Ubuntu 20.04下DeepStream 6.1.1完整安装指南（含CUDA 11.7和TensorRT 8.4配置）

Ubuntu 20.04下DeepStream 6.1.1完整安装指南含CUDA 11.7和TensorRT 8.4配置在计算机视觉和视频分析领域NVIDIA的DeepStream SDK已经成为开发者构建高效智能视频分析应用的首选工具。本文将详细介绍在Ubuntu 20.04系统上安装DeepStream 6.1.1的完整流程特别关注CUDA 11.7和TensorRT 8.4的正确配置方法。不同于一般的安装教程本指南将深入探讨各组件版本间的兼容性问题并提供实际安装过程中可能遇到的坑及其解决方案。1. 环境准备与系统检查在开始安装DeepStream之前确保您的系统满足以下基本要求操作系统Ubuntu 20.04 LTS推荐使用最新更新显卡NVIDIA GPU计算能力5.0及以上内存至少16GB RAM推荐32GB存储空间至少50GB可用空间首先更新系统并安装基础依赖sudo apt update sudo apt upgrade -y sudo apt install -y build-essential cmake git wget unzip检查NVIDIA显卡是否被系统识别lspci | grep -i nvidia如果输出中包含您的NVIDIA显卡信息说明硬件已被识别。接下来我们需要移除系统可能存在的旧版NVIDIA驱动sudo apt purge -y nvidia* sudo apt autoremove -y注意此步骤会删除所有已安装的NVIDIA驱动确保您有稳定的网络连接以下载新版驱动。2. 显卡驱动与CUDA 11.7安装2.1 安装NVIDIA显卡驱动DeepStream 6.1.1要求使用515.x或更高版本的NVIDIA驱动。以下是推荐的安装方法添加官方PPA仓库sudo add-apt-repository ppa:graphics-drivers/ppa sudo apt update查找适合您显卡的最新驱动版本ubuntu-drivers devices安装推荐驱动示例为515.76sudo apt install -y nvidia-driver-515安装完成后重启系统然后验证驱动是否正常工作nvidia-smi输出应显示类似以下信息----------------------------------------------------------------------------- | NVIDIA-SMI 515.76 Driver Version: 515.76 CUDA Version: 11.7 | |--------------------------------------------------------------------------- | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | || | 0 NVIDIA GeForce ... On | 00000000:01:00.0 On | N/A | | 30% 45C P8 15W / 250W | 987MiB / 8192MiB | 0% Default | | | | N/A | ---------------------------------------------------------------------------2.2 安装CUDA Toolkit 11.7DeepStream 6.1.1官方推荐使用CUDA 11.7版本。以下是安装步骤下载CUDA 11.7安装包wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda_11.7.1_515.65.01_linux.run运行安装程序sudo sh cuda_11.7.1_515.65.01_linux.run在安装界面中接受许可协议取消选中Driver选项因为已单独安装驱动确保选中CUDA Toolkit 11.7安装完成后添加环境变量echo export PATH/usr/local/cuda-11.7/bin:$PATH ~/.bashrc echo export LD_LIBRARY_PATH/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH ~/.bashrc source ~/.bashrc验证CUDA安装nvcc --version应显示类似以下输出nvcc: NVIDIA (R) Cuda compiler release 11.7, V11.7.993. 安装cuDNN 8.6和TensorRT 8.43.1 安装cuDNN 8.6cuDNN是NVIDIA提供的深度神经网络加速库必须与CUDA版本严格匹配。从NVIDIA开发者网站下载cuDNN 8.6.0 for CUDA 11.7需要注册账号安装.deb包假设下载文件为cudnn-local-repo-ubuntu2004-8.6.0.163_1.0-1_amd64.debsudo dpkg -i cudnn-local-repo-ubuntu2004-8.6.0.163_1.0-1_amd64.deb sudo cp /var/cudnn-local-repo-ubuntu2004-8.6.0.163/cudnn-local-*-keyring.gpg /usr/share/keyrings/ sudo apt update sudo apt install -y libcudnn88.6.0.163-1cuda11.7 libcudnn8-dev8.6.0.163-1cuda11.7验证cuDNN安装cat /usr/include/cudnn_version.h | grep CUDNN_MAJOR -A 2应显示类似#define CUDNN_MAJOR 8 #define CUDNN_MINOR 6 #define CUDNN_PATCHLEVEL 03.2 安装TensorRT 8.4TensorRT是NVIDIA的高性能深度学习推理库DeepStream依赖它来优化模型推理。下载TensorRT 8.4.3.1 for CUDA 11.7的.tar包解压并安装tar xvf TensorRT-8.4.3.1.Linux.x86_64-gnu.cuda-11.7.cudnn8.6.tar.gz cd TensorRT-8.4.3.1 sudo cp -r lib/* /usr/local/lib/ sudo cp -r include/* /usr/local/include/ echo export LD_LIBRARY_PATH$LD_LIBRARY_PATH:/path/to/TensorRT-8.4.3.1/lib ~/.bashrc source ~/.bashrc安装Python绑定可选但推荐cd python sudo pip3 install tensorrt-8.4.3.1-cp38-none-linux_x86_64.whl验证TensorRT安装python3 -c import tensorrt; print(tensorrt.__version__)应输出8.4.3.14. DeepStream 6.1.1安装与配置4.1 安装GStreamer依赖DeepStream基于GStreamer框架构建需要先安装相关依赖sudo apt install -y \ libssl1.1 \ libgstreamer1.0-0 \ gstreamer1.0-tools \ gstreamer1.0-plugins-good \ gstreamer1.0-plugins-bad \ gstreamer1.0-plugins-ugly \ gstreamer1.0-libav \ libgstreamer-plugins-base1.0-dev \ libgstrtspserver-1.0-0 \ libjansson4 \ libyaml-cpp-dev4.2 安装DeepStream SDK下载DeepStream 6.1.1 SDK.tbz2格式解压并安装sudo tar -xvf deepstream_sdk_v6.1.1_x86_64.tbz2 -C / cd /opt/nvidia/deepstream/deepstream-6.1 sudo ./install.sh sudo ldconfig添加环境变量echo export PATH/opt/nvidia/deepstream/deepstream-6.1/bin:$PATH ~/.bashrc source ~/.bashrc4.3 常见问题解决问题1libcudnn_ops_infer.so.8 is not a symbolic link解决方法sudo ln -sf /usr/local/cuda-11.7/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8.6.0 /usr/local/cuda-11.7/targets/x86_64-linux/lib/libcudnn_ops_infer.so.8 sudo ldconfig问题2GStreamer插件加载失败解决方法export GST_PLUGIN_PATH/opt/nvidia/deepstream/deepstream-6.1/lib/gst-plugins5. 验证安装与运行示例5.1 运行示例应用DeepStream提供了多个示例应用可以用来验证安装是否成功cd /opt/nvidia/deepstream/deepstream-6.1/samples/configs/deepstream-app deepstream-app -c source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt如果一切正常您应该能看到一个视频分析窗口显示实时物体检测和跟踪结果。5.2 性能优化建议为了获得最佳性能可以考虑以下调整内存分配策略export GST_DEBUGGST_MEMORY:4GPU利用率监控watch -n 0.5 nvidia-smiTensorRT优化使用FP16或INT8量化调整batch size以获得最佳吞吐量5.3 开发环境设置对于Python开发者建议设置虚拟环境python3 -m venv deepstream-env source deepstream-env/bin/activate pip install numpy opencv-pythonDeepStream Python绑定位于/opt/nvidia/deepstream/deepstream-6.1/lib/python3.8/site-packages可以将此路径添加到PYTHONPATH中export PYTHONPATH/opt/nvidia/deepstream/deepstream-6.1/lib/python3.8/site-packages:$PYTHONPATH