Simulation Analysis of the Effect of Camera Motion on Camera Pose and Object Coordinate Calculation Accuracy

초록

Recently, image acquisition has become common in various application fields such as self-driving cars, drones, robots, and deep learning. To extract accurate visual information from a set of images, it is essential to accurately calculate the camera pose and object coordinates, and it is important to plan the camera view geometry. To achieve this, analysis of the relationship between the geometry of the camera view and the calculation accuracy of camera pose and object coordinates is required, but this has not been well studied yet. Therefore, this study presents a method to quantify the effect of different types of camera motions on the calculation accuracy of camera pose and object coordinates. The tested camera motions are sidewalk motion, forward motion, and rotation motion. For fast analysis, we modeled the relationships among image coordinates, camera pose, and object coordinates in homography-based matrices and vectors. As a result of the experiment, it was found that accuracy due to sidewalk motion and forward motion is affected by baseline length. Additionally, rotation motions ranging from –20° to 10° were found to produce similar accuracies in camera pose and object coordinate calculations.

키워드