Compare computer visions algorithms for estimate 6DoF cameras pose

Abstract

Object tracking is a fundamental algorithm in robotics and variousother fields that involves estimating real world coordinates of objects from images.To achieve this, it is necessary to accurately determine the location and orientation of the capturing cameras. Typically, this process, known as “camera poseestimation,” has been manually performed and requires a meticulous calibrationprocedure. In this paper, we introduce a pipeline designed to estimate the camerapose for all available cameras based on different methods, that estimates objectcamera distance. The first method utilizes well-known tags that must be positionedin places visible from all cameras. The second approach is semi-automatic methodthat detect people moving within a scene and estimate 3D coordinates. The thirdalgorithm aims for full automation, computing depth maps derived from a singleimage. All these methods generate a cloud point in camera space used as input byan optimization algorithm that compute camera poses based on a metric that minimize a re-projection function. Finally, the entire process and some experimentalcases are presented.