B-spline based continuous-time multi-sensor calibration

Abstract

Autonomous robots, for example, self-driving cars, are often equipped with a set of heterogeneous sensors of different sensing modalities for estimation and perception. To correctly fuse multi-modal sensory information, accurate calibration of these sensors becomes critical. Recent developments in the field of sensor calibration have led to a growing interest in continuous-time trajectory representation over discrete-time formulation. In this project, we leverage B-Spline for continuous-time trajectory representation and aim to develop efficient and accurate multi-sensor extrinsic calibration for autonomous vehicles. The key idea of the proposed method is centered on the B-spline theory that naturally is suitable for fusing asynchronous measurements. Any pose on the trajectory can be represented based on the B-Spline theory as long as we have the correct control points of it and interpolate with times. To see this, we explain how to implement B-spline with sensor calibration and show the procedure in different simulated scenarios, validating that the proposed method is able to achieve accurate results.