160 likes | 324 Views
Fusion of Time-of-Flight Depth and Stereo for High Accuracy Depth Maps. Reporter : 鄒嘉恆 Date : 2009/11/17. Introduction. They provide real time depth estimates by combine stereo and Time-of-flight. Outline. Motivation Multi-sensor calibration Geometric calibration Photometric calibration
E N D
Fusion of Time-of-Flight Depth and Stereo for High Accuracy Depth Maps Reporter:鄒嘉恆 Date:2009/11/17
Introduction • They provide real time depth estimates by combine stereo and Time-of-flight.
Outline • Motivation • Multi-sensor calibration • Geometric calibration • Photometric calibration • Calibration verification • Sensor fusion • Experimental results • Conclusion
Motivation • Problem: • Laser scanner are too slow for real time use. • Stereo fails on texturelessscenes. • TOF is low resolution, noisy, and poorly calibrated.
Multi-sensor setup • SR3000: • 176x144 • Operational range up to 7 meters • 2 CCD stereo cameras • limit the calibration range from 1m to 1.4m
Multi-sensor calibration • Geometric calibration
Multi-sensor calibration • Photometric calibration
Multi-sensor calibration • Calibration verification-plane experiment
Multi-sensor calibration • Calibration verification-plane experiment
Multi-sensor calibration • Calibration verification-plane experiment • Calibration verification-box experiment
Sensor fusion • fd : the local evidence for node i • fs : a symmetric function measures the smoothness assumption about the scene • fr: the additional local evidence based on the measurement from the TOF sensor.
Experimental results • Using TOF • (A)local method • (B)local method with LUT • (C)global method • (D)global method with LUT • Using stereo • (E)local method • (F)global method • (G)local fusion method • (H)global fusion method
Conclusion • They present a simple and effective calibration method to improve the performance of time-of-flight sensors. • Their method can achieve an absolute accuracy of about 5mm over a range of one meter. • The fusion results reduce the over-all error by 50%.