INVESTIGATION OF DIFFERENT LOW-COST LAND VEHICLE NAVIGATION SYSTEMS BASED ON CPD SENSORS AND VEHICLE INFORMATION

Recently, many companies and research centres have been working on research and development of navigation technologies for selfdriving cars. Many navigation technologies were developed based on the fusion of various sensors. However, most of these techniques used expensive sensors and consequently increase the overall cost of such cars. Therefore, low-cost sensors are now a rich research topic in land vehicle navigation. Consumer Portable Devices (CPDs) such as smartphones and tablets are being widely used and contain many sensors (e.g. cameras, barometers, magnetometers, accelerometers, gyroscopes, and GNSS receivers) that can be used in the land vehicle navigation applications. This paper investigates various land vehicle navigation systems based on low-cost self-contained inertial sensors in CPD, vehicle information and on-board sensors with a focus on GNSS denied environment. Vehicle motion information such as forward speed is acquired from On-Board Diagnosis II (OBD-II) while the land vehicle heading change is estimated using CPD attached to the steering wheel. Additionally, a low-cost on-board GNSS/inertial integrated system is also employed. The paper investigates many navigation schemes such as different Dead Reckoning (DR) systems, Reduced Inertial Sensor System (RISS) based systems, and aided loosely coupled GNSS/inertial integrated system. An experimental road test is performed, and different simulated GNSS signal outages were applied to the data. The results show that the modified RISS system based on OBD-II velocity, onboard gyroscopes, accelerometers, and CPD-based heading change provides a better navigation estimation than the typical RISS system for 90s GNSS signal outage. On the other hand, typical inertial aided with CPD heading change, OBD-II velocity updates, and Non-Holonomic Constraint (NHC) provide the best navigation result.


INTRODUCTION
Global Navigation Satellite System (GNSS) is the most commonly used navigation component in land vehicles where it provides a long-term accurate estimate for position and velocity states (Aggarwal et al., 2008). Unfortunately, GNSS signals suffer from blockage and multipath (Bancroft, 2009) in some operating conditions such as urban areas (Venkatraman et al., 2010) and foliage regions where the navigation solution is blocked or deteriorated. Inertial Navigation System (INS) provides reliable short-term full navigation estimates (position, velocity, and attitudes) for land vehicles (Iqbal et al., 2010). However, the navigation solution is degraded after a short time due to the INS drift especially for Micro-Electro-Mechanical Systems (MEMS) based INS because of its accelerometers and gyroscopes large errors and noise characteristics (Abd Rabbou and El-Rabbany, 2015). Therefore, GNSS and INS are integrated to overcome the shortcomings of each sensor and to provide a more reliable navigation solution in both short and long-term periods (Niu et al., 2007). GNSS/INS has a defect when GNSS signal is blocked for a long time where INS standalone navigation solution will deteriorate quickly due to INS drift. Therefore, INS should be aided with other sensors to mitigate its large drift and to provide a better navigation estimation. Many aiding sensors are used for land vehicle navigation such as odometers, ultrasonic sensors (M Moussa et al., 2019) (Moussa et al., 2018), magnetometers (Won et al., 2015), Light Detection And Ranging (LIDAR) (Gao et al., 2015) (Tang et al., 2015), cameras (Zhenbo Liu, 2019) (Liu, et al., 2018), and Radio Detection And Ranging (RADAR) (Bo et al., 2018). Unfortunately, there are many drawbacks for using these sensors such as the magnetic interference in the case of magnetometers, the high price, high computational and processing cost, and the environmental effects in the case of LiDAR and cameras. Finally, sensor installment is a very hard process for the ultrasonic. Maps aiding navigation is used in many previous researches to help low-cost INS in GNSS denied environments (Attia, 2013). Such map aiding depends on the availability and update rate of the required maps. Consumer Portable Devices (CPDs) are widely used all over the globe. CPDs contain many sensors such as GNSS receivers, low-cost INS, magnetometer, barometer, and camera that can be used in many land vehicles applications such as navigation, lane localization (Song et al., 2017) (Zhu et al., 2017), environmental perception, safety driving monitoring, insurance telematics (Wahlström et al., 2017), and road surface condition monitoring (Sathe and Deshmukh, 2017). RISS is a Dead Reckoning (DR) navigation system which is categorized into 2D and 3D RISS where 2D RISS consists of one gyroscope and a source of land vehicle forward velocity such as odometer. On the other hand, 3D RISS consists of one gyroscope and two-axis accelerometers along with an odometer (Noureldin et al., 2013). RISS has been addressed in many previous researches, (Iqbal et al., 2008) described the mechanization of RISS as well as its integration with GNSS through Kalman Filter (KF). 3D RISS/GNSS integrated system through Particle Filter (PF) was addressed in (Georgy et al., 2010a). (Georgy et al., 2010b) integrated RISS along with GNSS through a tightly coupled fusion scheme with PF and compared the results with KF. Some researchers worked on enhancing the RISS in GNSS denied environment fusing other sensors such as magnetometers to offer a heading update to aid RISS during GNSS signal outage through EKF (Abosekeen et al., 2019). CAN (Controller Area Network) bus provides some useful information about the vehicle dynamics such as the forward velocity and steering angle data. However, the commercial On-Board Diagnostics (OBD-II) provides the velocity information and does not typically provide the steering angle data unless additional customized hardware and software designs are developed (Xiao et al., 2018). Therefore, a new method for estimating the steering angle was proposed by  through CPD accelerometers and then the land vehicle change of heading is estimated. The main objective of this paper is to investigate different lowcost land vehicle navigation systems based on CPD sensors and land vehicle information. These navigation systems are based on DR, RISS, and loosely coupled INS/GNSS integrated systems. The outcome of this paper will not only help land vehicle navigation applications but will also open the door for many mobile mapping applications based on CPD sensors.

METHODOLOGY
The methodology consists of five subsections: land vehicle heading change estimation using CPD accelerometers, DR navigation system based on heading change estimated by CPD accelerometers and OBD-II velocity, DR navigation system with gyroscope updates, 3D RISS, and typical loosely coupled GNSS/INS with CPD heading change and OBD-II velocity updates during GNSS signal outage.

CPD accelerometers Heading Change Estimation
Land vehicle heading change estimation using CPD accelerometers is described in details in . CPD is mounted on the vehicle steering wheel to estimate the steering angle through CPD accelerometers as shown in Figure  1. The steering angle computation should be compensated for the leveling of the onboard INS, the steering wheel inclination, the vehicle acceleration, and the vehicle inclination. However, all these factors may be ignored if only one CPD is used in the navigation state estimation, i.e. when no onboard INS is used. Therefore, the steering wheel angle is estimated as shown in equation 1.
Where δ is the steering wheel inclination angle, g is the gravity acceleration, axv', ayv', and azv' are the vehicle onboard accelerometers of x, y and z-axis compensated from the vehicle inclination. The vehicle heading change is estimated in equation 3.
Where VSR is the Vehicle Steering Ratio which should be constant for each type and model of vehicles. Δθvehicle is the change of heading. During GNSS signal availability, the reference navigation whether it is from GNSS only or GNSS/INS integrated system provides a reference heading change to model the errors (bias and scale factor) of the heading change estimated from CPD accelerometers as shown in Figure 3.

DR navigation system based on CPD heading change and OBD-II velocity
The evaluated dead reckoning scheme is based on the estimated heading change by the CPD accelerometers and the vehicle forward velocity information obtained from a commercial On-Board Diagnosis II (OBD-II). Figure 4 depicts this DR navigation system.

Figure 4. DR navigation system scheme based on CPD heading change and vehicle information
During GNSS signal availability, GNSS provides the land vehicle navigation solution in addition to the reference heading to calibrate the errors of the CPD heading change. On the other hand, During GNSS signal outages, the OBD-II forward velocity along with the change of heading estimated from the CPD forms a DR navigation system. It is worth mentioning here that this proposed navigation solution doesn't require any on-board sensors as it is based only on the CPD mounted on the steering wheel and a commercial OBD-II which makes it a very low-cost navigation system.

DR navigation system based on CPD heading change, gyroscopes measurements and OBD-II velocity
This proposed navigation system is based on the vehicle information gathered from a commercial OBD-II and the heading change computed from the CPD accelerometer and a calibrated z-gyroscope from an on-board low-cost INS as shown in Figure 5. An Integrated Heading Change (IHC) is computed using both the CPD heading change and the calibrated z-gyroscope angular rate. The DR system works during GNSS signal outage while the GNSS provides the navigation solution during its availability and calibrates the z-gyroscope as well the CPD heading change errors. The proposed navigation system is applied when there is an on-board INS, as well as a CPD, fixed on the steering wheel.

Typical 3D RISS navigation system
The main idea of the 3D RISS is described in (Noureldin, Aboelmagd, Karamat, Tashfeen and Georgy, 2013) and (Iqbal et al., 2008), it has been proposed to reduce some errors of the full inertial sensors system in addition to the reduction of the cost by using fewer sensors. 3D RISS systems consist of two accelerometers that are mounted in the forward and transverse moving directions in addition to a vehicle odometer as well as a vertical gyroscope which is mounted in the z-direction. The main function of the vertical gyroscope is to estimate the azimuth angle as shown in equation (4) and (5) (Noureldin, et al., 2013) tan sin Where A is the azimuth angle, ωz is the angular rotation rate in the z-direction measured by the vertical gyroscope, ωe is the earth angular rotation, ϕ is the latitude, ve is the velocity in the east direction, RN is the prime vertical radius of curvature, h is the height and dt is the epoch time interval. The rate of change of the azimuth angle is equal to the angular rotation measured by the vertical gyroscope considering two factors that should be compensated which are the earth rotation component in the z-direction and the change of the orientation of the Local Level Frame with respect to the Earth Fixed Frame (LLF w.r.t EFF) in the z-direction. The main function of the odometer is to measure the velocity in the vehicle forward direction and thus the velocity in the east, north, and up directions can be determined as shown in the following equations (Noureldin, et al., 2013).
sin cos Where ve, vn, and vu are the velocity in the east, north, and up directions respectively, while vod is the velocity measured by the odometer and p is the pitch angle. Finally, the function of the forward and the lateral accelerometers is to estimate the pitch (p) and the roll (r) angles. Originally, the forward accelerometer (fy) measures the forward vehicle acceleration which can be calculated using the odometer as (aod) as well as the gravity acceleration component in the forward direction and therefore the pitch angle can be calculated using equation (9). On the other hand, the roll angle can be estimated through equation (10).
Where fy and fx are the specific forces measured by the forward and the lateral accelerometers respectively. Finally, the position is calculated as follows: where RM is the meridian radius of curvature. The typical 3D RISS integration scheme is depicted in Figure 6. Figure 6. Typical 3D RISS navigation system scheme based on onboard sensors and vehicle information.

Modified 3D RISS navigation system
The proposed modified 3D RISS navigation system is based on OBD-II velocity, onboard gyroscopes, accelerometers, and CPD-based heading change as shown in Figure 7. The CPD is mounted on the steering wheel to estimate the steering wheel angle and then the vehicle heading change is calculated. The Integrated Azimuth Change is calculated using a weighted mean between the CPD heading change and the zgyroscope angular rate.

Loosely coupled GNSS/INS with different update types
Typical GNSS/INS loosely coupled integration scheme is based on GNSS that provides the navigation filter with position and velocity updates to aid the INS. During GNSS signal outages, INS should be aided to mitigate its large drift. Different update types are investigated which are: the forward velocity, the CPDbased change of heading, the z-gyroscope angular rate, the magnetometer heading, and Non-Holonomic Constraint (NHC) updates and different combinations between these updates. Figure 8 shows the GNSS/INS integration scheme with vehicle forward velocity and CPD heading change updates.
 where δP, δv, and δα are the position, the velocity, and the attitude error states respectively. ba and bg are the biases of the accelerometers and the gyroscopes respectively. Finally, Sa and Sg are the scale factor of the accelerometers and gyroscopes respectively. There are two main stages of the KF-prediction and update stages. The system model describes the time evolution of the navigation states and is responsible for the prediction stage. On the other hand, the measurement model provides updates to the navigation filter. The KF models and stages equations are depicted in Figure 9.

Where
x is the rate of change of the state vector, F is the dynamics matrix, x is the error state vector, G is the shaping matrix, and wk is white noise. ϕk,k+1 is the transition matrix, I is the identity matrix and Δt is the time interval, Qk is the process noise matrix which defines the uncertainty of the system model. Finally, Pk is the states' covariance matrix. (-) refers to the predicted elements, and finally, (+) refers to updated elements, zk is the observation vector, Hk is the design matrix, ηk is the measurement noise. Rk is the covariance matrix of the measurement noise which describes the uncertainty of the observations.

EXPERIMENTAL RESULTS
An experimental real data set was collected using Pixhawk 4 board which consists of a GNSS u-blox Neo-M8N receiver and ICM-20689 Invensense low-cost IMU in which its x-axis coincides with the forward vehicle direction. An iPhone 6 is used in the experiment which is mounted on the steering wheel of a Ford Focus car. A low-cost OBD-II interface (Uni-link Mini ELM327 OBD-II Bluetooth Scanner Tool) is used to access the regular odometer data from the vehicle. The experiment was performed in an outdoor parking lot where different simulated GNSS signal outages were applied to the trajectory to show the impact of proposed techniques on the final navigation solution. Pixhawk 4 GNSS/INS integration is the reference navigation solution with sub-meter accuracy. The experimental results are divided into two subsections which are the heading change regression model, and the heading change estimation results and different navigation systems results.

Heading Change Regression Model and Estimation Results
The steering wheel angle is estimated using CPD accelerometers where the CPD is attached to the steering wheel angle then the vehicle heading change is calculated using the estimated steering wheel angle. Linear regression is implemented to fit a model between the estimated heading change and the reference values. Figure 10 shows the regression model between the steering wheel angle and the reference heading change. Figure 10. CPD regression model between the steering angle and the reference change of heading The reference heading change is estimated either from the GNSS velocity or from GNSS/INS integration when using the typical loosely coupled integration scheme. Figure 11 exhibits the heading change estimated by the proposed method and the reference heading change. Figure 11. CPD change of heading versus the reference heading change The difference between the change of heading estimated by the CPD accelerometers and the reference heading change is computed to assess the proposed method. The RMSE of the heading change estimation is 1.54 degrees/sec for around 290 seconds.

Navigation States Estimation Results
Navigation estimation results section is divided into three subsections which are: the dead reckoning results, the 3D RISS results, and results of loosely coupled INS integration with different updates. The vehicle trajectory consists of three loops where the outages are chosen for the second and/or third loop.

DR based on CPD heading change and OBD-II velocity results
DR navigation system is based on the land vehicle heading change estimated from the CPD accelerometers that are mounted on the vehicle steering wheel. The velocity information is obtained from the regular odometer from OBD-II. Figures 12 and 13 show the DR trajectory along with the reference for two GNSS outage periods of 90 seconds and 180 seconds respectively. The DR trajectory is close to the reference trajectory during the 90 seconds outage period while it deviates at the beginning of the second 90 seconds.

DR based on CPD heading change, gyroscopes measurements and OBD-II velocity Results
This proposed navigation system depends on the CPD accelerometers as well as the on-board IMU gyroscope for providing heading change information to the system while the OBD-II provides the vehicle forward velocity. Figure 14, and 15 depicts the DR trajectory with the reference and the outage period. If the system depends only on the gyroscope in providing the heading change information along with the OBD-II velocity, the position RMSE is around 7.80 meters for 90 seconds GNSS signal outage. However, integrating the heading change information from both the CPD sensors and onboard gyroscopes provides a better position estimation.

Typical 3D RISS Navigation Results
Typical 3D RISS navigation system is based on the compensated on-board z gyroscope, the x and y accelerometers and the forward velocity from a regular odometer. Figures 16  and 17 show the typical 3D RISS trajectory along with the reference trajectory and the GNSS outage period for 90 seconds and 180 seconds GNSS signal outages respectively.  The navigation state estimation is improved where the average position RMSE is around 5.00 meters for 90 seconds GNSS signal outage.

Modified 3D RISS Navigation Results
Modified 3D RISS depends on the compensated on-board z gyroscope along with the CPD mounted on the steering wheel for providing the land vehicle heading change. Figure 18 and 19 exhibits the modified 3D RISS, reference trajectory and the outage period.  Aiding low-cost INS with OBD II velocity, NHC, and CPD heading change provides the best navigation state estimation where the RMSE of the position is around 3.1 and 6 meters for 90s and 180s GNSS signal outage respectively.

CONCLUSIONS
Different land vehicle navigation systems were implemented and investigated using CPD sensors, onboard sensors, and vehicle information. The navigation systems are based on different DR, RISS, and typical GNSS/INS integration systems with different updates. These navigation systems could be implemented in a GNSS denied environment such as in urban canyons, foliage areas, tunnels, and underground parking. Typical RISS shows better navigation results than the DR systems for 90s and 180s GNSS signal outage while modified RISS provides a better navigation solution than the typical RISS system. Finally, CPD heading change/OBD-II velocity/NHC integrated aiding system with INS provides the best navigation state estimation for various GNSS signal outages.