Modelling our way out of a featureless correspondence problem for automatic calibration of laser stripe mapping systems

This paper discusses a problem in predicting observations for laser stripe mapping systems that is akin to the correspondence problem for stereo imagery and structure from motion. Solving this problem is essential for allowing robust, efficient, featureless calibration of laser stripe extrinsic parameters using ordinary mapping data, similar to a bundle adjustment and referred to here as an automatic calibration. Such a calibration is important for laser stripe systems to calibrate them in-situ using natural scenes without additional hardware or calibration routines. We propose a simple modelling approach that ensures single predictions are made per observation and we highlight the implications for the design of laser stripe systems that would make simple laser line extraction algorithms more robust. We present the results of an automatic calibration using this modelling approach for simulated observations of entirely featureless terrain by a laser stripe mapping system; proving the concept of a robust, efficient, featureless, automatic calibration of laser stripe extrinsics.