Transcript thesis
Sukhum Sattaratnamai Advisor: Dr.Nattee Niparnan 1 Outline Introduction Objective Calibration Process Our Work Improving Laser Data Automate Data Collection Conclusion 2 LRF-Camera System p α p(u, v) d 3 LRF-Camera System pm AR | t P [R | t] p α d XL u YL v AR | t ZL 1 1 4 LRF-Camera Calibration Problem Definition Find the transformation [R |t ] of the camera w.r.t. LRF 5 Objective Related Work Proposal LRF-Camera Calibration Improving Laser Data Calibration of a multi-sensor system laser rangefinder/camera, 1995 More Accurate Result Extrinsic calibration of a camera and laser range finder (improves camera calibration), 2004 Filtering Laser Data Easier Process An algorithm for extrinsic parameters calibration of a camera and a laser range finder using line features, 2007 6 Objective Related Work Proposal Thesis LRF-Camera Calibration Improving Laser Data Improving Laser Data Calibration of a multi-sensor system laser rangefinder/camera, 1995 On Improving Laser Data for Extrinsic LRF/Camera Calibration, 2011 More Accurate Result Extrinsic calibration of a camera and laser range finder (improves camera calibration), 2004 Easier Process Filtering Laser Data Automated Process Automated Calibration Data Collection in LRF/Camera Calibration with Online Feedback, 2012 An algorithm for extrinsic parameters calibration of a camera and a laser range finder using line features, 2007 7 Objective Related Work Proposal Thesis LRF-Camera Calibration Improving Laser Data Improving Laser Data Calibration of a multi-sensor system laser rangefinder/camera, 1995 On Improving Laser Data for Extrinsic LRF/Camera Calibration, 2011 More Accurate Result Extrinsic calibration of a camera and laser range finder (improves camera calibration), 2004 Easier Process Filtering Laser Data Automated Process Automated Calibration Data Collection in LRF/Camera Calibration with Online Feedback, 2012 An algorithm for extrinsic parameters calibration of a camera and a laser range finder using line features, 2007 8 Calibration Process Start Data Collection Feature Detection Optimization Check Result End 9 Calibration Process Data Collection 10 Calibration Process Feature Detection 11 Calibration Process Projection Error E(M ext , pl ) P(M ext , pl ) pm 2 12 Calibration Process Optimization Simulated Annealing : Find global minimum Levenberg-Marquardt : Find local minimum 13 Calibration Process Result Project laser data onto an image 14 Our Work Improving Laser Data Automatic Data Collection 15 Improving Laser Data Angular Error [0, ) => 2 16 Simulation Angular Error [ 2 , 2) => 4 17 Simulation Laser Data Improvement Target Method average RMS S.D. of RMS * E(M ext , pl ) imp plain ratio 0.32 0.54 54.8% 0.007 0.032 22.1% imp 0.92 0.003 pm plain 2.04 0.019 ratio 48.0% 16.5% 18 Experiment Laser Range Finder Camera 19 Experiment Laser Data Improvement Target Method average RMS S.D. of RMS Stingray imp plain ratio 1.66 2.90 57.3% 0.02 0.04 40.7% imp 6.18 0.19 Legria plain ratio 11.38 54.3% 0.48 38.8% 20 Experiment Number of Data 21 Improving Laser Data Lower bound u f x ( X c / Z c ) cx v f y (Yc / Z c ) c y e0 f 4 22 Simulation Lower Bound การทดลอง 1 2 3 4 fx e0 270 540 270 540 0.5 0.5 1.0 1.0 0.59 1.18 1.18 2.36 RMS 0.55 1.09 1.09 2.18 Ratio 92.9 92.9 92.3 92.3 23 Automate Data Collection Start Feature Detection 5 นาที 2 นาที 30 นาที Optimization 1 วินาที Data Collection Check Result End 24 Automate Data Collection Feature Detection 25 Automate Data Collection False Detection => Tracking 26 Experiment Data Distribution บริ เวณ ซ้าย กลาง ขวา ทั้งหมด ปรับแก้ ค่าเฉลี่ย ค่าเบี่ยงเบน 0.44 0.002 0.39 0.003 0.50 0.003 0.44 0.005 ทดสอบ ค่าเฉลี่ย ค่าเบี่ยงเบน 1.03 0.063 0.78 0.037 1.16 0.078 0.69 0.011 27 Automate Data Collection Working Space Covering Data Bin (x, y, angle) 28 Automate Data Collection Moving Calibration Object => Velocity Metric 29 Experiment Velocity & Accuracy 30 Experiment Accuracy & Time 31 Experiment 32 Automate Data Collection User Interface Data Quality Metric Tracking, Velocity Data Distribution Data Bins, Current Bin, Target Bin 33 Automate Data Collection User Interface Result Laser Data Projection Acknowledge & Warning Sound Data Acquire, Tracking Lost 34 Conclusion Improved calibration method Reduce projection error to 50 percent Automatic data collection process Faster and easier for all user 35 36