Visual Odometry/Inertial Integration for Enhanced Land Vehicle Navigation in GNSS Denied Environment

Abstract

Self-driving vehicles have attained a huge research interest as it offers new opportunities for more secure and convenient transportation. Vision is considered one of the significant inputs to self-driving systems in terms of obstacle avoidance and navigation purposes. Recently, Consumer Portable Devices (CPDs) such as smartphones and tablets are continuously equipped with enhanced sensors and cameras that can be used in the navigation or obstacle avoidance applications.This paper presents an approach for enhancing the land vehicle navigation in GNSS denied environment by integrating the Visual Odometry (VO), On-Board Diagnostics information, and inertial sensors. The visual odometry is collected through a CPD camera (smartphone) which is mounted on the vehicle dashboard where the vehicle change of heading is estimated. On the other hand, the vehicle forward velocity information is acquired through On-Board Diagnostics II (OBD-II). In this paper, the proposed approach is a Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) loosely coupled integration scheme where the VO heading change and the forward OBD-II velocity are used to aid the low-cost INS during GNSS signal outage to limit its drift through Extended Kalman Filter (EKF). The employed visual odometry estimates the orientation change between successive frames which requires a feature extraction step to extract the features in every frame followed by primary matching to match the features and estimate the rotation between frames. Experimental road test is implemented, and the results show significant position accuracy enhancement using the proposed approach compared with typical Dead Reckoning (DR) solution where the average position Root Mean Square Error (RMSE) of the DR is around 14 meters for 60 s GNSS signal outage while the average position RMSE for the proposed navigation approach is 4.40 meters during the same GNSS signal outage.