GPS World, May 2018

CONNECTED CAR 18 FIGURE 6 Vehicles heading during lane change to left where H is defined as follows 19 Lane Estimation When the road segments have been accurately selected based on the filtered vehicles pose the projection of the vehicles positions on segment lanes can be easily calculated knowing the lane widths and number of lanes The sum of squared errors for each lane is then calculated by 20 where N is number of epochs and p v is the projection of vehicles position on lane The lane associated with the minimum error is selected as the designated lane Lane Change Detection If a lane change occurred within the processed buffer of data the least squares regression will not converge to the correct lane Therefore the buffer needs to be partitioned at the lane switch locations Thus a lane change detection module is developed In this work 30 GPS WORLD WWW GPSWORLD COM MAY 2018 FIGURE 8 Overall block diagram of the proposed system a lane change detection method is designed based on capturing the patterns of the vehicles orientation and raw gyroscope measurements The heading and raw gyroscope measurements during lane changes are shown in FIGURE 6 and FIGURE 7 The general pattern that the lane change module detects is a peak or a valley in azimuth accompanied by a peak valley or valley peak sequence in the gyroscope measurements To detect peaks and valleys the standard deviation of a moving window of data is calculated and compared to a peak valley threshold If both gyro and azimuth peak valley sequence are consistent and matched with the pattern described above a lane change is declared Two algorithm phases of processing are then applied Acquisition Phase GNSS and IMU measurements are fused in the main EKF and HMM map matching is performed and a lane is estimated The innovation sequence of the main EKF which is the difference between the predicted state and GNSS updates is calculated over a buffer of data If the innovation sequence is within a small threshold and no lane change has been detected the acquisition phase is concluded and the tracking phase begins Tracking Phase Two EKF filters are initiated One EKF accepts position updates from the projection of the vehicles position on the selected lane and the other EKF accepts GNSS position updates only A discrepancy measure is evaluated between the two EKF instances for a short window of time If this discrepancy measure is higher than a threshold a temporary GNSS deviation is assumed and the system keeps reporting the current lane as the designated lane If GNSS measurements started to be centered again on the new lane a lane change is confirmed and the output of the first EKF instance will be the correct state Otherwise this lane change is declared as false and the second EKF output is the correct output The overall block diagram of the proposed system is shown in FIGURE 8 FIGURE 7 Vehicles gyroscope measurements during lane change to left