonienelz/Downloads/capture_output60_countr02_1676545157.mp4')
ret, frame = vid_capture.read()

frame_height, frame_width = frame.shape[:2]
# 调整视频尺寸方便查看
frame = cv2.resize(frame, [frame_width // 2, frame_height // 2])
# 初始化视频写入器保存结果
output = cv2.VideoWriter(f'{tracker_type}_multi.avi',
                         cv2.VideoWriter_fourcc(*'XVID'), 60.0,
                         (frame_width // 2, frame_height // 2), True)
if not ret:
    print('无法读取视频')

# ---------------------------
# 替换手动选框：自动检测目标
# ---------------------------
# HOG检测行人，参数可根据场景调整
rects, weights = hog.detectMultiScale(frame, winStride=(4, 4), padding=(8, 8), scale=1.05)
# 非极大值抑制，去除重叠框避免重复追踪
rects = imutils.non_max_suppression(rects, probs=None, overlapThresh=0.65)

# 为每个检测到的目标初始化追踪器并添加到多追踪器
for (x, y, w, h) in rects:
    # 根据选择的追踪器类型创建实例
    if tracker_type == 'BOOSTING':
        tracker = cv2.legacy.TrackerBoosting_create()
    elif tracker_type == 'MIL':
        tracker = cv2.TrackerMIL_create()
    elif tracker_type == 'KCF':
        tracker = cv2.TrackerKCF_create()
    elif tracker_type == 'TLD':
        tracker = cv2.legacy.TrackerTLD_create()
    elif tracker_type == 'MEDIANFLOW':
        tracker = cv2.legacy.TrackerMedianFlow_create()
    elif tracker_type == 'MOSSE':
        tracker = cv2.legacy.TrackerMOSSE_create()
    elif tracker_type == "CSRT":
        tracker = cv2.TrackerCSRT_create()
    # 将追踪器和对应目标框添加到多追踪器
    multi_tracker.add(tracker, frame, (x, y, w, h))

# 开始追踪循环
while True:
    ret, frame = vid_capture.read()
    frame = cv2.resize(frame, [frame_width // 2, frame_height // 2])
    if not ret:
        print('视频读取异常')
        break
    timer = cv2.getTickCount()
    # 更新所有追踪器的位置
    ret, bboxes = multi_tracker.update(frame)
    fps = cv2.getTickFrequency() / (cv2.getTickCount() - timer)
    
    # 绘制所有追踪框
    if ret:
        for i, bbox in enumerate(bboxes):
            p1 = (int(bbox[0]), int(bbox[1]))
            p2 = (int(bbox[0] + bbox[2]), int(bbox[1] + bbox[3]))
            # 用不同颜色区分不同目标
            color = (0, 255*(i%3), 255*(i%2))
            cv2.rectangle(frame, p1, p2, color, 2, 1)
    else:
        cv2.putText(frame, "部分目标追踪失败", (100, 80),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.75, (0, 0, 255), 2)
    
    # 添加文字信息
    cv2.putText(frame, tracker_type + " 多目标追踪器", (100, 20),
                cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
    cv2.putText(frame, "FPS : " + str(int(fps)), (100, 50),
                cv2.FONT_HERSHEY_SIMPLEX, 0.75, (50, 170, 50), 2)
    cv2.imshow("多目标追踪", frame)
    output.write(frame)
    # 按ESC退出
    k = cv2.waitKey(1) & 0xff
    if k == 27: break

# 释放资源
vid_capture.release()
output.release()
cv2.destroyAllWindows()

关键修改说明 #

1. 自动检测替换手动选框：用HOG+SVM实现行人自动检测，若需追踪其他目标，可替换为Haar级联分类器（如车辆检测）或YOLO系列模型（通用目标检测）
2. 多追踪器管理：使用cv2.legacy.MultiTracker_create()替代单个追踪器，支持同时追踪多个目标
3. 目标去重：用imutils.non_max_suppression去除重叠检测框，避免重复追踪同一目标
4. 多目标可视化：为每个目标分配不同颜色的追踪框，方便区分

扩展优化建议 #

• 特定区域过滤：在检测后，可添加逻辑只保留位于指定区域内的目标框，实现"特定区域内追踪"的需求
• 高精度检测：如果需要更高精度，可使用YOLOv8的ONNX模型，OpenCV支持直接加载ONNX模型进行推理
• 失效重检测：添加追踪失效判断，当某个目标追踪失败时，重新运行检测找回目标

内容的提问来源于stack exchange，提问作者phytonnewbie