A Multiple Image-based Estimation Method of Road Surface Condition via Mobile Nets Using In-vehicle Camera Images

Date and Time: Tuesday, May 9, 2023: 10:00 AM - 12:00 PM
Location: Keck 100

 

Lead Presenter:  Hiroki Kinoshita, 
Affiliation:  Graduate School of Engineering, Hokkaido University
Social Media Handle: 

Lead Presenter Biography

Hiroki Kinoshita,
Graduate school of Engineering, Student, Hokkaido University
Email: hiorkiusa@eis.hokudai.ac.jp


Hiroki Kinoshita is a 1st grade in masters degree in graduate school of engineering, Hokkaido University. His research is on Road management in the cold snowy regions, especially by utilizing image and video analysis. 

Co-Authors

Sho Takahashi, Toru Hagiwara
Ph.D., Associate Professor, Ph.D., Professor, Hokkaido University
Email: stakahashi@eng.hokudai.ac.jp, hagiwara@eng.hokudai.ac.jp
Phone Number: +81-11-706-6215, +81-11-706-6214

Shinobu Azuma
Nexco-Engineering Hokkaido Company Limited
Email: s.azuma.sb@e-nexco.co.jp

Yuji Iwasaki, Teppei Mori, Yasushi Hanatsuka, Masamu Ishizuki
Digital Solution Center, Bridgestone
Email: {yuji.Iwasaki, teppei.mori, yasushi.hanatsuka, masamu.ishizuki}@bridgestone.com

Presentation Description 

Road surface conditions in the cold snowy regions change rapidly due to snowfalls and temperature change. To grasp the winter road surface condition, a construction of a novel method to estimate the road surface condition from images obtained by an in-vehicle camera is expected. Therefore, we verified the effectiveness of estimating the road surface conditions by a single image taken from an in-vehicle camera. However, as this method uses a single image to estimate the road surface conditions, some estimations were wrong as the road surface was being blocked by windshield wipers and surrounding vehicles. Therefore, in this presentation, a method to estimate the road surface conditions using multiple neighboring frames is proposed. The proposed method is expected to estimate the road surface condition more accurately compared to the estimation method which only uses a single image. As the proposed method estimates the road surface condition not only by single image, but also images of neighboring frame images, correct estimation is expected even if the estimating image is not estimated correctly due to errors such as windshield wiper and surrounding vehicles blocking the road surface. Also, an experiment is conducted to verify the effectiveness of the proposed method. The results of the experiment verifiy the effectiveness of the proposed method.

Extended Summary

Extended Abstract

Download Extended Abstract PDF 

Presentation Video