Muzzle Detection & Classify Fierce Dog

Graduation Project

video link: https://www.youtube.com/watch?v=4_CPPuSnH-c

---

Abstract

Currently, the population of households raising dogs in Korea reaches 27.7%. Many people follow the basic rules while raising dogs, but some people ignore the law of tie the dog's leash or wearing muzzle when they raise fierce dog. We have decided to proceed with a project on cases where fierce dogs do not wear muzzles among there issues.
We made Application (To take picture of fierce dog) and Server (Get the picture from Application and run deep learning models we trained) for this project. If person who use this app, take a picture of fierce dog without muzzle, (s)he will get a result information if dog is fierce dog or not and if it is fierce dog, then info of muzzle or not will returned too. Based on there information, our final goal is to allow a fierce dog without a muzzle to report to the police if it is determined that it is in the picture user takes.

---

Demo

1. Android Client(App) Dog in the picture

• Take picture of dog
• Send picture to Python server

2. Python Server

• Run Origin Yolov3 for input picture
• Run Custom Yolov3-tiny for detected dog picture
• Run Custom ResNet50 Classifier to determine No-muzzle dog picture
  

3. Android Client(App) Returned picture from Server

• Show the Returned picture(with context that explains the send picture)
• According to returned result, choose report to Police or not