Bidirectional pose estimation

This paper presents a method for relative pose estimation between two devices. Typically pose estimation is based on detecting positions of known feature points of a tracked object. I literally take a new viewpoint to this problem by considering two camera-enabled devices tracking each other. The method requires only two feature points, which can be unobtrusively integrated to small, portable devices such as mobile phones. Both devices detect each others feature points and share the positions over a wireless network connection. The combined position data can then be used to calculate the six degrees of freedom (DOF) transformation between the devices. I compared the pose estimation accuracy with a conventional method using a simulation and a real world experiment. The results show that the bidirectional method is significantly more accurate and robust to noise. It is also less affected by the target distance.