A tractable mechanism for external calibration in non-overlapping camera network

Abstract

To use a non-overlapping camera networks, the cameras' external parameters (i.e. the position and orientation of the camera) must be calibrated first. Among the popular approaches, the method proposed by Ali has been shown to work effectively with less constraints compared to alternative approaches. However, the poor robustness of the method, which may incur some practical problems, inspires us to develop a three-step mechanism to address these problems. First, the factors of the measurements impacting the error of the results are analyzed based on more than 10,000 synthetic trajectories. Second, the trajectories are classified into 4 reasonable types according to these factors. Finally, based on the statistics, this paper presents a tractable mechanism for selecting or generating the best route for the calibration target to take in order to provide improved calibration results for the cameras. The comparison between the errors associated with trajectories generated with or without the proposed mechanism are presented using a Cumulative Density Function (CDF) curve.