For the errors in the original depth data obtained by TOF depth camera, which lead to image distortion, the method of establishing BP neural network space registration model was proposed to calibrate the depth data deviation caused by non-systematic random error in the process of TOF camera depth data measurement and systematic error of TOF camera measurement principle. The results are compared and analyzed with the calibration results based on the principle of keyhole imaging, BP algorithm is adopted to establish the space registration of BP neural network model, and its precision is higher than the calibration results based on the principle of keyhole imaging, stronger generalization ability, better able to recover the real scene depth data. It provides a new idea and method for the calibration of depth data measured by TOF camera.
[1]
Stephan Hussmann,et al.
Investigation of different polar to Cartesian coordinate system conversion methods for ToF-cameras in the close-up range
,
2015,
2015 IEEE International Instrumentation and Measurement Technology Conference (I2MTC) Proceedings.
[2]
Stefan May,et al.
Calibration and registration for precise surface reconstruction with Time-Of-Flight cameras
,
2008,
Int. J. Intell. Syst. Technol. Appl..
[3]
Mirko Schmidt,et al.
SRA: Fast Removal of General Multipath for ToF Sensors
,
2014,
ECCV.
[4]
卢纯青 Lu Chunqing,et al.
3D information acquisition and error analysis based on TOF computational imaging
,
2018
.