Automatic extrinsic calibration of terrestrial laser scanner and digital camera by MoG image correlation

Abstract

The current terrestrial laser scanners (TLS) are generally equipped with digital cameras which can capture the scene along with the scanner. These two types of sensors offer complementary properties in modeling and visualization of real-world scenes. TLSs can provide geometric information of the real scene with accurate 3D coordinates of the point clouds; cameras are used to acquire high-resolution images and provide good texture information of the environment. Fusing the extracted information from these two sensors helps to create a better virtual representation of the real-world. For a TLS with several external cameras, their acquisition centers are not identical and the axis of their coordinate systems are not aligned either. This paper proposes an automatic camera and TLS extrinsic calibration approach using correspondences extracted from both measurements. To overcome the intrinsic difference between back-projected images of point clouds colored by intensities and the RGB camera images, we innovatively generate both magnitude of gradient images, enabling effective image correlation and accurate correspondence extraction. The 3 external cameras mounted on top, side and bottom of Leica RTC360 3D laser scanner are calibrated. Dependent on the distribution of observations, we achieve different calibration accuracy for each camera. With scans from multiple stations, the cameras obtain an offset accuracy of 0.12 – 0.36 mm and angular accuracy of 3.7 – 8.3″. After calibration, the excellent overlap of images from the two sensors further verifies the proposed method's success. The idea of correspondence identification demonstrated in this study can also be applied to the extrinsic calibration/registration of other types of scanner and digital cameras.

Keywords

• Terrestrial laser scanner
• Digital camera
• Automatic calibration
• Point cloud
• Image correlation