Skip to content

Latest commit

 

History

History
48 lines (34 loc) · 2.26 KB

smartphoneAPPS.md

File metadata and controls

48 lines (34 loc) · 2.26 KB

RoadCrackDetector

What is RoadCrackDetector?

RoadCrackDetector is a smartphone apps that detects damages on the road by utilizing deep neural network model.

スマートフォン上で深層学習モデルを動かすことで、道路路面の損傷画像を検出するアプリケーションです。

How to use RoadCrackDetector?

Basic Functions

  • Detect Road Damages when the car is running
  • Stop processing automatically when the car is stopping
  • Record road images every 1 second

Instalation location

  • Car dashboard

img1

img2

How to train the Crack Detection model?

(March 2017) We used yolo detector for training the model. The training dataset is more than 30,000 road images including road damages.. This application can just detect "damages".

(September 2017) We used SSD using MobileNet for training the model. Trained model with 9,053 road images with damages can be accessed here. This application can just detect "damages" and classify "damage types".

Android application

Apps with tiny YOLO(March 2017) RoadCrackDetector.apk(140MB) (Android 7.0 or higher is required)

Apps with MobileNet(September 2017) RoadCrackDetector.apk(240MB) (Android 7.0 or higher is required)

Experiments in some local governments in Japan(August 2017)

We did road inspection with our apps in Toga villege, please check our website! You can also watch the movie in the experiment as a demo(demo movie)

富山県の利賀村にて、本アプリケーションを用いて実際に路面点検を実施しました (参考)。
実験の様子をデモ動画として公開していますので、ご覧ください。