An Online Automatic Calibration Method Based on Feature Descriptor for Non-Overlapping Multi-Camera Systems

Long Zhang,Jingdong Zhang,Wen Zhang,Chaofan Zhang,Yong Liu

DOI: https://doi.org/10.1109/robio.2018.8665100

2018-01-01

Abstract:In this paper, a novel online automatic calibration method is presented for non-overlapping multi-camera systems. An improved feature descriptor-based calibration pattern is employed for both intrinsic and extrinsic calibration, which can be readily and automatically detected. For intrinsic calibration, two improvements to the calibration method proposed by Li et al. are made to yield a better accuracy. On the one hand, a more accurate parametric polynomial projection model is introduced to replace the unified projection model. On the other hand, a new optimization method is used, which extends the calibration procedure by replacing the residual function and joints refinement of all parameters. For extrinsic calibration, the easily exploited mapping-based method is employed to calibrate the relative poses of cameras with non-overlapping fields of view. All the cameras are tracked in the same map, and each camera's pose is initialized with the same calibration pattern. To our knowledge, this paper is the first feature descriptor-based online calibration method for the non-overlapping multi-camera system. To validate the proposed method, several experiments are performed to compare our work with other methods. Furthermore, the performance of our method is evaluated in our three-camera system, and an open source can be found online https://github.com/CASHIPS-ComputerVision.