Design And Development Of Spherical Camera Based Deep Learning Enabled Auto Lane Width Measurement For Road Safety Grading System

Tay, Choon Kiat (2017) Design And Development Of Spherical Camera Based Deep Learning Enabled Auto Lane Width Measurement For Road Safety Grading System. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (24 Pages) Design And Development Of Spherical Camera Based Deep Learning Enabled Auto Lane Width Measurement For Road Safety Grading System.pdf - Submitted Version Download (763kB)

Abstract

Design and Development of Spherical Camera based Deep Learning enabled Auto Lane Width Measurement for Road Safety Grading System is a project which provide a system that recognize the road objects using deep learning approach.This project is inspired by the Malaysian government commitment to provide a safer and higher quality road.Malaysian Institute of Road Safety Research (MIROS) funded this project RM100,000 to develop a system that can used for road survey and road width measurement. Spherical camera is suggested in this project because front facing camera limits the receptive field of the survey video.Deep Learning is famous of their robustness in recognize object.Therefore,this project proposes a new method to recognize the car and road marking and represent it in distinct colors.The predicted lane marking will further be used by lane width measurement. Deep Learning network used by this project is SegNet from University of Cambridge,UK.To suit our application,we further train the model using the spherical images dataset.The final global test accuracy hits 78.9% on spherical video recognition.Road width measurement is accurate when the lane is detected in the bottom area on the video.

Actions (login required)

View Item