ubuntu22.04@Jetson OpenCV安装

1. 源由

昨天用Jetson跑demo程序发现帧率很慢(只有10FPS左右)，按照视频文件怎么说应该有30FPS。但是为什么jetson orin nano跑不起来呢？？？

dnn_object_detection_embedded_device

想着，估计是GPU没有跑起来，正好凑着调试了下板子，记录、整理下资料。

2. 分析

从现象上看，OpenCV运行的时候没有跑GPU，从而导致高价买的Jetson Orin Nano连这么简单的Demo都跑不起来。

接下来，就先看下板子运行的环境：

• ubuntu22.04/jammy/aarch64

$ lsb_release -a
No LSB modules are available.
Distributor ID:	Ubuntu
Description:	Ubuntu 22.04.4 LTS
Release:	22.04
Codename:	jammy$ uname -a
Linux daniel-nvidia 5.15.122-tegra #1 SMP PREEMPT Mon Dec 18 21:24:25 PST 2023 aarch64 aarch64 aarch64 GNU/Linux

• OpenCV版本：4.8.0

$ opencv_version
4.8.0$ python3
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import cv2
>>> print(cv2.__version__)
4.8.0
>>> 

• 安装最新OpenCV4.9.0版本

1. 与之前《ubuntu22.04@laptop OpenCV安装》版本一致
2. 安装或者升级Jetson Orin Nano上OpenCV4.8.0版本
3. 官方论坛讨论：NVIDIA: install OpenCV for python3 in Jetson Nano
4. Google的网上资料：Install OpenCV on Jetson Nano
5. Google的教学视频

OpenCV with CUDA in Python on Jetson

3. 证实

要正式是否真的是OpenCV库对GPU的支持问题导致前面验证FPS帧率低，那么就需要通过jtop指令来查看。

有朋友问：你怎么知道的这些？
Ans: 其实我并不知道，是我搜出来的。其实就这么简单，并不是我知道的多，是网上这些问题很多人都已经早就解决了。即使真的有没有人解决的，那么也是有办法的，只不过思路和这个不太一样。

• NVIDIA: Jtop install fail
• NVIDIA: Where can I find these numbers in CUDA_ARCH_BIN?
• NVIDIA: Your GPU Compute Capability
• NVIDIA: OpenCV build script

3.1 jtop安装

$ pip3 --version
$ sudo -H pip3 install --no-cache-dir jetson-stats
$ sudo systemctl restart jtop.service
$ sudo reboot

注：一定要重启板子，否则无效。

3.2 jtop指令

$ jtop -h
usage: jtop [-h] [--health] [--error-log] [--no-warnings] [--restore] [--loop] [--color-filter] [-r REFRESH] [-p PAGE] [-v]jtop is system monitoring utility and runs on terminaloptions:-h, --help            show this help message and exit--health              Status jtop and fix (default: False)--error-log           Generate a log for GitHub (default: False)--no-warnings         Do not show warnings (default: False)--restore             Reset Jetson configuration (default: False)--loop                Automatically switch page every 5s (default: False)--color-filter        Change jtop base colors, you can use also JTOP_COLOR_FILTER=True (default: False)-r REFRESH, --refresh REFRESHrefresh interval (default: 1000)-p PAGE, --page PAGE  Open fix page (default: 1)-v, --version         show program's version number and exit

3.3 GPU支持情况

OpenCV: 4.8.0 with CUDA: NO，所以该版本不支持GPU。

4. 安装OpenCV

两个链接脚本供参考安装：

1. JetsonHacksNano/buildOpenCV
2. mdegans/nano_build_opencv

为了更好的配合我们的例子，就整合到SnapLearnOpenCV/scripts中供大家参考，上述脚本支持：

• 笔记本电脑(64位普通CPU)安装：install_opencv_for_laptop.sh – ubuntu22.04@laptop OpenCV安装
• AI最火的(Jetson Orin Nano)安装：install_opencv_for_jetson.sh

4.1 修改内容

$ git diff
diff --git a/build_opencv.sh b/build_opencv.sh
index c12bb40..891eea3 100755
--- a/build_opencv.sh
+++ b/build_opencv.sh
@@ -45,8 +45,8 @@ setup () {git_source () {echo "Getting version '$1' of OpenCV"
-    git clone --depth 1 --branch "$1" https://github.com/opencv/opencv.git
-    git clone --depth 1 --branch "$1" https://github.com/opencv/opencv_contrib.git
+    git clone --depth 1 --branch "$1" [e-mail]:opencv/opencv.git
+    git clone --depth 1 --branch "$1" [e-mail]:opencv/opencv_contrib.git}install_dependencies () {
@@ -63,9 +63,8 @@ install_dependencies () {libatlas-base-dev \libavcodec-dev \libavformat-dev \
-        libavresample-dev \libcanberra-gtk3-module \
-        libdc1394-22-dev \
+        libdc1394-dev \libeigen3-dev \libglew-dev \libgstreamer-plugins-base1.0-dev \
@@ -90,8 +89,6 @@ install_dependencies () {libxvidcore-dev \libx264-dev \pkg-config \
-        python-dev \
-        python-numpy \python3-dev \python3-numpy \python3-matplotlib \
@@ -103,14 +100,13 @@ install_dependencies () {configure () {local CMAKEFLAGS="-D BUILD_EXAMPLES=OFF
-        -D BUILD_opencv_python2=ON-D BUILD_opencv_python3=ON-D CMAKE_BUILD_TYPE=RELEASE-D CMAKE_INSTALL_PREFIX=${PREFIX}-D CUDA_ARCH_BIN=5.3,6.2,7.2,8.7-D CUDA_ARCH_PTX=-D CUDA_FAST_MATH=ON
-        -D CUDNN_VERSION='8.0'
+        -D CUDNN_VERSION='8.9'-D EIGEN_INCLUDE_PATH=/usr/include/eigen3-D ENABLE_NEON=ON-D OPENCV_DNN_CUDA=ON

4.2 Python2环境【不需要】

• python2环境：python-dev python-numpy
• OpenCV编译选项：-D BUILD_opencv_python2=ON

4.3 ubuntu22.04环境

• 无libavresample-dev库
• libdc1394-22-dev变更为libdc1394-dev

4.4 国内/本地环境问题

• 将https链接调整为git库链接

-    git clone --depth 1 --branch "$1" https://github.com/opencv/opencv.git
-    git clone --depth 1 --branch "$1" https://github.com/opencv/opencv_contrib.git
+    git clone --depth 1 --branch "$1" [e-mail]:opencv/opencv.git
+    git clone --depth 1 --branch "$1" [e-mail]:opencv/opencv_contrib.git

• git库采用git协议才能下载

$ git config --global "url.ssh://[e-mail]:443/.insteadOf" [e-mail]:

4.5 cudnn版本问题

《Linux 36.2@Jetson Orin Nano基础环境构建》安装的是8.9.4.25.

-        -D CUDNN_VERSION='8.0'
+        -D CUDNN_VERSION='8.9'

5. 总结

修改完上述脚本后，可以直接执行脚本指令：

$ git clone [e-mail]:mdegans/nano_build_opencv.git
$ cd nano_build_opencv
$ nano build_opencv.sh
$ ./build_opencv.sh或者
$ git clone [e-mail]:SnapDragonfly/SnapLearnOpenCV.git
$ cd SnapLearnOpenCV/scripts
$ ./install_opencv_for_jetson.sh

大致耗时约1个小时???没有计时，反正也挺快的。执行opencv_version python jtop指令：

$ echo $PATH
/home/daniel/.local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin$ opencv_version
4.9.0$ python 
Python 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import cv2
>>> print(cv2.__version__)
4.9.0
>>> $ jtop

最后，关于使用GPU和使用CPU对Demo的影响：

dnn_object_detection_jetson_orin_nano

注：本次环境安装为遇到默认路径导致的优先级问题。

6. 参考资料

【1】ubuntu22.04@laptop OpenCV安装
【2】Linux 36.2@Jetson Orin Nano基础环境构建
【3】ubuntu22.04@laptop OpenCV定制化安装
【4】Github操作网络异常笔记