Home // SPWID 2017, The Third International Conference on Smart Portable, Wearable, Implantable and Disability-oriented Devices and Systems // View article

Deep on Edge: Opportunistic Road Damage Detection with City Official Vehicles

Authors:
Makoto Kawano
Takuro Yonezawa
Jin Nakazawa

Keywords: Smart City; Deep Learning; Edge Computing; Image Recognition;

Abstract:
How can we inspect city conditions at low costs? City infrastructures, such as roads, are elements of great importance in urban lives. Roads require constant inspection and repair due to deterioration, but it is expensive to do so with manual labor. Therefore, these works should be done automatically so that the cost of inspecting or repairing becomes cheap. While there are several works to address these road issues, our study focuses on official city vehicles, especially garbage trucks, to detect damaged lane markings (lines) which is the simplest case of road deterioration. Since our proposed system is implemented on an edge computer, it is easy to attach our system to vehicles. In addition, our system utilizes a camera, and since garbage trucks almost run through the entire area of a city every day, we can constantly obtain road images covering wide areas. Our model, which we call Deep on Edge (DoE), is a deep convolutional neural network which detects damaged lines from images. In our experiments, to evaluate our system, we first compared the accuracy of line damage detection of DoE with other baseline methods. Our results show that DoE outperforms previous approaches. Then, we investigate whether our system can detect the line damage on a running car. With this demonstration, we show that our system would be useful in practice.

Pages: 5 to 10

Copyright: Copyright (c) IARIA, 2017

Publication date: June 25, 2017

Published in: conference

ISSN: 2519-8440

ISBN: 978-1-61208-569-2

Location: Venice, Italy

Dates: from June 25, 2017 to June 29, 2017