软硬件环境

  • windows 10 64bit
  • cuda 10.1
  • cudnn 7.6.35
  • anaconda with python 3.7
  • ubuntu 18.04 64bit
  • NVidia GTX 1070Ti
  • cmake 3.18.4
  • protobuf 3.8.0

简介

OpenPose人体姿态识别项目是美国卡耐基梅隆大学(CMU)基于卷积神经网络和监督学习并以caffe为框架开发的开源库。可以实现人体动作、面部表情、手指运动等姿态估计。适用于单人和多人，具有极好的鲁棒性，是世界上首个基于深度学习的实时多人二维姿态估计应用。人体姿态估计技术在体育健身、动作采集、人机交互、舆情监测等领域具有广阔的应用前景。

算法体验

如果仅仅是想体验下这个项目，官方提供了windows版的可执行文件下载，有gpu和cpu2个版本，地址是: https://github.com/CMU-Perceptual-Computing-Lab/openpose/releases/，解压后进入文件夹，下载项目中需要用到的各种模型

cd openpose\models
.\getModels.bat

javascript人体动作识别 人体实时动作识别_cmake


openpose

使用本地视频文件进行测试

cd ..
bin\OpenPoseDemo.exe --video examples\media\video.avi

javascript人体动作识别 人体实时动作识别_gpu_02


openpose

源码编译

我这里是有gpu环境，因此需要安装nvidia的驱动，cuda和cuDNN，具体安装方法前面的博文已经介绍过了，参考下面链接

  • nvidia驱动安装，https://xugaoxiang.com/2020/09/24/ubuntu-nvidia-driver/
  • CUDA和cuDNN安装， https://xugaoxiang.com/2019/12/13/ubuntu-cuda/
git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose.git
cd openpose
conda create -n openpose python=3.7
conda activate openpose
sudo apt install cmake build-essential
mkdir build
cd build
cmake ..
make -j `nproc`
cd ../models
./getModels.sh
cd ..
./build/example/openpose.bin --video examples/media/video.avi

javascript人体动作识别 人体实时动作识别_gpu_03


openpose

如果原来的系统中已经安装过了caffe，那么openpose可以直接使用而不需要再次安装了

cmake -DBUILD_CAFFE=OFF -DCaffe_INCLUDE_DIRS=$somewhere/caffe/include/ -DCaffe_LIBS=$somewhere/lib/libcaffe.so ..

如果需要在python中使用的话，cmake时需要加上参数-DBUILD_PYTHON=ON

如果不喜欢用命令行去配置cmake，可以使用cmake-gui

sudo apt install cmake-qt-gui

javascript人体动作识别 人体实时动作识别_卷积_04


openpose

模型下载

使用官方的模型下载脚本，速度非常的慢，这里将工程中用到的模型整理了，放在了百度网里，方便大家下载

链接：https://pan.baidu.com/s/15aZe7NN3jQerWN2uZRK0mw提取码：p1mj

pthread_create报错

Determining if the pthread_create exist failed with the following output:
Change Dir: /home/xugaoxiang/Works/github/openpose/build/CMakeFiles/CMakeTmp

Run Build Command:"/usr/bin/make" "cmTC_56b97/fast"
/usr/bin/make -f CMakeFiles/cmTC_56b97.dir/build.make CMakeFiles/cmTC_56b97.dir/build
make[1]: Entering directory '/home/xugaoxiang/Works/github/openpose/build/CMakeFiles/CMakeTmp'
/usr/bin/cmake: /home/xugaoxiang/anaconda3/lib/libcurl.so.4: no version information available (required by /usr/bin/cmake)
Building C object CMakeFiles/cmTC_56b97.dir/CheckSymbolExists.c.o
/usr/bin/cc   -fPIC    -o CMakeFiles/cmTC_56b97.dir/CheckSymbolExists.c.o   -c /home/xugaoxiang/Works/github/openpose/build/CMakeFiles/CMakeTmp/CheckSymbolExists.c
/usr/bin/cmake: /home/xugaoxiang/anaconda3/lib/libcurl.so.4: no version information available (required by /usr/bin/cmake)
Linking C executable cmTC_56b97
/usr/bin/cmake -E cmake_link_script CMakeFiles/cmTC_56b97.dir/link.txt --verbose=1
/usr/bin/cmake: /home/xugaoxiang/anaconda3/lib/libcurl.so.4: no version information available (required by /usr/bin/cmake)
/usr/bin/cc -fPIC     -rdynamic CMakeFiles/cmTC_56b97.dir/CheckSymbolExists.c.o  -o cmTC_56b97
CMakeFiles/cmTC_56b97.dir/CheckSymbolExists.c.o: In function `main':
CheckSymbolExists.c:(.text+0x1b): undefined reference to `pthread_create'
collect2: error: ld returned 1 exit status
CMakeFiles/cmTC_56b97.dir/build.make:97: recipe for target 'cmTC_56b97' failed
make[1]: *** [cmTC_56b97] Error 1
make[1]: Leaving directory '/home/xugaoxiang/Works/github/openpose/build/CMakeFiles/CMakeTmp'
Makefile:126: recipe for target 'cmTC_56b97/fast' failed
make: *** [cmTC_56b97/fast] Error 2

这是系统中缺失了pthread相关的库，解决方法

sudo apt-get install manpages-posix manpages-posix-dev

caffe代码拉不下来

-- Found glog    (include: /usr/include, library: /usr/lib/x86_64-linux-gnu/libglog.so)
-- Found Protobuf: /usr/local/lib/libprotobuf.so;-lpthread (found version "3.8.0")
-- Found OpenCV: /usr (found version "3.2.0")
-- Caffe will be downloaded from source now. NOTE: This process might take several minutes depending
        on your internet connection.
-- Caffe has already been downloaded.
error: pathspec 'c95002f' did not match any file(s) known to git.
-- Caffe will be built from source now.
-- Download the models.
-- Downloading BODY_25 model...
-- Model already exists.
-- Not downloading body (COCO) model
-- Not downloading body (MPI) model
-- Downloading face model...
-- Model already exists.
-- Downloading hand model...
-- Model already exists.
-- Models Downloaded.
-- Configuring done
-- Generating done
-- Build files have been written to: /home/xugaoxiang/Works/github/openpose/build

CMU自己维护了一个caffe的工程，但是找了一圈，并没有发现有c95002f这个tag，所以，在配置的时候拉不下来源码，其实clone最新的就可以了，因此去修改CMakeLists.txt，将c95002f替换成master就可以了

javascript人体动作识别 人体实时动作识别_卷积_05


openpose

protobuf版本不一致

Protobuf compiler version doesn't match library version 3.8.0

我这里使用的是anaconda的环境，默认安装的版本是3.11.4，与caffe要求的3.8.0不相符，使用conda安装3.8.0即可

conda install protobuf==3.8.0

或者也可以编译protobuf源码，步骤如下

wget https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz
tar xvf protobuf-3.8.0.tar.gz
cd protobuf-3.8.0
./autogen.sh
./configure
make
sudo make install

安装完毕之后，使用protoc --version来检查下版本是不是正确

atlas报错

CMake Error at /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:137 (message):
  Could NOT find Atlas (missing: Atlas_CLAPACK_INCLUDE_DIR
  Atlas_CBLAS_LIBRARY Atlas_BLAS_LIBRARY)
Call Stack (most recent call first):
  /usr/share/cmake-3.10/Modules/FindPackageHandleStandardArgs.cmake:378 (_FPHSA_FAILURE_MESSAGE)
  cmake/Modules/FindAtlas.cmake:43 (find_package_handle_standard_args)
  cmake/Dependencies.cmake:113 (find_package)
  CMakeLists.txt:46 (include)

因为系统中并没有安装atlas，使用apt安装

sudo apt install libatlas-base-dev

cmake版本问题

-- NVIDIA CUDA:
--   Target GPU(s)     :   Auto
--   GPU arch(s)       :   sm_61
--   cuDNN             :   Yes (ver. 7.6.5)
--
-- Install:
--   Install path      :   /home/xugaoxiang/Works/github/openpose/build/caffe
--
CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_cublas_device_LIBRARY (ADVANCED)
    linked by target "caffe" in directory /home/xugaoxiang/Works/github/openpose/3rdparty/caffe/src/caffe

-- Configuring incomplete, errors occurred!
See also "/home/xugaoxiang/Works/github/openpose/build/caffe/src/openpose_lib-build/CMakeFiles/CMakeOutput.log".
See also "/home/xugaoxiang/Works/github/openpose/build/caffe/src/openpose_lib-build/CMakeFiles/CMakeError.log".
CMakeFiles/openpose_lib.dir/build.make:105: recipe for target 'caffe/src/openpose_lib-stamp/openpose_lib-configure' failed
make[2]: *** [caffe/src/openpose_lib-stamp/openpose_lib-configure] Error 1
CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/openpose_lib.dir/all' failed
make[1]: *** [CMakeFiles/openpose_lib.dir/all] Error 2
Makefile:83: recipe for target 'all' failed
make: *** [all] Error 2

这个问题是由于系统中的cmake版本引起的，openpose对cmake版本的最低要求是3.12，具体可以参考链接 https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/prerequisites.md#ubuntu-prerequisites， 而ubuntu 18.04的默认版本是3.10.2。解决方法有2种，分别是

  • 使用snap安装
    安装方法非常简单，但是国内的速度是真的真的真的慢，以至于后来，实在是等不下去了，我直接拉源码编译了。愿意等的话，也是可以成功的
sudo apt remove --purge cmake
sudo snap install cmake --classic
  • 编译cmake的源码
    源码编译cmake，稍微复杂一点，步骤如下
sudo apt remove --purge cmake
sudo apt install libssl-dev
wget https://github.com/Kitware/CMake/releases/download/v3.18.4/cmake-3.18.4.tar.gz
tar xvf cmake-3.18.4
cd cmake-3.18.4
./bootstrap
make
sudo make install

libpangoft报错

在make的过程中报错

[ 51%] Linking CXX shared library libopenpose_core.so
[ 51%] Built target openpose_core
[ 51%] Linking CXX executable 1_custom_post_processing.bin
//home/xugaoxiang/anaconda3/lib/libpangoft2-1.0.so.0: undefined reference to `FcWeightToOpenTypeDouble'
//home/xugaoxiang/anaconda3/lib/libpangoft2-1.0.so.0: undefined reference to `FcWeightFromOpenTypeDouble'
collect2: error: ld returned 1 exit status
examples/tutorial_add_module/CMakeFiles/1_custom_post_processing.bin.dir/build.make:171: recipe for target 'examples/tutorial_add_module/1_custom_post_processing.bin' failed
make[5]: *** [examples/tutorial_add_module/1_custom_post_processing.bin] Error 1
CMakeFiles/Makefile2:1280: recipe for target 'examples/tutorial_add_module/CMakeFiles/1_custom_post_processing.bin.dir/all' failed
make[4]: *** [examples/tutorial_add_module/CMakeFiles/1_custom_post_processing.bin.dir/all] Error 2
[ 51%] Linking CXX executable 18_synchronous_custom_all_and_datum.bin
//home/xugaoxiang/anaconda3/lib/libpangoft2-1.0.so.0: undefined reference to `FcWeightToOpenTypeDouble'
//home/xugaoxiang/anaconda3/lib/libpangoft2-1.0.so.0: undefined reference to `FcWeightFromOpenTypeDouble'
collect2: error: ld returned 1 exit status
examples/tutorial_api_cpp/CMakeFiles/18_synchronous_custom_all_and_datum.bin.dir/build.make:171: recipe for target 'examples/tutorial_api_cpp/18_synchronous_custom_all_and_datum.bin' failed
make[5]: *** [examples/tutorial_api_cpp/18_synchronous_custom_all_and_datum.bin] Error 1
CMakeFiles/Makefile2:1307: recipe for target 'examples/tutorial_api_cpp/CMakeFiles/18_synchronous_custom_all_and_datum.bin.dir/all' failed
make[4]: *** [examples/tutorial_api_cpp/CMakeFiles/18_synchronous_custom_all_and_datum.bin.dir/all] Error 2
Makefile:148: recipe for target 'all' failed
make[3]: *** [all] Error 2
CMakeFiles/openpose_lib.dir/build.make:143: recipe for target 'openpose_lib' failed
make[2]: *** [openpose_lib] Error 2
CMakeFiles/Makefile2:94: recipe for target 'CMakeFiles/openpose_lib.dir/all' failed
make[1]: *** [CMakeFiles/openpose_lib.dir/all] Error 2
Makefile:102: recipe for target 'all' failed
make: *** [all] Error 2

解决方法是，使用系统的pangoft

cd $HOME/anaconda3/lib/
mv libpangoft2-1.0.so.0 libpangoft2-1.0.so.0.bak
ln -s /usr/lib/x86_64-linux-gnu/libpangoft2-1.0.so.0 libpangoft2-1.0.so.0

参考资料