Skip to content
yolov3, deep_sort and optical flow
Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
__pycache__
dataset add video Dec 14, 2018
deep_sort commit Dec 8, 2018
model_data commit Dec 8, 2018
optical_flow commit Dec 8, 2018
tools commit Dec 8, 2018
yolo3
LICENSE
README.md commit Dec 14, 2018
convert.py
tracking.py
yolo.py commit Dec 8, 2018
yolo.pyc commit Dec 8, 2018
yolov3.cfg

README.md

Multi-Object-Tracking

General Idea

A multi-object-tracking algorithm uses YOLO v3, deep_sort and optical flow based on Kanade–Lucas–Tomasi (KLT).

Methodology

  1. YOLO v3 detection
  2. deep_sort tracker update
  3. optical flow tracker update

Dependences

The code has been tested in python 3.5, ubuntu 16.04.

  1. tensorflow
  2. keras
  3. numpy
  4. sklearn
  5. scipy
  6. scikit-image
  7. opencv

How to run

  1. Download yolov3 model from YOLO website. Convert this model to a Keras model. For this project, we train a new yolov3 model and use Keras.save_model.
  2. Run script: python3.5 tracking.py

Results

  1. test result video 1: https://youtu.be/SKX-EcQnens
  2. test result video 2: https://youtu.be/56RKbOaInYI

Reference work

  1. keras YOLO v3: https://github.com/qqwweee/keras-yolo3
  2. deep_sort: https://github.com/nwojke/deep_sort
  3. YOLO v3 deep_sort integration: https://github.com/Qidian213/deep_sort_yolov3
  4. optical flow: https://github.com/ZheyuanXie/OpticalFlow
You can’t perform that action at this time.