文章说明

本教程主要如何在 Jetson Nano 下安装yolov8
测试环境： Jetson Nano + Ubuntu 20.04 + Logitech C920 WebCam

前提准备

最好安装Ubuntu 20.04系统，因为预装好OpenCV-4.8.0等依赖
系统安装教程链接：Nano入门教程基础篇-安装Ubuntu 20.04系统
yolov8的基本库版本要求

| 依赖库           | 最低版本              | 用途              
| torch         | ≥ 1.8.0（推荐 ≥ 2.0）  | PyTorch 深度学习框架  
| numpy         | ≥ 1.24.0              | 基础矩阵运算          
| opencv-python | ≥ 4.6.0               | 图像处理和视频流        
| matplotlib    | ≥ 3.7.0               | 结果可视化           
| seaborn       | ≥ 0.13.0              | 可选图表美化          
| pandas        | ≥ 2.0.0               | 数据统计（训练日志）     
| tqdm          | 最新版即可             | 训练/推理进度条        
| scipy         | ≥ 1.10.0              | 数学优化函数          
| PyYAML        | ≥ 5.3                 | 解析 .yaml 配置文件 
| thop          | ≥ 0.1.1               | 模型参数量计算（可选）

安装RCM工具

$ rm online_shell.sh ; wget https://gitee.com/ncnynl/commands/raw/master/online_shell.sh ; sudo chmod +x ./online_shell.sh; ./online_shell.sh rm;online_shell.sh

安装步骤

RCM脚本安装

RCM脚本一键安装

$ rcm -s install_yolov8

分步安装

查询当前CUDA版本

$ /usr/local/cuda/bin/nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Feb_28_22:34:44_PST_2021
Cuda compilation tools, release 10.2, V10.2.300
Build cuda_10.2_r440.TC440_70.29663091_0

安装对应版本的pytorch，可参考以下教程
安装其他依赖

$ pip3 install numpy>=1.24.0 matplotlib>=3.7.0 opencv-python>=4.8.0 seaborn>=0.13.0 tqdm pandas>=2.0.0 scipy>=1.10.0 thop>=0.1.1

安装yolov8

$ pip3 install ultralytics==8.1.0

测试步骤

图片识别测试

下载测试图片

$ mkdir -p ~/tools/yolov8
$ cd ~/tools/yolov8
$ wget https://raw.githubusercontent.com/ultralytics/ultralytics/refs/heads/main/ultralytics/assets/bus.jpg

运行测试

$ yolo task=detect mode=predict model=weight/yolov8n.pt source=bus.jpg save=true

查看识别结果

$ eog runs/detect/predict8/bus.jpg

请输入图片描述

相机实时识别测试

创建测试脚本

$ cd ~/tools/yolov8
$ vim yolov8_usbcam.py
#!/usr/bin/env python3

import cv2
import torch
import time
from ultralytics import YOLO
import sys

def main():
    # Select camera (default 0)
    cam_id = int(sys.argv[1]) if len(sys.argv) > 1 else 0
    device = "cuda" if torch.cuda.is_available() else "cpu"
    print(f"[INFO] Using device: {device}")

Load YOLOv8n model

model = YOLO("yolov8n.pt") if device == "cuda": model.to("cuda") # keep full precision (FP32) on Jetson Nano

Open USB camera

cap = cv2.VideoCapture(cam_id) if not cap.isOpened(): raise RuntimeError(f"Failed to open camera /dev/video{cam_id}")

Set camera resolution

width, height = 640, 480 cap.set(cv2.CAP_PROP_FRAME_WIDTH, width) cap.set(cv2.CAP_PROP_FRAME_HEIGHT, height)

print("[INFO] Press 'q' to quit.")

try: while True: ret, frame = cap.read() if not ret: print("[WARN] Cannot read frame from camera.") break

start = time.time()
    results = model.predict(frame, imgsz=640, verbose=False)
    end = time.time()

    # Draw detection results
    annotated = results[0].plot()

    # Show frame
    cv2.imshow("YOLOv8 USB Camera", annotated)

    # Print FPS
    fps = 1 / (end - start)
    print(f"FPS: {fps:.2f}", end="\r")

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

finally: cap.release() cv2.destroyAllWindows() print("\n[INFO] Program exited.")

if __name__ == "__main__":
    main()

连接相机，执行脚本

$ python3 yolov8_usbcam.py