Autonomous Ground Systems CROSS CORRELATING GROUND-LEVEL PANORAMAS WITH SATELLITE IMAGERY FOR GPS-DENIED LOCALIZATION OF AUTONOMOUS GROUND VEHICLES Calvin Cheung Stanley Baek U.S. Army Tank Department of Electrical and Computer Army Research Development Engineering Engineering Center University of Michigan Warren, MI Dearborn, MI 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited.
Autonomous Ground INTRODUCTION Systems • Localization – Where am I? ??? 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 2
Autonomous Ground INTRODUCTION Systems • Localization - GPS 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 3
Autonomous Ground INTRODUCTION Systems • Localization – GPS Degradation • Occlusion • Spoofing • Denial 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 4
Autonomous Ground INTRODUCTION Systems • Localization – GPS-Denied Localization 8/9/2018 5
Autonomous Ground RELATED WORK Systems Simultaneous Localization and Mapping Landmark Recognition *Image from http://www.pirobot.org/blog/0015/ Different Sensing Modalities SIFT Matching *Cornick, Koechling, Byron, Zheng. “Localizing Ground Penetrating RADAR: A Step Toward Robust Autonomous Ground Vehicle Localization” 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 6
Autonomous Ground LOCALIZATION PROCESS Systems 1. Capture a spherical panoramic image on a vehicle and warp it to produce a top-down view. 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 7
Autonomous Ground LOCALIZATION PROCESS Systems 2. Run Canny edge detection on both the top-down view and the satellite image and perform normalized cross correlation on the resulting images. 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 8
Autonomous Ground LOCALIZATION PROCESS Systems 3. Filter matches based on distance from the last known good point. 4. Apply Kalman filter step to matched point and report the location. 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 9
Autonomous Ground EXPERIMENT Systems • Panoramic images downloaded from Google Street view • Suburb of Sonoma Peak chosen to test localization process 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 10
Autonomous Ground RESULTS Systems : Ground Truth : Cross-correlation Matches : Kalman Filtered Results. 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 11
Autonomous Ground RESULTS Systems 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 12
Autonomous Ground RESULTS Systems Localization Performance RMSE for CC RMSE for CC RMSE for KF RMSE for KF KF Mean along along East/ along East/ West along Distance from North/South West (m) (m) North/South (m) Truth (m) (m) 72.32 480.33 71.97 398.49 253.78 Distance Filtering 97.72 194.70 36.33 136.21 115.65 Canny Edge Detection Canny Edge Detection, 36.47 81.53 20.69 52.91 46.95 Distance Filtering 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 13
Autonomous Ground CONCLUSIONS Systems • Developed a method of GPS-denied localization by performing normalized cross-correlation on satellite imagery and warped spherical panoramic • Using map data from Google Maps and Google Street View, we demonstrated localization ~50 m of ground truth • Method can work with existing GPS solutions (<7.8 m accuracy) to help maintain coordinates during intermittent GPS drops • In complete GPS-denied environments, the localization process can provide a great benefit for AGVs looking to approximate position 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 14
Autonomous Ground THANKS! Systems • Questions/comments? 8/9/2018 DISTRIBUTION STATEMENT A. Approved for public release; distribution unlimited. 15
Recommend
More recommend
