如何解决YOLO v8处理视频时无法绘制车辆检测框的问题

我使用YOLOv8开展项目，任务是识别车辆轮廓（以矩形框展示），并统计穿过画面中线的对向、同向车道车辆数量，最终汇总总数，需逐帧处理视频。目前检测输出正常，但视频画面未显示车辆轮廓，已尝试多种方法仍未解决。

检测输出示例 #

0: 384x640 27 cars, 1 truck, 1 tv, 93.0ms
Speed: 1.9ms preprocess, 93.0ms inference, 0.4ms postprocess per image at shape (1, 3, 384, 640)
0: 384x640 26 cars, 2 trucks, 88.1ms
Speed: 1.7ms preprocess, 88.1ms inference, 0.5ms postprocess per image at shape (1, 3, 384, 640)
0: 384x640 26 cars, 2 trucks, 104.2ms
Speed: 1.5ms preprocess, 104.2ms inference, 0.6ms postprocess per image at shape (1, 3, 384, 640)

当前代码 #

import cv2
from ultralytics import YOLO

model = YOLO('yolov8s.pt')

video_path = 'output2.avi'
video = cv2.VideoCapture(video_path)

if not video.isOpened():
    print("错误：无法打开视频。")
    exit()

current_frame = 0

while True:
    ret, frame = video.read()
    if not ret:
        break

    current_frame += 1

    results = model(frame)

    car_results = [detection for detection in results if detection[-1] in [2, 5, 7]]

    for result in car_results:
        x1, y1, x2, y2, confidence, class_id = result.xyxy[0]
        x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
        cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 0, 255), 2)

    cv2.imshow('Video', frame)

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

video.release()
cv2.destroyAllWindows()

问题分析与修复 #

核心问题是YOLOv8的results对象结构处理错误：

• model(frame)返回的是Results对象列表，每个元素对应单帧检测结果，不能直接遍历results取detection[-1]
• 正确做法是先获取单帧的Results对象，再通过boxes属性提取检测框信息

修复后的代码如下：

import cv2
from ultralytics import YOLO

model = YOLO('yolov8s.pt')

video_path = 'output2.avi'
video = cv2.VideoCapture(video_path)

if not video.isOpened():
    print("错误：无法打开视频。")
    exit()

current_frame = 0

while True:
    ret, frame = video.read()
    if not ret:
        break

    current_frame += 1

    # 获取当前帧的检测结果对象（YOLOv8返回列表，取第一个元素）
    results = model(frame)[0]

    # 过滤车辆类（2=轿车、5=巴士、7=卡车）
    for box in results.boxes:
        class_id = int(box.cls[0])
        if class_id in [2, 5, 7]:
            # 提取框坐标并转为整数
            x1, y1, x2, y2 = map(int, box.xyxy[0])
            # 绘制红色矩形框
            cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 0, 255), 2)

    cv2.imshow('Video', frame)

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

video.release()
cv2.destroyAllWindows()

关键修改点 #

1. 将results = model(frame)改为results = model(frame)[0]，明确获取当前帧的检测结果对象
2. 通过results.boxes遍历所有检测框，替代原错误的遍历逻辑
3. 用box.cls[0]获取类别ID、box.xyxy[0]获取框坐标，符合YOLOv8的API规范

内容的提问来源于stack exchange，提问作者Kkbehx