安装DeepStream
安装DeepStream
安装依赖
$ sudo apt install libjson-glib-dev libjansson4 libvtk6-dev python-vtk6 openssl libssl-dev ffmpeg libx11-dev gstreamer1.0-plugins-base gstreamer1.0-plugins-good gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-libav gstreamer1.0-doc gstreamer1.0-tools gstreamer1.0-x gstreamer1.0-alsa gstreamer1.0-gl gstreamer1.0-gtk3 gstreamer1.0-qt5 gstreamer1.0-pulseaudio gstreamer1.0-rtsp libgstreamer-plugins-base1.0-dev libgstreamer1.0-0 libgstreamer1.0-dev libgstrtspserver-1.0-0 libgstrtspserver-1.0-dev
sudo apt-get install libv4l-dev
cd /usr/include/linux
sudo ln -s ../libv4l1-videodev.h videodev.h
提前下好deepstream6
下载地址：https://developer.nvidia.com/deepstream-getting-started
sudo tar -zxvf deepstream_sdk_v6.0.0_x86_64.tbz2 -C /
命令行安装gcc
sudo apt-get install build-essential
源码编译安装gcc
下载地址：http://mirrors.nju.edu.cn/gnu/gcc/gcc-7.5.0/
安装依赖：
下载其他依赖文件，从这个地址都能找到:http://www.mirrorservice.org/sites/sourceware.org/pub/gcc/infrastructure/
- gmp-6.1.0.tar.bz2
- mpfr-3.1.4.tar.bz2
- mpc-1.0.3.tar.gz
- isl-0.16.1.tar.bz2
每个依赖都使用相同的安装步骤，如下：
./configure
make -j2
sudo make install
编译安装gcc
编译安装gcc
./configure --enable-checking=release --enable-languages=c,c++ --disable-multilib
make -j128
sudo make install
如果出现gcc和g++版本不一致的问题,很可能是配置的问题,因为目前系统里有多个gcc了,调用如下命令,即可解决
/usr/local/bin/gcc是我的安装路径
sudo update-alternatives --install /usr/bin/gcc gcc /usr/local/bin/gcc 100
sudo update-alternatives --install /usr/bin/gcc gcc /usr/local/bin/g++ 100
安装Cmake
tar -xvf cmake-3.14.5.tar
cd cmake-3.14.5
./bootstrap
make
make install
TensorRT-8.0.1.6
下载地址：https://developer.nvidia.com/nvidia-tensorrt-download
安装pycuda
# 第一种方式 手动下载安装<https://www.lfd.uci.edu/~gohlke/pythonlibs/#pycuda>
pip install {pycuda}.whl
# 第二种方式 直接pip
pip install pycuda==2021.1
安装tensorrt
#在home下新建文件夹，命名为tensorrt_tar，然后将下载的压缩文件拷贝进来解压
tar xzvf TensorRT-8.0.1.6.Linux.x86_64-gnu.cuda-11.3.cudnn8.2.tar.gz
#解压得到TensorRT-8.0.1.6的文件夹，将里边的lib和include添加都系统搜索路径中
mv TensorRT-8.0.1.6/ /usr/local/
mkdir /usr/lib/TensorRT-8.0.1.6
mkdir /usr/include/TensorRT-8.0.1.6
sudo ln -s /usr/local/TensorRT-8.0.1.6/lib /usr/lib/TensorRT-8.0.1.6/lib
sudo ln -s /usr/local/TensorRT-8.0.1.6/include /usr/include/TensorRT-8.0.1.6/include
# 安装TensorRT python API
cd /usr/local/TensorRT-8.0.1.6/python
pip install tensorrt-8.0.1.6-py2.py3-none-any.whl
# 安装UFF,支持tensorflow模型转化
cd TensorRT-8.0.1.6/uff
pip install uff-0.5.5-py2.py3-none-any.whl
# 安装graphsurgeon，支持自定义结构
cd TensorRT-8.0.1.6/graphsurgeon
pip install graphsurgeon-0.3.2-py2.py3-none-any.whl
安装librdkafka
git clone https://github.com/edenhill/librdkafka.git
cd librdkafka
git reset --hard 7101c2310341ab3f4675fc565f64f0967e135a6a
./configure
make
sudo make install
sudo cp /usr/local/lib/librdkafka* /opt/nvidia/deepstream/deepstream-6.0/lib
源码编译CPU版OpenCV
cd opencv-4.5.1
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local/opencv-4.5.1 ..
make -j128
sudo make install
源码编译GPU版OpenCV
** 使用源码安装opencv，注意要带cuda和gstreamer，这一步容易出现问题，将一些关键步骤做些说明，按照下边指令安装 **
安装依赖
sudo apt-get install libavcodec-dev libavformat-dev libavdevice-dev ffmpeg
libglew-dev OpenGL Extension Wrangler - 开发环境
libtiff5-dev 标记图像文件格式库 (TIFF)，开发文件
zlib1g-dev
libjpeg-dev
libpng12-dev
libjasper-dev
libavcodec-dev
libavformat-dev
libavutil-dev
libpostproc-dev
libswscale-dev
libeigen3-dev
libtbb-dev
libgtk2.0-dev
pkg-config
编译OpenCV
wget https://github.com/opencv/opencv/archive/3.4.0.zip -O opencv-3.4.0.zip
unzip opencv-3.4.0.zip
cd opencv-3.4.0
wget https://github.com/opencv/opencv_contrib/archive/3.4.0.zip -O opencv_contrib-3.4.0.zip
unzip opencv_contrib-3.4.0.zip
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_PNG=OFF -DBUILD_TIFF=OFF -DBUILD_TBB=OFF -DBUILD_JPEG=OFF -DBUILD_JASPER=OFF -DBUILD_ZLIB=OFF -DBUILD_EXAMPLES=OFF -DBUILD_opencv_java=OFF -DBUILD_opencv_python2=ON -DBUILD_opencv_python3=ON -DENABLE_PRECOMPILED_HEADERS=OFF -DWITH_OPENCL=OFF -DWITH_OPENMP=OFF -DWITH_FFMPEG=ON -DWITH_GSTREAMER=ON -DWITH_CUDA=ON -DWITH_GTK=ON -DWITH_VTK=ON -DWITH_TBB=ON -DWITH_1394=OFF -DWITH_OPENEXR=OFF -DOPENCV_EXTRA_MODULES_PATH=/opt/opencv-3.4.0/opencv_contrib-3.4.0/modules -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -DCUDA_ARCH_BIN=7.5 -DINSTALL_C_EXAMPLES=ON -DINSTALL_TESTS=OFF ..
make -j128
sudo make install
查看显卡CUDA_ARCH_BIN
- 方法1：利用cuda_sample查看
cd /usr/local/cuda/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery
输出其中的** CUDA Capability Major/Minor version number: 7.5 **
- 方法2：git下载代码编译
git clone https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps.git
cd deepstream_tlt_apps/TRT-OSS/x86
nvcc deviceQuery.cpp -o deviceQuery
./deviceQuery
检测版本
pkg-config --modversion opencv
pkg-config --cflags --libs opencv
# or
python
import cv2
cv2.getBuildInformation()
安装编译好的opencv python库
cd opencv-3.4.0/build/python_loader/
python setup.py install
可能的报错
1.找不到gst/gstxxx.hpp问题
在/usr/include下新建软连接
$ cd /usr/include
$ sudo ln -s gstreamer-1.0/gst/ gst
2.dynlink_nvcuvid.h问题
cuda10不再提供dynlink_nvcuvid.h功能，修改opencv-3.4.0/modules/cudacodec/src目录下的文件
modules/cudacodec/src/precomp.hpp
modules/cudacodec/src/video_decoder.hpp
modules/cudacodec/src/video_parser.hpp
modules/cudacodec/src/cuvid_video_source.hpp
modules/cudacodec/src/frame_queue.hpp
找到下边的代码
#if CUDA_VERSION >= 9000
#include <dynlink_nvcuvid.h>
#else
#include <nvcuvid.h>
#endif
替换为
#if CUDA_VERSION >= 9000 && CUDA_VERSION < 10000
#include <dynlink_nvcuvid.h>
#else
#include <nvcuvid.h>
#endif
如果改完还报错的话，需要下载 nvidia-sdk，下载链接：https://developer.nvidia.com/designworks/video_codec_sdk/downloads/v9.0
下载解压后将其中的 nvcuvid.h, cuviddec.h copy 到 /usr/local/cuda/include
然后重新make即可
3.ippicv_2017u3_lnx_intel64_general_20170822.tgz下载慢问题
https://github.com/opencv/opencv_3rdparty/tree/ippicv/master_20170822/ippicv
查看build目录下的CmakeDownloadLog.txt，里边记录了下载地址，我这里的路径是
use_cache "/home/nvidia/opencv_3.4.0/opencv-3.4.0/.cache"
do_unpack "ippicv_2017u3_lnx_intel64_general_20170822.tgz" "4e0352ce96473837b1d671ce87f17359" "https://raw.githubusercontent.com/opencv/opencv_3rdparty/dfe3162c237af211e98b8960018b564bc209261d/ippicv/ippicv_2017u3_lnx_intel64_general_20170822.tgz" "/home/nvidia/opencv_3.4.0/opencv-3.4.0/build/3rdparty/ippicv"
#check_md5 "/home/nvidia/opencv_3.4.0/opencv-3.4.0/.cache/ippicv/4e0352ce96473837b1d671ce87f17359-ippicv_2017u3_lnx_intel64_general_20170822.tgz"
可以看到是把这个文件下载到/home/nvidia/opencv_3.3.1/opencv-3.3.1/.cache/ippicv/下，并命名为4e0352ce96473837b1d671ce87f17359-ippicv_2017u3_lnx_intel64_general_20170822.tgz，我们只需要把手动下载的文件然后复制到该路径下即可
4 cblas_xxx问题
如果出现编译问题“undefined reference to `cblas_sgemm(CBLAS_ORDER, CBLAS_TRANSPOSE, CBLAS_TRANSPOSE, int, int, int, float, float const, int, float const, int, float, float*, int)'”
可能需要重新编译BLAS，CBLAS，LPACK，参考链接《blas、lapack、cblas在Ubuntu上的安装》<https://www.jianshu.com/p/33c4aea6117b>
5 boostdesc_bgm.i等问题
在opencv_contrib目录中报错fatal error: boostdesc_bgm.i: No such file or directory
这个跟2.4.2类似，是有些包下载不成功导致的，可以直接下载https://github.com/nanmi/myresource/tree/main/opencv_3rdparts/下面的资源，参考链接：<https://github.com/opencv/opencv_contrib/issues/1301>，然后拷贝到opencv_contrib/modules/xfeatures2d/src/ 路径下，重新编译即可
6 多版本问题
如果之前安装过其它版本没有卸载，可以采用《Ubuntu下多个版本OpenCV管理（Multiple Opencv version）》来指定编译时用到的opencv版本，连接：<https://www.cnblogs.com/cmt/p/14580194.html?from=https%3A%2F%2Fwww.cnblogs.com%2Fxzd1575%2Fp%2F5555523.html>
7 opencv2/xfeatures2d/cuda.hpp问题
修改/opencv-3.4.0/modules/stitching/CMakeLists.txt文件，在开头增加下边内容，然后重新cmake和make
INCLUDE_DIRECTORIES("/<your location>/opencv_contrib-3.4.0/modules/xfeatures2d/include")
如果修改后还报错，则找到报错的文件，将头文件改成绝对路径
8 "xxxx/test_detectors_invariance.impl.hpp"问题
一般是说在features2d/test目录下没有XXX.hpp什么的，处理方式是将opencv-3.3.1/modules/features2d/test该目录下对于的缺少文件复制到opencv_contrib-3.3.1/modules/xfeatures2d/test该目录下，然后修改报错的文件的#include，将前面的地址删除，就让其在本地找
例如 :报错说在文件test_rotation_and_scale_invariance.cpp中找不到#include "xxxx/test_detectors_invariance.impl.hpp"，那么就在opencv-3.3.1/modules/features2d/test下去找test_detectors_invariance.impl.hpp文件，将其复制到opencv_contrib-3.3.1/modules/xfeatures2d/test目录，然后打开test_rotation_and_scale_invariance.cpp文件，修改#include "xxxx/test_detectors_invariance.impl.hpp"为#include "test_detectors_invariance.impl.hpp"即可
如果觉得难得每个文件去找，那么干脆将目录中的所有文件复制过去，之后就该对于报错文件的#include位置就好了。
安装DeepStream
cd /opt/nvidia/deepstream/deepstream-6.0/
sudo ./install.sh
sudo ldconfig
# 查看安装的版本
deepstream-app --version-all
可能的报错
1. 如果运行时报错提示找不到一些库，如libnvdsgst_meta.so，则需要把deepstream-6.0/lib添加到系统lib路径中，如下
sudo vim /etc/ld.so.conf
/opt/nvidia/deepstream/deepstream-6.0/lib/ //在文本后边添加该路径
sudo ldconfig //执行ldconfig立即生效
2. 使用这个shell命令来测试 deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display_int8.txt
如果报错：
** ERROR: <create_multi_source_bin:714>: Failed to create element 'src_bin_muxer'
** ERROR: <create_multi_source_bin:777>: create_multi_source_bin failed
** ERROR: <create_pipeline:1045>: create_pipeline failed
** ERROR: <main:632>: Failed to create pipeline
Quitting
App run failed
则是因为gstreamer缓存问题，运行下边指令删除即可，运行成功后，会显示检测画面。
rm ${HOME}/.cache/gstreamer-1.0/registry.*
如果是在服务器上运行，没有显示界面的话会报错如下
No EGL Display
nvbufsurftransform: Could not get EGL display connection
需要修改环境变量，如下
vim ~/.bashrc
#打开后在最后边加下边语句
unset DISPLAY
export DISPLAY=:0
#或者 export DISPLAY=:1
#保存退出
source ~/.bashrc
rm ${HOME}/.cache/gstreamer-1.0/registry.*
#然后再运行
deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display_int8.txt
增加deepstream的Python绑定
安装依赖
sudo apt install python-gi-dev
安装python-gi-dev之后就会有pyobject等内容了
安装Deepstream python 版
安装依赖
- CUDA Toolkit
- cuDNN library
- TensorRT
- GStreamer
- PyGObject
- NumPy
下载deepstream的python绑定源码
绑定是使用pybind11来封装的
git clone --recursive https://github.com/NVIDIA-AI-IOT/deepstream_python_apps.git
下载好deepstream_python_apps/3rdparty/pybind11
安装gst-python
cd deepstream_python_apps/gst-python
./autogen.sh
make
make install
或者
sudo apt-get install python-gst-1.0
编译ds的绑定成whl的python包文件
修改deepstream_python_apps/bindings下的cmake文件，
设置DS_PATHdeepstream6.2的路径，示例check_variable_set(DS_PATH ${CMAKE_SOURCE_DIR}/../../opt/nvidia/deepstream/deepstream-6.2)
cd deepstream_python_apps/bindings/
mkdir build
cd build
cmake ..
make
安装pyds
pip3 install ./pyds-1.1.4-py3-none*.whl
示例
deepstream_python_apps/apps/
