AGX Xavier 搭建360环视教程【二、环境配置】

AGX Xavier 场景下的 【OpenCV + CUDA + GStreamer】 重装 & 编译的2025年稳定方案，不要采用ffmpeg方案。

✅ 1️⃣ 先卸载老版本

AGX 自带很多预装包，原则：卸载干净，避免旧库和新编译冲突。

🔹 卸载 OpenCV

dpkg -l | grep opencv
sudo apt-get remove --purge libopencv* python3-opencv

# 自己 make install 装过的也一起删
sudo rm -rf /usr/local/include/opencv4
sudo rm -rf /usr/local/lib/libopencv*
sudo rm -rf /usr/local/lib/pkgconfig/opencv4.pc
sudo rm -rf /usr/local/lib/python3*/dist-packages/cv2*


# 卸载 pip 版本（如果有）
pip3 uninstall opencv-python opencv-contrib-python



sudo ldconfig

✅ 2️⃣ 检查 CUDA 和 GStreamer

AGX Xavier 自带 CUDA，JetPack 4.5 默认是 CUDA 10.2 + GStreamer 1.14.x
→ 不要自己乱装 CUDA，保持 JetPack 自带就行。

检查一下：

nvcc --version
gst-launch-1.0 --version

确认有：

CUDA compilation tools, release 10.2, V10.2.89
GStreamer 1.14.x

✅ 3️⃣ 安装依赖

sudo apt-get update

# 编译工具
sudo apt-get install -y build-essential cmake git pkg-config

# GStreamer (一定要有)
sudo apt-get install -y libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-libav

# 图像编解码
sudo apt-get install -y libjpeg-dev libpng-dev libtiff-dev libavcodec-dev libavformat-dev libswscale-dev libv4l-dev

# OpenGL 和 V4L2
sudo apt-get install -y libgtk-3-dev libcanberra-gtk* libtbb2 libtbb-dev libdc1394-22-dev

# Python
sudo apt-get install -y python3-dev python3-numpy

✅ 4️⃣ 编译 OpenCV (推荐 4.5.5)

🔹 拉源码

cd ~
git clone https://github.com/opencv/opencv.git
cd opencv
git checkout 4.5.5

cd ~
git clone https://github.com/opencv/opencv_contrib.git
cd opencv_contrib
git checkout 4.5.5


cd ~/opencv
mkdir build && cd build

将opencv_contrib放在opencv的文件夹下。后面配置CMake路径时注意统一。

解决ADE被墙

 cmake的过程中，会因为下载ADE被墙，而停顿，因此只好手动下载。

wget https://github.com/opencv/ade/archive/refs/tags/v0.1.1f.zip
# 解压到 opencv_contrib/modules/ade 或者 opencv/3rdparty/ade
unzip v0.1.1f.zip -d v0.1.1f

opencv/
 ├── 3rdparty/
 │    ├── ade/
 │         ├── CMakeLists.txt
 │         ├── sources
           ├── v0.1.1f
           ├── v0.1.1f.zip
             ...

只要在 opencv/3rdparty/ade/ 里能找到它的 CMakeLists.txt

ADE_DIR 必须指向 解压后的源码根目录。

并在CMake时加上一句：

-D ADE_DIR=$HOME/opencv/3rdparty/ade/v0.1.1f

 解决NVIDIA_OPTICAL_FLOW被墙

🔹 CMake 推荐配置（AGX Xavier 专属 🚀）

cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local -D OPENCV_EXTRA_MODULES_PATH=~/opencv/opencv_contrib/modules -D WITH_CUDA=ON -D WITH_CUDNN=ON -D OPENCV_DNN_CUDA=ON -D ENABLE_FAST_MATH=ON -D CUDA_FAST_MATH=ON -D WITH_CUBLAS=ON -D WITH_GSTREAMER=ON -D WITH_FFMPEG=ON -D WITH_GSTREAMER_0_10=OFF -D WITH_LIBV4L=ON -D WITH_OPENGL=ON -D WITH_QT=OFF -D BUILD_opencv_python2=OFF -D BUILD_opencv_python3=ON -D PYTHON3_EXECUTABLE=$(which python3) -D PYTHON3_INCLUDE_DIR=$(python3 -c "from sysconfig import get_paths as gp; print(gp()['include'])") -D PYTHON3_LIBRARY=$(python3 -c "from sysconfig import get_paths as gp; print(gp()['stdlib'])") -D OPENCV_GENERATE_PKGCONFIG=ON -D BUILD_TESTS=OFF -D OPENCV_ENABLE_NONFREE=ON -D BUILD_EXAMPLES=OFF -D  OPENCV_FORCE_3RDPARTY_BUILD=ON -D ADE_DIR=$HOME/opencv/3rdparty/ade/v0.1.1f ..

需要包含：

-D BUILD_opencv_python3=ON \
-D PYTHON3_EXECUTABLE=$(which python3) \
-D PYTHON3_INCLUDE_DIR=$(python3 -c "from sysconfig import get_paths as gp; print(gp()['include'])") \
-D PYTHON3_LIBRARY=$(python3 -c "from sysconfig import get_paths as gp; print(gp()['stdlib'])") \

 否则会提示：

 python3 -c "import cv2; print(cv2.getBuildInformation())" | grep CUDATraceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'cv2'
nvidia@agxA:~/opencv/build$ python3 -c "import cv2; print(cv2.getBuildInformation())" | grep GStreamer
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'cv2'

🔹 编译 & 安装

make -j$(nproc)   # Xavier AGX 建议一次用 4-6 核，别一次全开，避免 OOM
sudo make install
sudo ldconfig

✅ 5️⃣ 验证

# OpenCV
python3 -c "import cv2; print(cv2.getBuildInformation())" | grep CUDA
python3 -c "import cv2; print(cv2.getBuildInformation())" | grep GStreamer

# GStreamer
gst-inspect-1.0 | grep nv

如果报错如下：

 python3 -c "import cv2; print(cv2.getBuildInformation())" | grep CUDATraceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'cv2'
nvidia@agxA:~/opencv/build$ python3 -c "import cv2; print(cv2.getBuildInformation())" | grep GStreamer
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'cv2'

cd ~/opencv/build
find . -name "cv2*.so"


# 预计输出如下：
# ./lib/python3/cv2.cpython-36m-aarch64-linux-gnu.so

 如果没找到，说明你 cmake 没开：

-D BUILD_opencv_python3=ON

如果找到了，手动拷贝到 Python 路径，先看看python的版本。然后找到指定的路径下。

which python3
which pip3
python3 --version

从下面指令中，找到 site-packages 路径：

python3 -m site

 输出如下：

sys.path = [
    '/home/nvidia/opencv/build',
    '/home/nvidia/ws_intersection/devel/lib/python2.7/dist-packages',
    '/home/nvidia/ws_lego_loam/devel/lib/python2.7/dist-packages',
    '/home/nvidia/ws_can/devel/lib/python2.7/dist-packages',
    '/home/nvidia/ws_drv_other/devel/lib/python2.7/dist-packages',
    '/opt/ros/melodic/lib/python2.7/dist-packages',
    '/usr/lib/python36.zip',
    '/usr/lib/python3.6',
    '/usr/lib/python3.6/lib-dynload',
    '/home/nvidia/.local/lib/python3.6/site-packages',
    '/home/nvidia/.local/lib/python3.6/site-packages/torchvision-0.11.1-py3.6-linux-aarch64.egg',
    '/usr/local/lib/python3.6/dist-packages',
    '/usr/lib/python3/dist-packages',
    '/usr/lib/python3.6/dist-packages',
]
USER_BASE: '/home/nvidia/.local' (exists)
USER_SITE: '/home/nvidia/.local/lib/python3.6/site-packages' (exists)
ENABLE_USER_SITE: True

将so文件，拷贝到指定路径下：

sudo cp ./lib/python3/cv2*.so /usr/local/lib/python3.6/dist-packages/

注意要和你的 Python 版本对得上！

关键点总结：

1️⃣ JetPack 4.5 的 CUDA/GStreamer 都需要自己编时打开
2️⃣ ADE 可手动解压替代自动下载
3️⃣ JNI、Tesseract 非刚需可以跳过或后装
4️⃣ BUILD_opencv_python3=ON 和 Python 路径一定配好
5️⃣ 不要混用 pip 的 opencv-python，它是 CPU 版，没 GPU