2020-11-26

在项目需要用到 jetson开发平台时，很方便地安装了deepstream，记录自己的使用心得，概念的话可以参考其他的blog。

deepstream的特点

• sdk没有开源，所以使用起来可能会遇到麻烦。
• IVA = stream pipeline编解码 + nvinfer预测模型 （本质是tensorrt），后续会记录自己的tensorrt使用记录。。。留坑。
• 支持python和C++
• 据说可以在x86上run，但我之前测试遇到错误，尚未解决。目前是在jetson成功测试的。

参考

参考https://www.jianshu.com/p/aeaa36e92ca8
NVIDIA DeepStream SDK

demo测试

参考 https://blog.csdn.net/tosonw/article/details/104154090

1. 输出３０路视频分析结果

参考：https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/blob/master/back-to-back-detectors/README.md

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream/samples/configs/deepstream-app$ deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display_int8.txt

2 人脸，汽车，车牌检测

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream-5.0/sources/apps/sample_apps/deepstream_reference_apps/back-to-back-detectors$ ./back-to-back-detectors /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.h264

---

sample apps

---

python目标检测

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream-5.0/sources/deepstream_python_apps/apps/deepstream-test1$ python3 deepstream_test_1.py /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.h264 

c++ 目标检测

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-test1$ ./deepstream-test1-app /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.h264 

级联element：目标检测+目标跟踪

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-test2$ python3 deepstream_test_2.py /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.h264 

rtsp视频流检测 fps=7.8

nvidia@miivii-tegra:/opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-test3$ python3 deepstream_test_3.py rtsp://admin:q3202336@192.168.88.64:554/h264/ch1/main/av_stream

总结： 目前看来有点乱，后续再花时间整理一次，deepstream使用起来比较舒服，但是sdk不开源可能有点限制，jetson上还可以参考

jetson-inference