Fine Pose Estimation Research Articles

A hybrid tracking method for surgical augmented reality

Camera pose estimation is one of the most important, but also one of the most challenging tasks in augmented reality. Without a highly accurate estimation of the position and orientation of the digital video camera, it is impossible to render a spatially correct overlay of graphical information. This requirement is even more crucial in medical applications, where virtual objects typically have to be correctly aligned with the patient. Many experimental AR systems use specialized tracking devices, which usually are not certified for medical settings. We have developed an AR framework for surgical applications based on existing medical equipment. A surgical navigation device delivers tracking information measured by a built-in infrared camera system, which is the basis for the pose estimation of the AR video camera. However, depending on the conditions in the environment, this infrared pose data can contain discernible tracking errors. One main drawback of the medical tracking device is the fact that, while it delivers a very high positional accuracy, the reported camera orientation can contain a relatively large error. In this article, we present a hybrid tracking scheme for medical augmented reality based on a certified medical tracking system. The final pose estimation takes the initial infrared tracking data as well as salient features in the camera image into account. The vision-based component of the tracking algorithm relies on a pre-defined graphical model of the observed scene. The infrared and vision-based tracking data are tightly integrated into a unified pose estimation algorithm. This algorithm is based on an iterative numerical optimization method. We describe an implementation of the algorithm and present experimental data showing that our new method is capable of delivering a more accurate pose estimation.

Vertebral pose estimation using edge-based pattern matching and stereoradiographic 3D reconstruction of the spine

We propose a method using pattern matching techniques to significantly reduce interaction with the user when performing the evaluation of axial rotation of vertebrae. Our method works from semiautomated 3D reconstruction results obtained by using two orthogonal radiographs of the patient and a state-of-the-art stereoradiographic technique. Axial rotations of the vertebrae are first estimated from the local curvature of the spine and statistical knowledge. The final pose estimation is then obtained from a point-to-point pattern matching method using a criterion based on minimum distance and gradient phase orientation. The matching takes place between edges extracted in the images and retroprojected contours obtained from surface rendering of the spine.

Feature space trajectory methods for active computer vision

We advance new active object recognition algorithms that classify rigid objects and estimate their pose from intensity images. Our algorithms automatically detect if the class or pose of an object is ambiguous in a given image, reposition the sensor as needed, and incorporate data from multiple object views in determining the final object class and pose estimate. A probabilistic feature space trajectory (FST) in a global eigenspace is used to represent 3D distorted views of an object and to estimate the class and pose of an input object. Confidence measures for the class and pose estimates, derived using the probabilistic FST object representation, determine when additional observations are required as well as where the sensor should be positioned to provide the most useful information. We demonstrate the ability to use FSTs constructed from images rendered from computer-aided design models to recognize real objects in real images and present test results for a set of metal machined parts.