Arvind Ramanandan

Real-Time Computer Vision/DGPS-Aided Inertial Navigation System for Lane-Level Vehicle Navigation

Many intelligent transportation system (ITS) applications will increasingly rely on lane-level vehicle positioning that requires high accuracy, bandwidth, availability, and integrity. Lane-level positioning methods must reliably work in real time in a wide range of environments, spanning rural to urban areas. Traditional positioning sensors such as the Global Navigation Satellite Systems may have poor performance in dense urban areas, where obstacles block satellite signals. This paper presents a sensor fusion technique that uses computer vision and differential pseudorange Global Positioning System (DGPS) measurements to aid an inertial navigation system (INS) in challenging environments where GPS signals are limited and/or unreliable. To supplement limited DGPS measurements, this method uses mapped landmarks that were measured through a priori observations (e.g., traffic light location data), taking advantage of existing infrastructure that is abundant within suburban/urban environments. For example, traffic lights are easily detected by color vision sensors in both day and night conditions. A tightly coupled estimation process is employed to use observables from satellite signals and known feature observables from a camera to correct an INS that is formulated as an extended Kalman filter. A traffic light detection method is also outlined, where the projected feature uncertainty ellipse is utilized to perform data association between a predicted feature and a set of detected features. Real-time experimental results from real-world settings are presented to validate the proposed localization method.

Inertial Navigation Aiding by Stationary Updates

Sensor-aided inertial navigation has successfully been used for decades for localization of a roving body. When the rover is known to be stationary, artificial “stationary” measurements (i.e., zero velocity and/or zero angular rate) may be imposed. This corrects the velocity, attitude, and inertial measurement unit (IMU) biases, which decreases the rate of drift of the position and attitude. Implementation requires reliable automated tests to detect periods when the vehicle is stationary. Due to cost concerns, methods that use sensors that are already on the vehicle are preferred. This paper reviews existing stationary detection methods and proposes a new frequency domain approach, using only IMU data, to detect stationarity, with specifications and analysis for land vehicles. The performance of this new approach is evaluated in both theory and practice. In addition, this paper presents analytic and numeric evaluations of the observability of the inertial navigation system (INS) error states with stationary updates. Improvements in localization performance in an INS with stationary detection and aiding is shown experimentally.

High-precision lane-level road map building for vehicle navigation

Lane relative vehicle navigation and control requires accurate lane-relative positioning of the vehicle. This relative position can be computed by comparing the vehicle absolute position with analytic roadway maps, which requires both high-accuracy positioning of the vehicle and high-accuracy lane maps. Carrier Phase Differential GPS (CPDGPS) aided INS or CPDGPS aided encoders is capable of estimating vehicle absolute position (relative to earth center) with centimeter level accuracy; however, to the best of the authors knowledge, the accuracy of lane level maps is currently not sufficient. In this paper, we first consider the structure of lane level maps that are compatible with standard practices of GIS road modeling. Then, various analytic lane definition are discussed. We also present a method of building lane level maps from high-accuracy positioning data along the lane center. The data is segmented according to road intersections. Shape points (vertices) as a function of arclength are located based on changes in estimated curvature. For each segment, the parameters are estimated by least-square criteria and can be refined as new datasets become available. This process is shown by an example.

A near-real time nonlinear state estimation approach with application to initialization of navigation systems

The performance of any linearization based estimation algorithm like the Extended Kalman Filter (EKF) relies heavily on the accuracy of the nominal trajectory about which the system is linearized. When the linearization assumption does not hold, such an algorithm behaves in an unpredictable fashion and metrics of estimation error (i.e. state covariance) are invalid. This paper presents methods to identify in real-time those parts of the state vector whose uncertainties cause significant deviations from the linearized model and proposes a near-real time approach to address the issue. One important class of applications is initialization of navigation systems; therefore, as an example the paper applies the results of the theory to a simplified 7 state, two dimensional GPS aided INS. The near-real time approach is demonstrated in simulation.

Near-real-time GPS integer ambiguity resolution

Real-time decimeter accuracy GPS positioning can be achieved using carrier phase measurements. This requires fast and reliable on-the-fly integer ambiguity resolution. However, in GPS challenged areas (e.g. Urban canyons, tunnels, thick canopy etc.) the GPS receiver may not be able to track a sufficient number of satellites to resolve the integer ambiguities within one epoch. In this paper, we would like to find the optimal solution by combining the measurements from several epochs. In this paper, we present the theoretical derivation for a fast and efficient method for GPS integer ambiguity resolution with multiple GPS epoch measurements. Simulation results show the effectiveness of the proposed approach.

Observability analysis of an inertial navigation system with stationary updates

Stationary measurement updates offer the possibility of containing errors in velocity, biases and attitude even during periods of GPS unavailability, given the information that the rover is not moving. Automated detection of appropriate instances of stationarity is possible, given a recently proposed frequency domain method based on IMU data. In this paper we focus on analytical and numerical evaluation of observability of error states of an INS aided with stationary updates. The null space of the continuous time observability gramian is evaluated for various motion scenarios typically occurring in a land vehicle.

INS aided GPS integer ambiguity resolution

Real-time high precision GPS positioning is based on carrier phase measurements, which requires fast and precise on-the-fly integer ambiguity resolution. In some navigation applications, external sensors are available that provide auxiliary measurements. For example, in GPS/INS navigation systems, the inertial sensors allow computation of prior position estimates when GPS signals become available. This information can be used to aid GPS integer ambiguity resolution, offering a higher probability of obtaining correct integers, especially in challenging GPS conditions (e.g., few satellites, high measurement noise). This paper describes a fast and efficient technique for integer ambiguity resolution when auxiliary information from INS is available. The theoretical derivation will be presented and simulation result will show the effectiveness of the proposed method.

Observability analysis of INS and lever-arm error states with CDGPS - Camera aiding

Lane level guidance is required for several Intelligent Transportation Systems (ITS) applications. To perform lane level guidance, vehicle localization accuracies on the decimeter level are required. Carrier phase Differential GPS integrated Inertial Navigation System (CDGPS - INS) is able to achieve such accuracies given a clear view of the sky and differential correction availability; however, when using an inexpensive Inertial Measurement Unit (IMU), the performance deteriorates beyond the meter level in tens of seconds in GPS denied areas. Another popular method of navigation in the literature is INS with feature-based aiding. One of the fundamental difficulties in integrating any feature-based sensor with INS is the necessity to calibrate, with commensurate accuracy, the lever arm vector and rotation matrix, i.e. extrinsic parameters, referred here as frame transformation parameters, from the INS to the feature sensor frame. This calibration can occur during the systems operation when the corresponding error state is observable. The primary objective of this paper is to analyze observability of INS error states in an integrated CDGPS - vision - INS approach and compare it with observability in CDGPS - INS and a vision - INS approaches.