/YOLOv4_MOT_TensorRT

People detection, tracking, and counting. Deploy on NVIDIA Jetson nano with TensorRT.

Primary LanguagePython

Introduction

  • deployed on NVIDIA Jetson
  • detecting people
  • tracking detected people
  • counting in/out of people

People detection

  • using YOLOv4-tiny as object detection model
  • model is trained in Darknet framework and converted to TensorRT inference engine with 608 x 608 input size

Prepare to run

  • a cam or a video with people to replace crowd.mp4 in main.py
  • modify the main.py: replace 'crowd.mp4' by your video
  • modify the main.py: point_1_y and point_2_y to define a line to represent the boundary for in/out counting
  • download the engine file: 'yolov4_tiny_608.engine'

Run python

  • simply type: python3 main.py