Modern smartphones contain embedded global navigation satellite systems (GNSSs), inertial measurement units (IMUs), cameras, and other sensors which are capable of providing user position, velocity, and attitude. However, it is difficult to utilize the actual navigation performance capabilities of smartphones due to the low-cost and disparate sensors, software technologies adopted by manufacturers, and the significant influence of environmental conditions. In this study, we proposed a scheme that integrated sensor data from smartphone IMUs, GNSS chipsets, and cameras using an extended Kalman filter (EKF) to enhance the navigation performance. The visual data from the camera was preprocessed using oriented FAST (Features from accelerated segment test) and rotated BRIEF (Binary robust independent elementary features)-simultaneous localization and mapping (ORB-SLAM), rescaled by applying GNSS measurements, and converted to velocity data before being utilized to update the integration filter. In order to verify the performance of the integrated system, field test data was collected in a downtown area of Tainan City, Taiwan. Experimental results indicated that visual data contributed significantly to improving the accuracy of the navigation performance, demonstrating improvements of 43.0% and 51.3% in position and velocity, respectively. It was verified that the proposed integrated system, which used data from smartphone sensors, was efficient in terms of increasing navigation accuracy in GNSS-challenging environments.
Global navigation satellite systems (GNSSs) are commonly used for navigation and mapping applications. However, in GNSS-hostile environments, where the GNSS signal is noisy or blocked, the navigation information provided by a GNSS is inaccurate or unavailable. To overcome these issues, this study proposed a real-time visual odometry (VO)/GNSS integrated navigation system. An on-line smoothing method based on the extended Kalman filter (EKF) and the Rauch-Tung-Striebel (RTS) smoother was proposed. VO error modelling was also proposed to estimate the VO error and compensate the incoming measurements. Field tests were performed in various GNSS-hostile environments, including under a tree canopy and an urban area. An analysis of the test results indicates that with the EKF used for data fusion, the root-mean-square error (RMSE) of the three-dimensional position is about 80 times lower than that of the VO-only solution. The on-line smoothing and error modelling made the results more accurate, allowing seamless on-line navigation information. The efficiency of the proposed methods in terms of cost and accuracy compared to the conventional inertial navigation system (INS)/GNSS integrated system was demonstrated.