Perception of Autonomous Vehicle for Localization Using Camera and GPS

Abstract

This paper is an attempt to give solution for autonomous vehicle localization using Camera and GPS. However, GPS accuracy in outdoor localization has less accuracy in different environmental conditions. The main objective of the system is to provide vehicle position with respect to its surroundings with help of Camera based localization solution for autonomous vehicle using the aide of GPS/GNSS. In this project, we propose different solutions for vehicle localization with use of pole-like landmarks over a predefined path with assigned GPS coordinates to it and also a unique identification assigned poles. The process is performed in real-time by a camera system which detects the poles with help of different CNN models as the primary sensor and using GPS as secondary information medium along with calculations of distance between the system and the pole by Euclidean method which reduces the computational load system. Results are illustrated from the comparisons between localization result and reference geo-location data in the data set. The system will be able to predict the location with an accuracy of 85% plus and reducing the error of conventional GPS that is 4–10 m to 40–60 cm mean average. The experimental results from the localization methods look to be a promising alternative.