VIDEO SYSTEM

FAST,EASY-LEARNING AI VIDEO PROJECT

• This project, built upon PyAV (for video streaming) and TensorRT (for YOLO model inference), achieves real-time analysis and display of 8 video streams simultaneously. It encompasses the following core functionalities:

  • API-style invocation interface

  • Video stream acquisition

  • Model inference execution

  The implementation leverages TensorRT to optimize inference speed, ensuring high-performance processing across all concurrent streams.

INTRODUCTION

• please install cuda,cudnn and tensorrt before learning this project
  
• here is the installation of TensorRT

USAGE

1. model export

• running the below code after installing Tensorrt

    trtexec --onnx=model.onnx \
    --fp16 \
    --saveEngine=model_int8.engine \
    --workspace=4096

• move the exported engine file to /video_system/weights

2. environment installation

• recommend conda environment

  pip install -r requirement.txt

3. run

• main.py can run locally with a cv2.imshow
• video_stream.py run with push a rtsp streamer but need a NVR( mediamtx ...)

4. run port_test.py